daviddoji/ISLP_labs
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

v2.2 versions of labs except Ch10

Browse Source
This commit is contained in:
Jonathan Taylor
2024-06-04 18:07:35 -07:00
parent e5bbb1a5bc
commit 29526fb7bc
25 changed files with 19373 additions and 10042 deletions
Download Patch File Download Diff File Expand all files Collapse all files

163
Ch02-statlearn-lab.Rmd
View File

@@ -1,23 +1,16 @@
---
jupyter:
jupytext:
cell_metadata_filter: -all
formats: Rmd,ipynb
text_representation:
extension: .Rmd
format_name: rmarkdown
format_version: '1.2'
jupytext_version: 1.14.7
---
# Introduction to Python
<a target="_blank" href="https://colab.research.google.com/github/intro-stat-learning/ISLP_labs/blob/v2.2/Ch02-statlearn-lab.ipynb">
<img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/>
</a>
# Chapter 2
# Lab: Introduction to Python
[![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/intro-stat-learning/ISLP_labs/v2.2?labpath=Ch02-statlearn-lab.ipynb)
## Getting Started
@@ -73,21 +66,21 @@ inputs. For example, the
print('fit a model with', 11, 'variables')
```
The following command will provide information about the `print()` function.
```{python}
# print?
print?
```
Adding two integers in `Python` is pretty intuitive.
```{python}
3 + 5
```
In `Python`, textual data is handled using
*strings*. For instance, `"hello"` and
`'hello'`
@@ -98,7 +91,7 @@ We can concatenate them using the addition `+` symbol.
"hello" + " " + "world"
```
A string is actually a type of *sequence*: this is a generic term for an ordered list.
The three most important types of sequences are lists, tuples, and strings.
We introduce lists now.
@@ -114,7 +107,7 @@ x = [3, 4, 5]
x
```
Note that we used the brackets
`[]` to construct this list.
@@ -126,14 +119,14 @@ y = [4, 9, 7]
x + y
```
The result may appear slightly counterintuitive: why did `Python` not add the entries of the lists
element-by-element?
In `Python`, lists hold *arbitrary* objects, and are added using *concatenation*.
In fact, concatenation is the behavior that we saw earlier when we entered `"hello" + " " + "world"`.
This example reflects the fact that
`Python` is a general-purpose programming language. Much of `Python`'s data-specific
functionality comes from other packages, notably `numpy`
@@ -148,8 +141,8 @@ See [docs.scipy.org/doc/numpy/user/quickstart.html](https://docs.scipy.org/doc/n
As mentioned earlier, this book makes use of functionality that is contained in the `numpy`
*library*, or *package*. A package is a collection of modules that are not necessarily included in
the base `Python` distribution. The name `numpy` is an abbreviation for *numerical Python*.
To access `numpy`, we must first `import` it.
```{python}
@@ -193,7 +186,7 @@ x
The object `x` has several
*attributes*, or associated objects. To access an attribute of `x`, we type `x.attribute`, where we replace `attribute`
@@ -203,7 +196,7 @@ For instance, we can access the `ndim` attribute of `x` as follows.
```{python}
x.ndim
```
The output indicates that `x` is a two-dimensional array.
Similarly, `x.dtype` is the *data type* attribute of the object `x`. This indicates that `x` is
comprised of 64-bit integers:
@@ -227,7 +220,7 @@ documentation associated with the function `fun`, if it exists.
We can try this for `np.array()`.
```{python}
# np.array?
np.array?
```
This documentation indicates that we could create a floating point array by passing a `dtype` argument into `np.array()`.
@@ -245,7 +238,7 @@ at its `shape` attribute.
x.shape
```
A *method* is a function that is associated with an
object.
@@ -282,10 +275,10 @@ x_reshape = x.reshape((2, 3))
print('reshaped x:\n', x_reshape)
```
The previous output reveals that `numpy` arrays are specified as a sequence
of *rows*. This is called *row-major ordering*, as opposed to *column-major ordering*.
`Python` (and hence `numpy`) uses 0-based
indexing. This means that to access the top left element of `x_reshape`,
@@ -315,13 +308,13 @@ print('x_reshape after we modify its top left element:\n', x_reshape)
print('x after we modify top left element of x_reshape:\n', x)
```
Modifying `x_reshape` also modified `x` because the two objects occupy the same space in memory.
We just saw that we can modify an element of an array. Can we also modify a tuple? It turns out that we cannot --- and trying to do so introduces
an *exception*, or error.
@@ -330,8 +323,8 @@ my_tuple = (3, 4, 5)
my_tuple[0] = 2
```
We now briefly mention some attributes of arrays that will come in handy. An array's `shape` attribute contains its dimension; this is always a tuple.
The `ndim` attribute yields the number of dimensions, and `T` provides its transpose.
@@ -339,7 +332,7 @@ The `ndim` attribute yields the number of dimensions, and `T` provides its tran
x_reshape.shape, x_reshape.ndim, x_reshape.T
```
Notice that the three individual outputs `(2,3)`, `2`, and `array([[5, 4],[2, 5], [3,6]])` are themselves output as a tuple.
We will often want to apply functions to arrays.
@@ -350,22 +343,22 @@ square root of the entries using the `np.sqrt()` function:
np.sqrt(x)
```
We can also square the elements:
```{python}
x**2
```
We can compute the square roots using the same notation, raising to the power of $1/2$ instead of 2.
```{python}
x**0.5
```
Throughout this book, we will often want to generate random data.
The `np.random.normal()` function generates a vector of random
normal variables. We can learn more about this function by looking at the help page, via a call to `np.random.normal?`.
@@ -382,7 +375,7 @@ x = np.random.normal(size=50)
x
```
We create an array `y` by adding an independent $N(50,1)$ random variable to each element of `x`.
```{python}
@@ -394,7 +387,7 @@ correlation between `x` and `y`.
```{python}
np.corrcoef(x, y)
```
If you're following along in your own `Jupyter` notebook, then you probably noticed that you got a different set of results when you ran the past few
commands. In particular,
each
@@ -407,7 +400,7 @@ print(np.random.normal(scale=5, size=2))
```
In order to ensure that our code provides exactly the same results
each time it is run, we can set a *random seed*
using the
@@ -423,7 +416,7 @@ print(rng.normal(scale=5, size=2))
rng2 = np.random.default_rng(1303)
print(rng2.normal(scale=5, size=2))
```
Throughout the labs in this book, we use `np.random.default_rng()` whenever we
perform calculations involving random quantities within `numpy`. In principle, this
should enable the reader to exactly reproduce the stated results. However, as new versions of `numpy` become available, it is possible
@@ -446,7 +439,7 @@ np.mean(y), y.mean()
```{python}
np.var(y), y.var(), np.mean((y - y.mean())**2)
```
Notice that by default `np.var()` divides by the sample size $n$ rather
than $n-1$; see the `ddof` argument in `np.var?`.
@@ -455,7 +448,7 @@ than $n-1$; see the `ddof` argument in `np.var?`.
```{python}
np.sqrt(np.var(y)), np.std(y)
```
The `np.mean()`, `np.var()`, and `np.std()` functions can also be applied to the rows and columns of a matrix.
To see this, we construct a $10 \times 3$ matrix of $N(0,1)$ random variables, and consider computing its row sums.
@@ -469,14 +462,14 @@ Since arrays are row-major ordered, the first axis, i.e. `axis=0`, refers to its
```{python}
X.mean(axis=0)
```
The following yields the same result.
```{python}
X.mean(0)
```
## Graphics
In `Python`, common practice is to use the library
@@ -542,7 +535,7 @@ As an alternative, we could use the `ax.scatter()` function to create a scatter
fig, ax = subplots(figsize=(8, 8))
ax.scatter(x, y, marker='o');
```
Notice that in the code blocks above, we have ended
the last line with a semicolon. This prevents `ax.plot(x, y)` from printing
text to the notebook. However, it does not prevent a plot from being produced.
@@ -583,7 +576,7 @@ fig.set_size_inches(12,3)
fig
```
Occasionally we will want to create several plots within a figure. This can be
achieved by passing additional arguments to `subplots()`.
@@ -612,8 +605,8 @@ Type `subplots?` to learn more about
To save the output of `fig`, we call its `savefig()`
method. The argument `dpi` is the dots per inch, used
to determine how large the figure will be in pixels.
@@ -623,7 +616,7 @@ fig.savefig("Figure.png", dpi=400)
fig.savefig("Figure.pdf", dpi=200);
```
We can continue to modify `fig` using step-by-step updates; for example, we can modify the range of the $x$-axis, re-save the figure, and even re-display it.
@@ -675,7 +668,7 @@ fig, ax = subplots(figsize=(8, 8))
ax.imshow(f);
```
## Sequences and Slice Notation
@@ -689,8 +682,8 @@ seq1 = np.linspace(0, 10, 11)
seq1
```
The function `np.arange()`
returns a sequence of numbers spaced out by `step`. If `step` is not specified, then a default value of $1$ is used. Let's create a sequence
that starts at $0$ and ends at $10$.
@@ -700,7 +693,7 @@ seq2 = np.arange(0, 10)
seq2
```
Why isn't $10$ output above? This has to do with *slice* notation in `Python`.
Slice notation
is used to index sequences such as lists, tuples and arrays.
@@ -742,7 +735,7 @@ See the documentation `slice?` for useful options in creating slices.
## Indexing Data
To begin, we create a two-dimensional `numpy` array.
@@ -752,7 +745,7 @@ A = np.array(np.arange(16)).reshape((4, 4))
A
```
Typing `A[1,2]` retrieves the element corresponding to the second row and third
column. (As usual, `Python` indexes from $0.$)
@@ -760,7 +753,7 @@ column. (As usual, `Python` indexes from $0.$)
A[1,2]
```
The first number after the open-bracket symbol `[`
refers to the row, and the second number refers to the column.
@@ -772,7 +765,7 @@ The first number after the open-bracket symbol `[`
A[[1,3]]
```
To select the first and third columns, we pass in `[0,2]` as the second argument in the square brackets.
In this case we need to supply the first argument `:`
which selects all rows.
@@ -781,7 +774,7 @@ which selects all rows.
A[:,[0,2]]
```
Now, suppose that we want to select the submatrix made up of the second and fourth
rows as well as the first and third columns. This is where
indexing gets slightly tricky. It is natural to try to use lists to retrieve the rows and columns:
@@ -790,21 +783,21 @@ indexing gets slightly tricky. It is natural to try to use lists to retrieve th
A[[1,3],[0,2]]
```
Oops --- what happened? We got a one-dimensional array of length two identical to
```{python}
np.array([A[1,0],A[3,2]])
```
Similarly, the following code fails to extract the submatrix comprised of the second and fourth rows and the first, third, and fourth columns:
```{python}
A[[1,3],[0,2,3]]
```
We can see what has gone wrong here. When supplied with two indexing lists, the `numpy` interpretation is that these provide pairs of $i,j$ indices for a series of entries. That is why the pair of lists must have the same length. However, that was not our intent, since we are looking for a submatrix.
One easy way to do this is as follows. We first create a submatrix by subsetting the rows of `A`, and then on the fly we make a further submatrix by subsetting its columns.
@@ -815,7 +808,7 @@ A[[1,3]][:,[0,2]]
```
There are more efficient ways of achieving the same result.
@@ -827,7 +820,7 @@ idx = np.ix_([1,3],[0,2,3])
A[idx]
```
Alternatively, we can subset matrices efficiently using slices.
@@ -841,7 +834,7 @@ A[1:4:2,0:3:2]
```
Why are we able to retrieve a submatrix directly using slices but not using lists?
Its because they are different `Python` types, and
are treated differently by `numpy`.
@@ -857,7 +850,7 @@ Slices can be used to extract objects from arbitrary sequences, such as strings,
### Boolean Indexing
In `numpy`, a *Boolean* is a type that equals either `True` or `False` (also represented as $1$ and $0$, respectively).
@@ -874,7 +867,7 @@ keep_rows[[1,3]] = True
keep_rows
```
Note that the elements of `keep_rows`, when viewed as integers, are the same as the
values of `np.array([0,1,0,1])`. Below, we use `==` to verify their equality. When
applied to two arrays, the `==` operation is applied elementwise.
@@ -883,7 +876,7 @@ applied to two arrays, the `==` operation is applied elementwise.
np.all(keep_rows == np.array([0,1,0,1]))
```
(Here, the function `np.all()` has checked whether
all entries of an array are `True`. A similar function, `np.any()`, can be used to check whether any entries of an array are `True`.)
@@ -895,14 +888,14 @@ The former retrieves the first, second, first, and second rows of `A`.
A[np.array([0,1,0,1])]
```
By contrast, `keep_rows` retrieves only the second and fourth rows of `A` --- i.e. the rows for which the Boolean equals `TRUE`.
```{python}
A[keep_rows]
```
This example shows that Booleans and integers are treated differently by `numpy`.
@@ -926,7 +919,7 @@ A[idx_mixed]
```
For more details on indexing in `numpy`, readers are referred
to the `numpy` tutorial mentioned earlier.
@@ -979,7 +972,7 @@ files. Before loading data into `Python`, it is a good idea to view it using
a text editor or other software, such as Microsoft Excel.
We now take a look at the column of `Auto` corresponding to the variable `horsepower`:
@@ -1000,7 +993,7 @@ We see the culprit is the value `?`, which is being used to encode missing value
To fix the problem, we must provide `pd.read_csv()` with an argument called `na_values`.
Now, each instance of `?` in the file is replaced with the
value `np.nan`, which means *not a number*:
@@ -1012,8 +1005,8 @@ Auto = pd.read_csv('Auto.data',
Auto['horsepower'].sum()
```
The `Auto.shape` attribute tells us that the data has 397
observations, or rows, and nine variables, or columns.
@@ -1021,7 +1014,7 @@ observations, or rows, and nine variables, or columns.
Auto.shape
```
There are
various ways to deal with missing data.
In this case, since only five of the rows contain missing
@@ -1032,7 +1025,7 @@ Auto_new = Auto.dropna()
Auto_new.shape
```
### Basics of Selecting Rows and Columns
@@ -1043,7 +1036,7 @@ Auto = Auto_new # overwrite the previous value
Auto.columns
```
Accessing the rows and columns of a data frame is similar, but not identical, to accessing the rows and columns of an array.
Recall that the first argument to the `[]` method
@@ -1328,8 +1321,8 @@ Auto.plot.scatter('horsepower', 'mpg', ax=axes[1]);
```
Note also that the columns of a data frame can be accessed as attributes: try typing in `Auto.horsepower`.
We now consider the `cylinders` variable. Typing in `Auto.cylinders.dtype` reveals that it is being treated as a quantitative variable.
However, since there is only a small number of possible values for this variable, we may wish to treat it as
qualitative. Below, we replace
@@ -1348,7 +1341,7 @@ fig, ax = subplots(figsize=(8, 8))
Auto.boxplot('mpg', by='cylinders', ax=ax);
```
The `hist()` method can be used to plot a *histogram*.
```{python}
@@ -1395,7 +1388,7 @@ Auto['cylinders'].describe()
Auto['mpg'].describe()
```
To exit `Jupyter`, select `File / Close and Halt`.
To exit `Jupyter`, select `File / Shut Down`.

5233
Ch02-statlearn-lab.ipynb
View File

File diff suppressed because one or more lines are too long

52
Ch03-linreg-lab.Rmd
View File

@@ -1,21 +1,13 @@
---
jupyter:
jupytext:
cell_metadata_filter: -all
formats: Rmd,ipynb
main_language: python
text_representation:
extension: .Rmd
format_name: rmarkdown
format_version: '1.2'
jupytext_version: 1.14.7
---
# Linear Regression
<a target="_blank" href="https://colab.research.google.com/github/intro-stat-learning/ISLP_labs/blob/v2.2/Ch03-linreg-lab.ipynb">
<img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/>
</a>
[![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/intro-stat-learning/ISLP_labs/v2.2?labpath=Ch03-linreg-lab.ipynb)
# Chapter 3
# Lab: Linear Regression
## Importing packages
We import our standard libraries at this top
@@ -27,7 +19,7 @@ import pandas as pd
from matplotlib.pyplot import subplots
```
### New imports
Throughout this lab we will introduce new functions and libraries. However,
@@ -103,7 +95,7 @@ A.sum()
```
## Simple Linear Regression
In this section we will construct model
@@ -125,7 +117,7 @@ Boston = load_data("Boston")
Boston.columns
```
Type `Boston?` to find out more about these data.
We start by using the `sm.OLS()` function to fit a
@@ -140,7 +132,7 @@ X = pd.DataFrame({'intercept': np.ones(Boston.shape[0]),
X[:4]
```
We extract the response, and fit the model.
```{python}
@@ -162,7 +154,7 @@ method, and returns such a summary.
summarize(results)
```
Before we describe other methods for working with fitted models, we outline a more useful and general framework for constructing a model matrix~`X`.
### Using Transformations: Fit and Transform
@@ -233,8 +225,8 @@ The fitted coefficients can also be retrieved as the
results.params
```
The `get_prediction()` method can be used to obtain predictions, and produce confidence intervals and
prediction intervals for the prediction of `medv` for given values of `lstat`.
@@ -339,7 +331,7 @@ As mentioned above, there is an existing function to add a line to a plot --- `a
Next we examine some diagnostic plots, several of which were discussed
in Section 3.3.3.
in Section~\ref{Ch3:problems.sec}.
We can find the fitted values and residuals
of the fit as attributes of the `results` object.
Various influence measures describing the regression model
@@ -404,7 +396,7 @@ terms = Boston.columns.drop('medv')
terms
```
We can now fit the model with all the variables in `terms` using
the same model matrix builder.
@@ -415,7 +407,7 @@ results = model.fit()
summarize(results)
```
What if we would like to perform a regression using all of the variables but one? For
example, in the above regression output, `age` has a high $p$-value.
So we may wish to run a regression excluding this predictor.
@@ -436,7 +428,7 @@ We can access the individual components of `results` by name
and
`np.sqrt(results.scale)` gives us the RSE.
Variance inflation factors (section 3.3.3) are sometimes useful
Variance inflation factors (section~\ref{Ch3:problems.sec}) are sometimes useful
to assess the effect of collinearity in the model matrix of a regression model.
We will compute the VIFs in our multiple regression fit, and use the opportunity to introduce the idea of *list comprehension*.
@@ -490,7 +482,7 @@ model2 = sm.OLS(y, X)
summarize(model2.fit())
```
## Non-linear Transformations of the Predictors
The model matrix builder can include terms beyond
@@ -557,7 +549,7 @@ ax = subplots(figsize=(8,8))[1]
ax.scatter(results3.fittedvalues, results3.resid)
ax.set_xlabel('Fitted value')
ax.set_ylabel('Residual')
ax.axhline(0, c='k', ls='--')
ax.axhline(0, c='k', ls='--');
```
We see that when the quadratic term is included in the model,
@@ -565,7 +557,7 @@ there is little discernible pattern in the residuals.
In order to create a cubic or higher-degree polynomial fit, we can simply change the degree argument
to `poly()`.
## Qualitative Predictors
Here we use the `Carseats` data, which is included in the

476
Ch03-linreg-lab.ipynb
View File

File diff suppressed because one or more lines are too long

48
Ch04-classification-lab.Rmd
View File

@@ -1,24 +1,16 @@
---
jupyter:
jupytext:
cell_metadata_filter: -all
formats: Rmd,ipynb
main_language: python
text_representation:
extension: .Rmd
format_name: rmarkdown
format_version: '1.2'
jupytext_version: 1.14.7
---
# Chapter 4
# Logistic Regression, LDA, QDA, and KNN
<a target="_blank" href="https://colab.research.google.com/github/intro-stat-learning/ISLP_labs/blob/v2.2/Ch04-classification-lab.ipynb">
<img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/>
</a>
[![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/intro-stat-learning/ISLP_labs/v2.2?labpath=Ch04-classification-lab.ipynb)
# Lab: Logistic Regression, LDA, QDA, and KNN
@@ -404,7 +396,7 @@ lda.fit(X_train, L_train)
```
Here we have used the list comprehensions introduced
in Section 3.6.4. Looking at our first line above, we see that the right-hand side is a list
in Section~\ref{Ch3-linreg-lab:multivariate-goodness-of-fit}. Looking at our first line above, we see that the right-hand side is a list
of length two. This is because the code `for M in [X_train, X_test]` iterates over a list
of length two. While here we loop over a list,
the list comprehension method works when looping over any iterable object.
@@ -453,7 +445,7 @@ lda.scalings_
```
These values provide the linear combination of `Lag1` and `Lag2` that are used to form the LDA decision rule. In other words, these are the multipliers of the elements of $X=x$ in (4.24).
These values provide the linear combination of `Lag1` and `Lag2` that are used to form the LDA decision rule. In other words, these are the multipliers of the elements of $X=x$ in (\ref{Ch4:bayes.multi}).
If $-0.64\times `Lag1` - 0.51 \times `Lag2` $ is large, then the LDA classifier will predict a market increase, and if it is small, then the LDA classifier will predict a market decline.
```{python}
@@ -462,7 +454,7 @@ lda_pred = lda.predict(X_test)
```
As we observed in our comparison of classification methods
(Section 4.5), the LDA and logistic
(Section~\ref{Ch4:comparison.sec}), the LDA and logistic
regression predictions are almost identical.
```{python}
@@ -521,7 +513,7 @@ The LDA classifier above is the first classifier from the
`sklearn` library. We will use several other objects
from this library. The objects
follow a common structure that simplifies tasks such as cross-validation,
which we will see in Chapter 5. Specifically,
which we will see in Chapter~\ref{Ch5:resample}. Specifically,
the methods first create a generic classifier without
referring to any data. This classifier is then fit
to data with the `fit()` method and predictions are
@@ -807,7 +799,7 @@ feature_std.std()
```
Notice that the standard deviations are not quite $1$ here; this is again due to some procedures using the $1/n$ convention for variances (in this case `scaler()`), while others use $1/(n-1)$ (the `std()` method). See the footnote on page 200.
Notice that the standard deviations are not quite $1$ here; this is again due to some procedures using the $1/n$ convention for variances (in this case `scaler()`), while others use $1/(n-1)$ (the `std()` method). See the footnote on page~\pageref{Ch4-varformula}.
In this case it does not matter, as long as the variables are all on the same scale.
Using the function `train_test_split()` we now split the observations into a test set,
@@ -874,7 +866,7 @@ This is double the rate that one would obtain from random guessing.
The number of neighbors in KNN is referred to as a *tuning parameter*, also referred to as a *hyperparameter*.
We do not know *a priori* what value to use. It is therefore of interest
to see how the classifier performs on test data as we vary these
parameters. This can be achieved with a `for` loop, described in Section 2.3.8.
parameters. This can be achieved with a `for` loop, described in Section~\ref{Ch2-statlearn-lab:for-loops}.
Here we use a for loop to look at the accuracy of our classifier in the group predicted to purchase
insurance as we vary the number of neighbors from 1 to 5:
@@ -901,7 +893,7 @@ As a comparison, we can also fit a logistic regression model to the
data. This can also be done
with `sklearn`, though by default it fits
something like the *ridge regression* version
of logistic regression, which we introduce in Chapter 6. This can
of logistic regression, which we introduce in Chapter~\ref{Ch6:varselect}. This can
be modified by appropriately setting the argument `C` below. Its default
value is 1 but by setting it to a very large number, the algorithm converges to the same solution as the usual (unregularized)
logistic regression estimator discussed above.
@@ -945,7 +937,7 @@ confusion_table(logit_labels, y_test)
```
## Linear and Poisson Regression on the Bikeshare Data
Here we fit linear and Poisson regression models to the `Bikeshare` data, as described in Section 4.6.
Here we fit linear and Poisson regression models to the `Bikeshare` data, as described in Section~\ref{Ch4:sec:pois}.
The response `bikers` measures the number of bike rentals per hour
in Washington, DC in the period 2010--2012.
@@ -986,7 +978,7 @@ variables constant, there are on average about 7 more riders in
February than in January. Similarly there are about 16.5 more riders
in March than in January.
The results seen in Section 4.6.1
The results seen in Section~\ref{sec:bikeshare.linear}
used a slightly different coding of the variables `hr` and `mnth`, as follows:
```{python}
@@ -1040,7 +1032,7 @@ np.allclose(M_lm.fittedvalues, M2_lm.fittedvalues)
```
To reproduce the left-hand side of Figure 4.13
To reproduce the left-hand side of Figure~\ref{Ch4:bikeshare}
we must first obtain the coefficient estimates associated with
`mnth`. The coefficients for January through November can be obtained
directly from the `M2_lm` object. The coefficient for December
@@ -1080,7 +1072,7 @@ ax_month.set_ylabel('Coefficient', fontsize=20);
```
Reproducing the right-hand plot in Figure 4.13 follows a similar process.
Reproducing the right-hand plot in Figure~\ref{Ch4:bikeshare} follows a similar process.
```{python}
coef_hr = S2[S2.index.str.contains('hr')]['coef']
@@ -1115,7 +1107,7 @@ M_pois = sm.GLM(Y, X2, family=sm.families.Poisson()).fit()
```
We can plot the coefficients associated with `mnth` and `hr`, in order to reproduce Figure 4.15. We first complete these coefficients as before.
We can plot the coefficients associated with `mnth` and `hr`, in order to reproduce Figure~\ref{Ch4:bikeshare.pois}. We first complete these coefficients as before.
```{python}
S_pois = summarize(M_pois)

2169
Ch04-classification-lab.ipynb
View File

File diff suppressed because one or more lines are too long

50
Ch05-resample-lab.Rmd
View File

@@ -1,23 +1,15 @@
---
jupyter:
jupytext:
cell_metadata_filter: -all
formats: Rmd,ipynb
main_language: python
text_representation:
extension: .Rmd
format_name: rmarkdown
format_version: '1.2'
jupytext_version: 1.14.7
---
# Chapter 5
# Cross-Validation and the Bootstrap
<a target="_blank" href="https://colab.research.google.com/github/intro-stat-learning/ISLP_labs/blob/v2.2/Ch05-resample-lab.ipynb">
<img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/>
</a>
[![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/intro-stat-learning/ISLP_labs/v2.2?labpath=Ch05-resample-lab.ipynb)
# Lab: Cross-Validation and the Bootstrap
In this lab, we explore the resampling techniques covered in this
chapter. Some of the commands in this lab may take a while to run on
your computer.
@@ -235,7 +227,7 @@ for i, d in enumerate(range(1,6)):
cv_error
```
As in Figure 5.4, we see a sharp drop in the estimated test MSE between the linear and
As in Figure~\ref{Ch5:cvplot}, we see a sharp drop in the estimated test MSE between the linear and
quadratic fits, but then no clear improvement from using higher-degree polynomials.
Above we introduced the `outer()` method of the `np.power()`
@@ -276,7 +268,7 @@ cv_error
Notice that the computation time is much shorter than that of LOOCV.
(In principle, the computation time for LOOCV for a least squares
linear model should be faster than for $K$-fold CV, due to the
availability of the formula (5.2) for LOOCV;
availability of the formula~(\ref{Ch5:eq:LOOCVform}) for LOOCV;
however, the generic `cross_validate()` function does not make
use of this formula.) We still see little evidence that using cubic
or higher-degree polynomial terms leads to a lower test error than simply
@@ -322,7 +314,7 @@ incurred by picking different random folds.
## The Bootstrap
We illustrate the use of the bootstrap in the simple example
{of Section 5.2,} as well as on an example involving
{of Section~\ref{Ch5:sec:bootstrap},} as well as on an example involving
estimating the accuracy of the linear regression model on the `Auto`
data set.
### Estimating the Accuracy of a Statistic of Interest
@@ -337,8 +329,8 @@ in a dataframe.
To illustrate the bootstrap, we
start with a simple example.
The `Portfolio` data set in the `ISLP` package is described
in Section 5.2. The goal is to estimate the
sampling variance of the parameter $\alpha$ given in formula (5.7). We will
in Section~\ref{Ch5:sec:bootstrap}. The goal is to estimate the
sampling variance of the parameter $\alpha$ given in formula~(\ref{Ch5:min.var}). We will
create a function
`alpha_func()`, which takes as input a dataframe `D` assumed
to have columns `X` and `Y`, as well as a
@@ -357,7 +349,7 @@ def alpha_func(D, idx):
```
This function returns an estimate for $\alpha$
based on applying the minimum
variance formula (5.7) to the observations indexed by
variance formula (\ref{Ch5:min.var}) to the observations indexed by
the argument `idx`. For instance, the following command
estimates $\alpha$ using all 100 observations.
@@ -427,7 +419,7 @@ intercept and slope terms for the linear regression model that uses
`horsepower` to predict `mpg` in the `Auto` data set. We
will compare the estimates obtained using the bootstrap to those
obtained using the formulas for ${\rm SE}(\hat{\beta}_0)$ and
${\rm SE}(\hat{\beta}_1)$ described in Section 3.1.2.
${\rm SE}(\hat{\beta}_1)$ described in Section~\ref{Ch3:secoefsec}.
To use our `boot_SE()` function, we must write a function (its
first argument)
@@ -474,7 +466,7 @@ demonstrate its utility on 10 bootstrap samples.
```{python}
rng = np.random.default_rng(0)
np.array([hp_func(Auto,
rng.choice(392,
rng.choice(Auto.index,
392,
replace=True)) for _ in range(10)])
@@ -496,7 +488,7 @@ This indicates that the bootstrap estimate for ${\rm SE}(\hat{\beta}_0)$ is
0.85, and that the bootstrap
estimate for ${\rm SE}(\hat{\beta}_1)$ is
0.0074. As discussed in
Section 3.1.2, standard formulas can be used to compute
Section~\ref{Ch3:secoefsec}, standard formulas can be used to compute
the standard errors for the regression coefficients in a linear
model. These can be obtained using the `summarize()` function
from `ISLP.sm`.
@@ -510,7 +502,7 @@ model_se
The standard error estimates for $\hat{\beta}_0$ and $\hat{\beta}_1$
obtained using the formulas from Section 3.1.2 are
obtained using the formulas from Section~\ref{Ch3:secoefsec} are
0.717 for the
intercept and
0.006 for the
@@ -518,13 +510,13 @@ slope. Interestingly, these are somewhat different from the estimates
obtained using the bootstrap. Does this indicate a problem with the
bootstrap? In fact, it suggests the opposite. Recall that the
standard formulas given in
{Equation 3.8 on page 82}
{Equation~\ref{Ch3:se.eqn} on page~\pageref{Ch3:se.eqn}}
rely on certain assumptions. For example,
they depend on the unknown parameter $\sigma^2$, the noise
variance. We then estimate $\sigma^2$ using the RSS. Now although the
formula for the standard errors do not rely on the linear model being
correct, the estimate for $\sigma^2$ does. We see
{in Figure 3.8 on page 108} that there is
{in Figure~\ref{Ch3:polyplot} on page~\pageref{Ch3:polyplot}} that there is
a non-linear relationship in the data, and so the residuals from a
linear fit will be inflated, and so will $\hat{\sigma}^2$. Secondly,
the standard formulas assume (somewhat unrealistically) that the $x_i$
@@ -537,7 +529,7 @@ the results from `sm.OLS`.
Below we compute the bootstrap standard error estimates and the
standard linear regression estimates that result from fitting the
quadratic model to the data. Since this model provides a good fit to
the data (Figure 3.8), there is now a better
the data (Figure~\ref{Ch3:polyplot}), there is now a better
correspondence between the bootstrap estimates and the standard
estimates of ${\rm SE}(\hat{\beta}_0)$, ${\rm SE}(\hat{\beta}_1)$ and
${\rm SE}(\hat{\beta}_2)$.

378
Ch05-resample-lab.ipynb
View File

@@ -2,20 +2,29 @@
"cells": [
{
"cell_type": "markdown",
"id": "85ad9863",
"id": "dc2d635a",
"metadata": {},
"source": []
},
{
"cell_type": "markdown",
"id": "6dde3cef",
"metadata": {},
"source": [
"# Cross-Validation and the Bootstrap\n",
"\n",
"# Chapter 5\n",
"\n"
"<a target=\"_blank\" href=\"https://colab.research.google.com/github/intro-stat-learning/ISLP_labs/blob/v2.2/Ch05-resample-lab.ipynb\">\n",
"<img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/>\n",
"</a>\n",
"\n",
"[![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/intro-stat-learning/ISLP_labs/v2.2?labpath=Ch05-resample-lab.ipynb)"
]
},
{
"cell_type": "markdown",
"id": "ac8b08af",
"id": "a9fd4324",
"metadata": {},
"source": [
"# Lab: Cross-Validation and the Bootstrap\n",
"In this lab, we explore the resampling techniques covered in this\n",
"chapter. Some of the commands in this lab may take a while to run on\n",
"your computer.\n",
@@ -26,9 +35,14 @@
{
"cell_type": "code",
"execution_count": 1,
"id": "e7712cfe",
"id": "f1deb5cc",
"metadata": {
"execution": {},
"execution": {
"iopub.execute_input": "2024-06-04T23:19:13.493284Z",
"iopub.status.busy": "2024-06-04T23:19:13.492950Z",
"iopub.status.idle": "2024-06-04T23:19:14.143174Z",
"shell.execute_reply": "2024-06-04T23:19:14.142882Z"
},
"lines_to_next_cell": 2
},
"outputs": [],
@@ -44,7 +58,7 @@
},
{
"cell_type": "markdown",
"id": "784a2ba3",
"id": "afa08b62",
"metadata": {},
"source": [
"There are several new imports needed for this lab."
@@ -53,9 +67,14 @@
{
"cell_type": "code",
"execution_count": 2,
"id": "21c2ed4f",
"id": "268c41b3",
"metadata": {
"execution": {},
"execution": {
"iopub.execute_input": "2024-06-04T23:19:14.144884Z",
"iopub.status.busy": "2024-06-04T23:19:14.144773Z",
"iopub.status.idle": "2024-06-04T23:19:14.146541Z",
"shell.execute_reply": "2024-06-04T23:19:14.146330Z"
},
"lines_to_next_cell": 2
},
"outputs": [],
@@ -71,7 +90,7 @@
},
{
"cell_type": "markdown",
"id": "9ac3acd5",
"id": "1c04f8e4",
"metadata": {},
"source": [
"## The Validation Set Approach\n",
@@ -92,9 +111,14 @@
{
"cell_type": "code",
"execution_count": 3,
"id": "8af59641",
"id": "22f44ae0",
"metadata": {
"execution": {}
"execution": {
"iopub.execute_input": "2024-06-04T23:19:14.147809Z",
"iopub.status.busy": "2024-06-04T23:19:14.147730Z",
"iopub.status.idle": "2024-06-04T23:19:14.152606Z",
"shell.execute_reply": "2024-06-04T23:19:14.152414Z"
}
},
"outputs": [],
"source": [
@@ -106,7 +130,7 @@
},
{
"cell_type": "markdown",
"id": "e76383f0",
"id": "318fe69f",
"metadata": {},
"source": [
"Now we can fit a linear regression using only the observations corresponding to the training set `Auto_train`."
@@ -115,9 +139,14 @@
{
"cell_type": "code",
"execution_count": 4,
"id": "d9b0b7c8",
"id": "0c32e917",
"metadata": {
"execution": {}
"execution": {
"iopub.execute_input": "2024-06-04T23:19:14.153847Z",
"iopub.status.busy": "2024-06-04T23:19:14.153757Z",
"iopub.status.idle": "2024-06-04T23:19:14.157537Z",
"shell.execute_reply": "2024-06-04T23:19:14.157339Z"
}
},
"outputs": [],
"source": [
@@ -130,7 +159,7 @@
},
{
"cell_type": "markdown",
"id": "d196dd08",
"id": "7e883b8f",
"metadata": {},
"source": [
"We now use the `predict()` method of `results` evaluated on the model matrix for this model\n",
@@ -140,9 +169,14 @@
{
"cell_type": "code",
"execution_count": 5,
"id": "3e77d831",
"id": "86ce4f85",
"metadata": {
"execution": {}
"execution": {
"iopub.execute_input": "2024-06-04T23:19:14.158717Z",
"iopub.status.busy": "2024-06-04T23:19:14.158637Z",
"iopub.status.idle": "2024-06-04T23:19:14.162177Z",
"shell.execute_reply": "2024-06-04T23:19:14.161910Z"
}
},
"outputs": [
{
@@ -165,7 +199,7 @@
},
{
"cell_type": "markdown",
"id": "f4369ee6",
"id": "f2ecdee6",
"metadata": {},
"source": [
"Hence our estimate for the validation MSE of the linear regression\n",
@@ -179,9 +213,14 @@
{
"cell_type": "code",
"execution_count": 6,
"id": "0aa4bfcc",
"id": "50a66a97",
"metadata": {
"execution": {}
"execution": {
"iopub.execute_input": "2024-06-04T23:19:14.163466Z",
"iopub.status.busy": "2024-06-04T23:19:14.163397Z",
"iopub.status.idle": "2024-06-04T23:19:14.165323Z",
"shell.execute_reply": "2024-06-04T23:19:14.165076Z"
}
},
"outputs": [],
"source": [
@@ -205,7 +244,7 @@
},
{
"cell_type": "markdown",
"id": "0271dc50",
"id": "a255779c",
"metadata": {},
"source": [
"Let’s use this function to estimate the validation MSE\n",
@@ -217,9 +256,14 @@
{
"cell_type": "code",
"execution_count": 7,
"id": "a0dbd55f",
"id": "d49b6999",
"metadata": {
"execution": {}
"execution": {
"iopub.execute_input": "2024-06-04T23:19:14.166563Z",
"iopub.status.busy": "2024-06-04T23:19:14.166497Z",
"iopub.status.idle": "2024-06-04T23:19:14.177198Z",
"shell.execute_reply": "2024-06-04T23:19:14.176975Z"
}
},
"outputs": [
{
@@ -245,7 +289,7 @@
},
{
"cell_type": "markdown",
"id": "a7401536",
"id": "9d7b8fc1",
"metadata": {},
"source": [
"These error rates are $23.62, 18.76$, and $18.80$, respectively. If we\n",
@@ -256,9 +300,14 @@
{
"cell_type": "code",
"execution_count": 8,
"id": "885136a4",
"id": "dac8bd54",
"metadata": {
"execution": {}
"execution": {
"iopub.execute_input": "2024-06-04T23:19:14.178405Z",
"iopub.status.busy": "2024-06-04T23:19:14.178321Z",
"iopub.status.idle": "2024-06-04T23:19:14.188650Z",
"shell.execute_reply": "2024-06-04T23:19:14.188432Z"
}
},
"outputs": [
{
@@ -287,7 +336,7 @@
},
{
"cell_type": "markdown",
"id": "00785402",
"id": "61f2c12d",
"metadata": {},
"source": [
"Using this split of the observations into a training set and a validation set,\n",
@@ -301,7 +350,7 @@
},
{
"cell_type": "markdown",
"id": "21c071b8",
"id": "f22daa51",
"metadata": {},
"source": [
"## Cross-Validation\n",
@@ -334,9 +383,14 @@
{
"cell_type": "code",
"execution_count": 9,
"id": "6d957d8c",
"id": "601ae443",
"metadata": {
"execution": {},
"execution": {
"iopub.execute_input": "2024-06-04T23:19:14.189993Z",
"iopub.status.busy": "2024-06-04T23:19:14.189906Z",
"iopub.status.idle": "2024-06-04T23:19:14.876368Z",
"shell.execute_reply": "2024-06-04T23:19:14.876129Z"
},
"lines_to_next_cell": 0
},
"outputs": [
@@ -365,7 +419,7 @@
},
{
"cell_type": "markdown",
"id": "c17e2bc8",
"id": "ebadc35f",
"metadata": {},
"source": [
"The arguments to `cross_validate()` are as follows: an\n",
@@ -381,7 +435,7 @@
},
{
"cell_type": "markdown",
"id": "5c7901f2",
"id": "25f47b99",
"metadata": {},
"source": [
"We can repeat this procedure for increasingly complex polynomial fits.\n",
@@ -397,9 +451,14 @@
{
"cell_type": "code",
"execution_count": 10,
"id": "e2b5ce95",
"id": "11226c85",
"metadata": {
"execution": {},
"execution": {
"iopub.execute_input": "2024-06-04T23:19:14.877800Z",
"iopub.status.busy": "2024-06-04T23:19:14.877726Z",
"iopub.status.idle": "2024-06-04T23:19:15.384419Z",
"shell.execute_reply": "2024-06-04T23:19:15.384193Z"
},
"lines_to_next_cell": 0
},
"outputs": [
@@ -430,10 +489,10 @@
},
{
"cell_type": "markdown",
"id": "03706248",
"id": "a3a920ae",
"metadata": {},
"source": [
"As in Figure 5.4, we see a sharp drop in the estimated test MSE between the linear and\n",
"As in Figure~\\ref{Ch5:cvplot}, we see a sharp drop in the estimated test MSE between the linear and\n",
"quadratic fits, but then no clear improvement from using higher-degree polynomials.\n",
"\n",
"Above we introduced the `outer()` method of the `np.power()`\n",
@@ -449,9 +508,14 @@
{
"cell_type": "code",
"execution_count": 11,
"id": "1dda1bd7",
"id": "64b64d97",
"metadata": {
"execution": {}
"execution": {
"iopub.execute_input": "2024-06-04T23:19:15.385768Z",
"iopub.status.busy": "2024-06-04T23:19:15.385690Z",
"iopub.status.idle": "2024-06-04T23:19:15.387686Z",
"shell.execute_reply": "2024-06-04T23:19:15.387484Z"
}
},
"outputs": [
{
@@ -475,7 +539,7 @@
},
{
"cell_type": "markdown",
"id": "f5092f1b",
"id": "71385c1b",
"metadata": {},
"source": [
"In the CV example above, we used $K=n$, but of course we can also use $K<n$. The code is very similar\n",
@@ -486,9 +550,14 @@
{
"cell_type": "code",
"execution_count": 12,
"id": "fb25fa70",
"id": "ca0f972f",
"metadata": {
"execution": {},
"execution": {
"iopub.execute_input": "2024-06-04T23:19:15.389014Z",
"iopub.status.busy": "2024-06-04T23:19:15.388934Z",
"iopub.status.idle": "2024-06-04T23:19:15.407438Z",
"shell.execute_reply": "2024-06-04T23:19:15.407194Z"
},
"lines_to_next_cell": 0
},
"outputs": [
@@ -520,13 +589,13 @@
},
{
"cell_type": "markdown",
"id": "c4ec6afb",
"id": "8b234093",
"metadata": {},
"source": [
"Notice that the computation time is much shorter than that of LOOCV.\n",
"(In principle, the computation time for LOOCV for a least squares\n",
"linear model should be faster than for $K$-fold CV, due to the\n",
"availability of the formula (5.2) for LOOCV;\n",
"availability of the formula~(\\ref{Ch5:eq:LOOCVform}) for LOOCV;\n",
"however, the generic `cross_validate()` function does not make\n",
"use of this formula.) We still see little evidence that using cubic\n",
"or higher-degree polynomial terms leads to a lower test error than simply\n",
@@ -535,7 +604,7 @@
},
{
"cell_type": "markdown",
"id": "5edf407f",
"id": "fb4487a4",
"metadata": {},
"source": [
"The `cross_validate()` function is flexible and can take\n",
@@ -546,9 +615,14 @@
{
"cell_type": "code",
"execution_count": 13,
"id": "d78795cd",
"id": "080cdb29",
"metadata": {
"execution": {},
"execution": {
"iopub.execute_input": "2024-06-04T23:19:15.408750Z",
"iopub.status.busy": "2024-06-04T23:19:15.408677Z",
"iopub.status.idle": "2024-06-04T23:19:15.413979Z",
"shell.execute_reply": "2024-06-04T23:19:15.413762Z"
},
"lines_to_next_cell": 2
},
"outputs": [
@@ -576,7 +650,7 @@
},
{
"cell_type": "markdown",
"id": "a081be63",
"id": "b2f4b4cf",
"metadata": {},
"source": [
"One can estimate the variability in the test error by running the following:"
@@ -585,9 +659,14 @@
{
"cell_type": "code",
"execution_count": 14,
"id": "0407ad56",
"id": "7c46de2b",
"metadata": {
"execution": {}
"execution": {
"iopub.execute_input": "2024-06-04T23:19:15.415225Z",
"iopub.status.busy": "2024-06-04T23:19:15.415158Z",
"iopub.status.idle": "2024-06-04T23:19:15.437526Z",
"shell.execute_reply": "2024-06-04T23:19:15.437302Z"
}
},
"outputs": [
{
@@ -614,7 +693,7 @@
},
{
"cell_type": "markdown",
"id": "b66db3cb",
"id": "07165f0e",
"metadata": {},
"source": [
"Note that this standard deviation is not a valid estimate of the\n",
@@ -625,7 +704,7 @@
"\n",
"## The Bootstrap\n",
"We illustrate the use of the bootstrap in the simple example\n",
" {of Section 5.2,} as well as on an example involving\n",
" {of Section~\\ref{Ch5:sec:bootstrap},} as well as on an example involving\n",
"estimating the accuracy of the linear regression model on the `Auto`\n",
"data set.\n",
"### Estimating the Accuracy of a Statistic of Interest\n",
@@ -640,8 +719,8 @@
"To illustrate the bootstrap, we\n",
"start with a simple example.\n",
"The `Portfolio` data set in the `ISLP` package is described\n",
"in Section 5.2. The goal is to estimate the\n",
"sampling variance of the parameter $\\alpha$ given in formula (5.7). We will\n",
"in Section~\\ref{Ch5:sec:bootstrap}. The goal is to estimate the\n",
"sampling variance of the parameter $\\alpha$ given in formula~(\\ref{Ch5:min.var}). We will\n",
"create a function\n",
"`alpha_func()`, which takes as input a dataframe `D` assumed\n",
"to have columns `X` and `Y`, as well as a\n",
@@ -654,9 +733,14 @@
{
"cell_type": "code",
"execution_count": 15,
"id": "f04f15bd",
"id": "a4b6d9b3",
"metadata": {
"execution": {},
"execution": {
"iopub.execute_input": "2024-06-04T23:19:15.438786Z",
"iopub.status.busy": "2024-06-04T23:19:15.438714Z",
"iopub.status.idle": "2024-06-04T23:19:15.441484Z",
"shell.execute_reply": "2024-06-04T23:19:15.441268Z"
},
"lines_to_next_cell": 0
},
"outputs": [],
@@ -670,12 +754,12 @@
},
{
"cell_type": "markdown",
"id": "c88bd6a4",
"id": "9d50058e",
"metadata": {},
"source": [
"This function returns an estimate for $\\alpha$\n",
"based on applying the minimum\n",
" variance formula (5.7) to the observations indexed by\n",
" variance formula (\\ref{Ch5:min.var}) to the observations indexed by\n",
"the argument `idx`. For instance, the following command\n",
"estimates $\\alpha$ using all 100 observations."
]
@@ -683,9 +767,14 @@
{
"cell_type": "code",
"execution_count": 16,
"id": "f98c0323",
"id": "81498a11",
"metadata": {
"execution": {}
"execution": {
"iopub.execute_input": "2024-06-04T23:19:15.442843Z",
"iopub.status.busy": "2024-06-04T23:19:15.442765Z",
"iopub.status.idle": "2024-06-04T23:19:15.445171Z",
"shell.execute_reply": "2024-06-04T23:19:15.444944Z"
}
},
"outputs": [
{
@@ -705,7 +794,7 @@
},
{
"cell_type": "markdown",
"id": "58a78f00",
"id": "4f5d0aab",
"metadata": {},
"source": [
"Next we randomly select\n",
@@ -717,9 +806,14 @@
{
"cell_type": "code",
"execution_count": 17,
"id": "bcd40175",
"id": "64fe1cb6",
"metadata": {
"execution": {},
"execution": {
"iopub.execute_input": "2024-06-04T23:19:15.446422Z",
"iopub.status.busy": "2024-06-04T23:19:15.446354Z",
"iopub.status.idle": "2024-06-04T23:19:15.448793Z",
"shell.execute_reply": "2024-06-04T23:19:15.448579Z"
},
"lines_to_next_cell": 2
},
"outputs": [
@@ -744,7 +838,7 @@
},
{
"cell_type": "markdown",
"id": "e6058be4",
"id": "91a635fe",
"metadata": {},
"source": [
"This process can be generalized to create a simple function `boot_SE()` for\n",
@@ -755,9 +849,14 @@
{
"cell_type": "code",
"execution_count": 18,
"id": "ab6602cd",
"id": "dd16bbae",
"metadata": {
"execution": {},
"execution": {
"iopub.execute_input": "2024-06-04T23:19:15.450062Z",
"iopub.status.busy": "2024-06-04T23:19:15.449992Z",
"iopub.status.idle": "2024-06-04T23:19:15.451958Z",
"shell.execute_reply": "2024-06-04T23:19:15.451742Z"
},
"lines_to_next_cell": 0
},
"outputs": [],
@@ -782,7 +881,7 @@
},
{
"cell_type": "markdown",
"id": "d94d383e",
"id": "ac4e17ed",
"metadata": {},
"source": [
"Notice the use of `_` as a loop variable in `for _ in range(B)`. This is often used if the value of the counter is\n",
@@ -795,9 +894,14 @@
{
"cell_type": "code",
"execution_count": 19,
"id": "4a323513",
"id": "b42b4585",
"metadata": {
"execution": {}
"execution": {
"iopub.execute_input": "2024-06-04T23:19:15.453190Z",
"iopub.status.busy": "2024-06-04T23:19:15.453118Z",
"iopub.status.idle": "2024-06-04T23:19:15.631597Z",
"shell.execute_reply": "2024-06-04T23:19:15.631370Z"
}
},
"outputs": [
{
@@ -821,7 +925,7 @@
},
{
"cell_type": "markdown",
"id": "22343f53",
"id": "6c5464d7",
"metadata": {},
"source": [
"The final output shows that the bootstrap estimate for ${\\rm SE}(\\hat{\\alpha})$ is $0.0912$.\n",
@@ -835,7 +939,7 @@
"`horsepower` to predict `mpg` in the `Auto` data set. We\n",
"will compare the estimates obtained using the bootstrap to those\n",
"obtained using the formulas for ${\\rm SE}(\\hat{\\beta}_0)$ and\n",
"${\\rm SE}(\\hat{\\beta}_1)$ described in Section 3.1.2.\n",
"${\\rm SE}(\\hat{\\beta}_1)$ described in Section~\\ref{Ch3:secoefsec}.\n",
"\n",
"To use our `boot_SE()` function, we must write a function (its\n",
"first argument)\n",
@@ -856,9 +960,14 @@
{
"cell_type": "code",
"execution_count": 20,
"id": "0220f3af",
"id": "6bc11784",
"metadata": {
"execution": {},
"execution": {
"iopub.execute_input": "2024-06-04T23:19:15.632802Z",
"iopub.status.busy": "2024-06-04T23:19:15.632725Z",
"iopub.status.idle": "2024-06-04T23:19:15.634450Z",
"shell.execute_reply": "2024-06-04T23:19:15.634222Z"
},
"lines_to_next_cell": 0
},
"outputs": [],
@@ -872,7 +981,7 @@
},
{
"cell_type": "markdown",
"id": "df0c7f05",
"id": "2a6ea3ce",
"metadata": {},
"source": [
"This is not quite what is needed as the first argument to\n",
@@ -886,9 +995,14 @@
{
"cell_type": "code",
"execution_count": 21,
"id": "62037dcb",
"id": "740cd50c",
"metadata": {
"execution": {},
"execution": {
"iopub.execute_input": "2024-06-04T23:19:15.635644Z",
"iopub.status.busy": "2024-06-04T23:19:15.635575Z",
"iopub.status.idle": "2024-06-04T23:19:15.637097Z",
"shell.execute_reply": "2024-06-04T23:19:15.636867Z"
},
"lines_to_next_cell": 0
},
"outputs": [],
@@ -898,7 +1012,7 @@
},
{
"cell_type": "markdown",
"id": "61fbe248",
"id": "ed6d19e2",
"metadata": {},
"source": [
"Typing `hp_func?` will show that it has two arguments `D`\n",
@@ -914,25 +1028,30 @@
{
"cell_type": "code",
"execution_count": 22,
"id": "b8bdb7a4",
"id": "ffb3ec50",
"metadata": {
"execution": {},
"execution": {
"iopub.execute_input": "2024-06-04T23:19:15.638287Z",
"iopub.status.busy": "2024-06-04T23:19:15.638220Z",
"iopub.status.idle": "2024-06-04T23:19:15.656475Z",
"shell.execute_reply": "2024-06-04T23:19:15.656261Z"
},
"lines_to_next_cell": 0
},
"outputs": [
{
"data": {
"text/plain": [
"array([[39.88064456, -0.1567849 ],\n",
" [38.73298691, -0.14699495],\n",
" [38.31734657, -0.14442683],\n",
" [39.91446826, -0.15782234],\n",
" [39.43349349, -0.15072702],\n",
" [40.36629857, -0.15912217],\n",
" [39.62334517, -0.15449117],\n",
" [39.0580588 , -0.14952908],\n",
" [38.66688437, -0.14521037],\n",
" [39.64280792, -0.15555698]])"
"array([[39.12226577, -0.1555926 ],\n",
" [37.18648613, -0.13915813],\n",
" [37.46989244, -0.14112749],\n",
" [38.56723252, -0.14830116],\n",
" [38.95495707, -0.15315141],\n",
" [39.12563927, -0.15261044],\n",
" [38.45763251, -0.14767251],\n",
" [38.43372587, -0.15019447],\n",
" [37.87581142, -0.1409544 ],\n",
" [37.95949036, -0.1451333 ]])"
]
},
"execution_count": 22,
@@ -943,14 +1062,14 @@
"source": [
"rng = np.random.default_rng(0)\n",
"np.array([hp_func(Auto,\n",
" rng.choice(392,\n",
" rng.choice(Auto.index,\n",
" 392,\n",
" replace=True)) for _ in range(10)])\n"
]
},
{
"cell_type": "markdown",
"id": "2a831036",
"id": "c6d09d96",
"metadata": {},
"source": [
"Next, we use the `boot_SE()` {} function to compute the standard\n",
@@ -960,17 +1079,22 @@
{
"cell_type": "code",
"execution_count": 23,
"id": "36808258",
"id": "7d561f70",
"metadata": {
"execution": {},
"execution": {
"iopub.execute_input": "2024-06-04T23:19:15.657733Z",
"iopub.status.busy": "2024-06-04T23:19:15.657659Z",
"iopub.status.idle": "2024-06-04T23:19:17.204871Z",
"shell.execute_reply": "2024-06-04T23:19:17.204614Z"
},
"lines_to_next_cell": 2
},
"outputs": [
{
"data": {
"text/plain": [
"intercept 0.848807\n",
"horsepower 0.007352\n",
"intercept 0.731176\n",
"horsepower 0.006092\n",
"dtype: float64"
]
},
@@ -989,14 +1113,14 @@
},
{
"cell_type": "markdown",
"id": "38c65fbf",
"id": "a834f240",
"metadata": {},
"source": [
"This indicates that the bootstrap estimate for ${\\rm SE}(\\hat{\\beta}_0)$ is\n",
"0.85, and that the bootstrap\n",
"estimate for ${\\rm SE}(\\hat{\\beta}_1)$ is\n",
"0.0074. As discussed in\n",
"Section 3.1.2, standard formulas can be used to compute\n",
"Section~\\ref{Ch3:secoefsec}, standard formulas can be used to compute\n",
"the standard errors for the regression coefficients in a linear\n",
"model. These can be obtained using the `summarize()` function\n",
"from `ISLP.sm`."
@@ -1005,9 +1129,14 @@
{
"cell_type": "code",
"execution_count": 24,
"id": "c9aea297",
"id": "3888aa0a",
"metadata": {
"execution": {},
"execution": {
"iopub.execute_input": "2024-06-04T23:19:17.206302Z",
"iopub.status.busy": "2024-06-04T23:19:17.206223Z",
"iopub.status.idle": "2024-06-04T23:19:17.221631Z",
"shell.execute_reply": "2024-06-04T23:19:17.221444Z"
},
"lines_to_next_cell": 2
},
"outputs": [
@@ -1032,11 +1161,11 @@
},
{
"cell_type": "markdown",
"id": "d870ad6b",
"id": "aefc0575",
"metadata": {},
"source": [
"The standard error estimates for $\\hat{\\beta}_0$ and $\\hat{\\beta}_1$\n",
"obtained using the formulas from Section 3.1.2 are\n",
"obtained using the formulas from Section~\\ref{Ch3:secoefsec} are\n",
"0.717 for the\n",
"intercept and\n",
"0.006 for the\n",
@@ -1044,13 +1173,13 @@
"obtained using the bootstrap. Does this indicate a problem with the\n",
"bootstrap? In fact, it suggests the opposite. Recall that the\n",
"standard formulas given in\n",
" {Equation 3.8 on page 82}\n",
" {Equation~\\ref{Ch3:se.eqn} on page~\\pageref{Ch3:se.eqn}}\n",
"rely on certain assumptions. For example,\n",
"they depend on the unknown parameter $\\sigma^2$, the noise\n",
"variance. We then estimate $\\sigma^2$ using the RSS. Now although the\n",
"formula for the standard errors do not rely on the linear model being\n",
"correct, the estimate for $\\sigma^2$ does. We see\n",
" {in Figure 3.8 on page 108} that there is\n",
" {in Figure~\\ref{Ch3:polyplot} on page~\\pageref{Ch3:polyplot}} that there is\n",
"a non-linear relationship in the data, and so the residuals from a\n",
"linear fit will be inflated, and so will $\\hat{\\sigma}^2$. Secondly,\n",
"the standard formulas assume (somewhat unrealistically) that the $x_i$\n",
@@ -1063,7 +1192,7 @@
"Below we compute the bootstrap standard error estimates and the\n",
"standard linear regression estimates that result from fitting the\n",
"quadratic model to the data. Since this model provides a good fit to\n",
"the data (Figure 3.8), there is now a better\n",
"the data (Figure~\\ref{Ch3:polyplot}), there is now a better\n",
"correspondence between the bootstrap estimates and the standard\n",
"estimates of ${\\rm SE}(\\hat{\\beta}_0)$, ${\\rm SE}(\\hat{\\beta}_1)$ and\n",
"${\\rm SE}(\\hat{\\beta}_2)$."
@@ -1072,17 +1201,22 @@
{
"cell_type": "code",
"execution_count": 25,
"id": "79c56529",
"id": "acc3e32c",
"metadata": {
"execution": {}
"execution": {
"iopub.execute_input": "2024-06-04T23:19:17.222887Z",
"iopub.status.busy": "2024-06-04T23:19:17.222785Z",
"iopub.status.idle": "2024-06-04T23:19:19.351574Z",
"shell.execute_reply": "2024-06-04T23:19:19.351317Z"
}
},
"outputs": [
{
"data": {
"text/plain": [
"intercept 2.067840\n",
"poly(horsepower, degree=2, raw=True)[0] 0.033019\n",
"poly(horsepower, degree=2, raw=True)[1] 0.000120\n",
"intercept 1.538641\n",
"poly(horsepower, degree=2, raw=True)[0] 0.024696\n",
"poly(horsepower, degree=2, raw=True)[1] 0.000090\n",
"dtype: float64"
]
},
@@ -1101,7 +1235,7 @@
},
{
"cell_type": "markdown",
"id": "9fccbbbd",
"id": "e8a2fd2b",
"metadata": {},
"source": [
"We compare the results to the standard errors computed using `sm.OLS()`."
@@ -1110,9 +1244,14 @@
{
"cell_type": "code",
"execution_count": 26,
"id": "4d0b4edc",
"id": "dca5340c",
"metadata": {
"execution": {},
"execution": {
"iopub.execute_input": "2024-06-04T23:19:19.352904Z",
"iopub.status.busy": "2024-06-04T23:19:19.352827Z",
"iopub.status.idle": "2024-06-04T23:19:19.360147Z",
"shell.execute_reply": "2024-06-04T23:19:19.359948Z"
},
"lines_to_next_cell": 0
},
"outputs": [
@@ -1138,7 +1277,7 @@
},
{
"cell_type": "markdown",
"id": "9a86ff6e",
"id": "e98297be",
"metadata": {},
"source": [
"\n",
@@ -1149,8 +1288,13 @@
"metadata": {
"jupytext": {
"cell_metadata_filter": "-all",
"formats": "Rmd,ipynb",
"main_language": "python"
"main_language": "python",
"notebook_metadata_filter": "-all"
},
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
@@ -1162,7 +1306,7 @@
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.10.12"
"version": "3.12.3"
}
},
"nbformat": 4,

123
Ch06-varselect-lab.Rmd
View File

@@ -1,21 +1,13 @@
---
jupyter:
jupytext:
cell_metadata_filter: -all
formats: Rmd,ipynb
main_language: python
text_representation:
extension: .Rmd
format_name: rmarkdown
format_version: '1.2'
jupytext_version: 1.14.7
---
# Linear Models and Regularization Methods
<a target="_blank" href="https://colab.research.google.com/github/intro-stat-learning/ISLP_labs/blob/v2.2/Ch06-varselect-lab.ipynb">
<img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/>
</a>
[![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/intro-stat-learning/ISLP_labs/v2.2?labpath=Ch06-varselect-lab.ipynb)
# Chapter 6
# Lab: Linear Models and Regularization Methods
In this lab we implement many of the techniques discussed in this chapter.
We import some of our libraries at this top
level.
@@ -35,7 +27,7 @@ from functools import partial
```
We again collect the new imports
needed for this lab.
needed for this lab. Readers will also have to have installed `l0bnb` using `pip install l0bnb`.
```{python}
from sklearn.pipeline import Pipeline
@@ -45,11 +37,10 @@ from ISLP.models import \
(Stepwise,
sklearn_selected,
sklearn_selection_path)
# !pip install l0bnb
from l0bnb import fit_path
```
We have installed the package `l0bnb` on the fly. Note the escaped `!pip install` --- this is run as a separate system command.
## Subset Selection Methods
Here we implement methods that reduce the number of parameters in a
model by restricting the model to a subset of the input variables.
@@ -74,7 +65,7 @@ Hitters = load_data('Hitters')
np.isnan(Hitters['Salary']).sum()
```
We see that `Salary` is missing for 59 players. The
`dropna()` method of data frames removes all of the rows that have missing
values in any variable (by default --- see `Hitters.dropna?`).
@@ -84,9 +75,9 @@ Hitters = Hitters.dropna();
Hitters.shape
```
We first choose the best model using forward selection based on $C_p$ (6.2). This score
We first choose the best model using forward selection based on $C_p$ (\ref{Ch6:eq:cp}). This score
is not built in as a metric to `sklearn`. We therefore define a function to compute it ourselves, and use
it as a scorer. By default, `sklearn` tries to maximize a score, hence
our scoring function computes the negative $C_p$ statistic.
@@ -111,7 +102,7 @@ sigma2 = OLS(Y,X).fit().scale
```
The function `sklearn_selected()` expects a scorer with just three arguments --- the last three in the definition of `nCp()` above. We use the function `partial()` first seen in Section 5.3.3 to freeze the first argument with our estimate of $\sigma^2$.
The function `sklearn_selected()` expects a scorer with just three arguments --- the last three in the definition of `nCp()` above. We use the function `partial()` first seen in Section~\ref{Ch5-resample-lab:the-bootstrap} to freeze the first argument with our estimate of $\sigma^2$.
```{python}
neg_Cp = partial(nCp, sigma2)
@@ -119,7 +110,7 @@ neg_Cp = partial(nCp, sigma2)
```
We can now use `neg_Cp()` as a scorer for model selection.
Along with a score we need to specify the search strategy. This is done through the object
`Stepwise()` in the `ISLP.models` package. The method `Stepwise.first_peak()`
@@ -133,7 +124,7 @@ strategy = Stepwise.first_peak(design,
max_terms=len(design.terms))
```
We now fit a linear regression model with `Salary` as outcome using forward
selection. To do so, we use the function `sklearn_selected()` from the `ISLP.models` package. This takes
a model from `statsmodels` along with a search strategy and selects a model with its
@@ -147,7 +138,7 @@ hitters_MSE.fit(Hitters, Y)
hitters_MSE.selected_state_
```
Using `neg_Cp` results in a smaller model, as expected, with just 10 variables selected.
```{python}
@@ -158,7 +149,7 @@ hitters_Cp.fit(Hitters, Y)
hitters_Cp.selected_state_
```
### Choosing Among Models Using the Validation Set Approach and Cross-Validation
As an alternative to using $C_p$, we might try cross-validation to select a model in forward selection. For this, we need a
@@ -180,7 +171,7 @@ strategy = Stepwise.fixed_steps(design,
full_path = sklearn_selection_path(OLS, strategy)
```
We now fit the full forward-selection path on the `Hitters` data and compute the fitted values.
```{python}
@@ -189,8 +180,8 @@ Yhat_in = full_path.predict(Hitters)
Yhat_in.shape
```
This gives us an array of fitted values --- 20 steps in all, including the fitted mean for the null model --- which we can use to evaluate
in-sample MSE. As expected, the in-sample MSE improves each step we take,
indicating we must use either the validation or cross-validation
@@ -279,7 +270,7 @@ ax.legend()
mse_fig
```
To repeat the above using the validation set approach, we simply change our
`cv` argument to a validation set: one random split of the data into a test and training. We choose a test size
of 20%, similar to the size of each test set in 5-fold cross-validation.`skm.ShuffleSplit()`
@@ -309,7 +300,7 @@ ax.legend()
mse_fig
```
### Best Subset Selection
Forward stepwise is a *greedy* selection procedure; at each step it augments the current set by including one additional variable. We now apply best subset selection to the `Hitters`
@@ -337,7 +328,7 @@ path = fit_path(X,
max_nonzeros=X.shape[1])
```
The function `fit_path()` returns a list whose values include the fitted coefficients as `B`, an intercept as `B0`, as well as a few other attributes related to the particular path algorithm used. Such details are beyond the scope of this book.
```{python}
@@ -363,7 +354,7 @@ Since we
standardize first, in order to find coefficient
estimates on the original scale, we must *unstandardize*
the coefficient estimates. The parameter
$\lambda$ in (6.5) and (6.7) is called `alphas` in `sklearn`. In order to
$\lambda$ in (\ref{Ch6:ridge}) and (\ref{Ch6:LASSO}) is called `alphas` in `sklearn`. In order to
be consistent with the rest of this chapter, we use `lambdas`
rather than `alphas` in what follows. {At the time of publication, ridge fits like the one in code chunk [22] issue unwarranted convergence warning messages; we expect these to disappear as this package matures.}
@@ -405,7 +396,7 @@ soln_path.index.name = 'negative log(lambda)'
soln_path
```
We plot the paths to get a sense of how the coefficients vary with $\lambda$.
To control the location of the legend we first set `legend` to `False` in the
plot method, adding it afterward with the `legend()` method of `ax`.
@@ -429,14 +420,14 @@ beta_hat = soln_path.loc[soln_path.index[39]]
lambdas[39], beta_hat
```
Let’s compute the $\ell_2$ norm of the standardized coefficients.
```{python}
np.linalg.norm(beta_hat)
```
In contrast, here is the $\ell_2$ norm when $\lambda$ is 2.44e-01.
Note the much larger $\ell_2$ norm of the
coefficients associated with this smaller value of $\lambda$.
@@ -490,7 +481,7 @@ results = skm.cross_validate(ridge,
-results['test_score']
```
The test MSE is 1.342e+05. Note
that if we had instead simply fit a model with just an intercept, we
would have predicted each test observation using the mean of the
@@ -527,7 +518,7 @@ grid.best_params_['ridge__alpha']
grid.best_estimator_
```
Alternatively, we can use 5-fold cross-validation.
```{python}
@@ -540,7 +531,7 @@ grid.best_params_['ridge__alpha']
grid.best_estimator_
```
Recall we set up the `kfold` object for 5-fold cross-validation on page 298. We now plot the cross-validated MSE as a function of $-\log(\lambda)$, which has shrinkage decreasing from left
Recall we set up the `kfold` object for 5-fold cross-validation on page~\pageref{line:choos-among-models}. We now plot the cross-validated MSE as a function of $-\log(\lambda)$, which has shrinkage decreasing from left
to right.
```{python}
@@ -553,7 +544,7 @@ ax.set_xlabel('$-\log(\lambda)$', fontsize=20)
ax.set_ylabel('Cross-validated MSE', fontsize=20);
```
One can cross-validate different metrics to choose a parameter. The default
metric for `skl.ElasticNet()` is test $R^2$.
Let’s compare $R^2$ to MSE for cross-validation here.
@@ -565,7 +556,7 @@ grid_r2 = skm.GridSearchCV(pipe,
grid_r2.fit(X, Y)
```
Finally, let’s plot the results for cross-validated $R^2$.
```{python}
@@ -577,7 +568,7 @@ ax.set_xlabel('$-\log(\lambda)$', fontsize=20)
ax.set_ylabel('Cross-validated $R^2$', fontsize=20);
```
### Fast Cross-Validation for Solution Paths
The ridge, lasso, and elastic net can be efficiently fit along a sequence of $\lambda$ values, creating what is known as a *solution path* or *regularization path*. Hence there is specialized code to fit
@@ -597,7 +588,7 @@ pipeCV = Pipeline(steps=[('scaler', scaler),
pipeCV.fit(X, Y)
```
Let’s produce a plot again of the cross-validation error to see that
it is similar to using `skm.GridSearchCV`.
@@ -613,7 +604,7 @@ ax.set_xlabel('$-\log(\lambda)$', fontsize=20)
ax.set_ylabel('Cross-validated MSE', fontsize=20);
```
We see that the value of $\lambda$ that results in the
smallest cross-validation error is 1.19e-02, available
as the value `tuned_ridge.alpha_`. What is the test MSE
@@ -623,7 +614,7 @@ associated with this value of $\lambda$?
np.min(tuned_ridge.mse_path_.mean(1))
```
This represents a further improvement over the test MSE that we got
using $\lambda=4$. Finally, `tuned_ridge.coef_`
has the coefficients fit on the entire data set
@@ -640,7 +631,7 @@ not perform variable selection!
### Evaluating Test Error of Cross-Validated Ridge
Choosing $\lambda$ using cross-validation provides a single regression
estimator, similar to fitting a linear regression model as we saw in
Chapter 3. It is therefore reasonable to estimate what its test error
Chapter~\ref{Ch3:linreg}. It is therefore reasonable to estimate what its test error
is. We run into a problem here in that cross-validation will have
*touched* all of its data in choosing $\lambda$, hence we have no
further data to estimate test error. A compromise is to do an initial
@@ -679,7 +670,7 @@ results = skm.cross_validate(pipeCV,
```
### The Lasso
We saw that ridge regression with a wise choice of $\lambda$ can
@@ -728,13 +719,13 @@ ax.set_ylabel('Standardized coefficiients', fontsize=20);
```
The smallest cross-validated error is lower than the test set MSE of the null model
and of least squares, and very similar to the test MSE of 115526.71 of ridge
regression (page 305) with $\lambda$ chosen by cross-validation.
regression (page~\pageref{page:MSECVRidge}) with $\lambda$ chosen by cross-validation.
```{python}
np.min(tuned_lasso.mse_path_.mean(1))
```
Let’s again produce a plot of the cross-validation error.
@@ -759,7 +750,7 @@ variables.
tuned_lasso.coef_
```
As in ridge regression, we could evaluate the test error
of cross-validated lasso by first splitting into
test and training sets and internally running
@@ -770,17 +761,17 @@ this as an exercise.
## PCR and PLS Regression
### Principal Components Regression
Principal components regression (PCR) can be performed using
`PCA()` from the `sklearn.decomposition`
module. We now apply PCR to the `Hitters` data, in order to
predict `Salary`. Again, ensure that the missing values have
been removed from the data, as described in Section 6.5.1.
been removed from the data, as described in Section~\ref{Ch6-varselect-lab:lab-1-subset-selection-methods}.
We use `LinearRegression()` to fit the regression model
here. Note that it fits an intercept by default, unlike
the `OLS()` function seen earlier in Section 6.5.1.
the `OLS()` function seen earlier in Section~\ref{Ch6-varselect-lab:lab-1-subset-selection-methods}.
```{python}
pca = PCA(n_components=2)
@@ -791,7 +782,7 @@ pipe.fit(X, Y)
pipe.named_steps['linreg'].coef_
```
When performing PCA, the results vary depending
on whether the data has been *standardized* or not.
As in the earlier examples, this can be accomplished
@@ -805,7 +796,7 @@ pipe.fit(X, Y)
pipe.named_steps['linreg'].coef_
```
We can of course use CV to choose the number of components, by
using `skm.GridSearchCV`, in this
case fixing the parameters to vary the
@@ -820,7 +811,7 @@ grid = skm.GridSearchCV(pipe,
grid.fit(X, Y)
```
Let’s plot the results as we have for other methods.
```{python}
@@ -835,7 +826,7 @@ ax.set_xticks(n_comp[::2])
ax.set_ylim([50000,250000]);
```
We see that the smallest cross-validation error occurs when
17
components are used. However, from the plot we also see that the
@@ -859,18 +850,18 @@ cv_null = skm.cross_validate(linreg,
-cv_null['test_score'].mean()
```
The `explained_variance_ratio_`
attribute of our `PCA` object provides the *percentage of variance explained* in the predictors and in the response using
different numbers of components. This concept is discussed in greater
detail in Section 12.2.
detail in Section~\ref{Ch10:sec:pca}.
```{python}
pipe.named_steps['pca'].explained_variance_ratio_
```
Briefly, we can think of
this as the amount of information about the predictors
that is captured using $M$ principal components. For example, setting
@@ -893,7 +884,7 @@ pls = PLSRegression(n_components=2,
pls.fit(X, Y)
```
As was the case in PCR, we will want to
use CV to choose the number of components.
@@ -906,7 +897,7 @@ grid = skm.GridSearchCV(pls,
grid.fit(X, Y)
```
As for our other methods, we plot the MSE.
```{python}
@@ -921,7 +912,7 @@ ax.set_xticks(n_comp[::2])
ax.set_ylim([50000,250000]);
```
CV error is minimized at 12,
though there is little noticable difference between this point and a much lower number like 2 or 3 components.

11389
Ch06-varselect-lab.ipynb
View File

File diff suppressed because one or more lines are too long

120
Ch07-nonlin-lab.Rmd
View File

@@ -1,22 +1,14 @@
---
jupyter:
jupytext:
cell_metadata_filter: -all
formats: Rmd,ipynb
main_language: python
text_representation:
extension: .Rmd
format_name: rmarkdown
format_version: '1.2'
jupytext_version: 1.14.7
---
# Non-Linear Modeling
<a target="_blank" href="https://colab.research.google.com/github/intro-stat-learning/ISLP_labs/blob/v2.2/Ch07-nonlin-lab.ipynb">
<img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/>
</a>
[![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/intro-stat-learning/ISLP_labs/v2.2?labpath=Ch07-nonlin-lab.ipynb)
# Chapter 7
# Lab: Non-Linear Modeling
In this lab, we demonstrate some of the nonlinear models discussed in
this chapter. We use the `Wage` data as a running example, and show that many of the complex non-linear fitting procedures discussed can easily be implemented in \Python.
this chapter. We use the `Wage` data as a running example, and show that many of the complex non-linear fitting procedures discussed can easily be implemented in `Python`.
As usual, we start with some of our standard imports.
@@ -30,7 +22,7 @@ from ISLP.models import (summarize,
ModelSpec as MS)
from statsmodels.stats.anova import anova_lm
```
We again collect the new imports
needed for this lab. Many of these are developed specifically for the
`ISLP` package.
@@ -51,9 +43,9 @@ from ISLP.pygam import (approx_lam,
anova as anova_gam)
```
## Polynomial Regression and Step Functions
We start by demonstrating how Figure 7.1 can be reproduced.
We start by demonstrating how Figure~\ref{Ch7:fig:poly} can be reproduced.
Let's begin by loading the data.
```{python}
@@ -62,10 +54,10 @@ y = Wage['wage']
age = Wage['age']
```
Throughout most of this lab, our response is `Wage['wage']`, which
we have stored as `y` above.
As in Section 3.6.6, we will use the `poly()` function to create a model matrix
As in Section~\ref{Ch3-linreg-lab:non-linear-transformations-of-the-predictors}, we will use the `poly()` function to create a model matrix
that will fit a $4$th degree polynomial in `age`.
```{python}
@@ -74,16 +66,16 @@ M = sm.OLS(y, poly_age.transform(Wage)).fit()
summarize(M)
```
This polynomial is constructed using the function `poly()`,
which creates
a special *transformer* `Poly()` (using `sklearn` terminology
for feature transformations such as `PCA()` seen in Section 6.5.3) which
for feature transformations such as `PCA()` seen in Section \ref{Ch6-varselect-lab:principal-components-regression}) which
allows for easy evaluation of the polynomial at new data points. Here `poly()` is referred to as a *helper* function, and sets up the transformation; `Poly()` is the actual workhorse that computes the transformation. See also
the
discussion of transformations on
page 129.
page~\pageref{Ch3-linreg-lab:using-transformations-fit-and-transform}.
In the code above, the first line executes the `fit()` method
using the dataframe
@@ -96,7 +88,7 @@ on the second line, as well as in the plotting function developed below.
We now create a grid of values for `age` at which we want
predictions.
@@ -146,10 +138,10 @@ def plot_wage_fit(age_df,
We include an argument `alpha` to `ax.scatter()`
to add some transparency to the points. This provides a visual indication
of density. Notice the use of the `zip()` function in the
`for` loop above (see Section 2.3.8).
`for` loop above (see Section~\ref{Ch2-statlearn-lab:for-loops}).
We have three lines to plot, each with different colors and line
types. Here `zip()` conveniently bundles these together as
iterators in the loop. {In `Python` speak, an "iterator" is an object with a finite number of values, that can be iterated on, as in a loop.}
iterators in the loop. {In `Python`{} speak, an "iterator" is an object with a finite number of values, that can be iterated on, as in a loop.}
We now plot the fit of the fourth-degree polynomial using this
function.
@@ -164,7 +156,7 @@ plot_wage_fit(age_df,
With polynomial regression we must decide on the degree of
the polynomial to use. Sometimes we just wing it, and decide to use
second or third degree polynomials, simply to obtain a nonlinear fit. But we can
@@ -195,7 +187,7 @@ anova_lm(*[sm.OLS(y, X_).fit()
for X_ in Xs])
```
Notice the `*` in the `anova_lm()` line above. This
function takes a variable number of non-keyword arguments, in this case fitted models.
When these models are provided as a list (as is done here), it must be
@@ -220,8 +212,8 @@ that `poly()` creates orthogonal polynomials.
summarize(M)
```
Notice that the p-values are the same, and in fact the square of
the t-statistics are equal to the F-statistics from the
`anova_lm()` function; for example:
@@ -230,8 +222,8 @@ the t-statistics are equal to the F-statistics from the
(-11.983)**2
```
However, the ANOVA method works whether or not we used orthogonal
polynomials, provided the models are nested. For example, we can use
`anova_lm()` to compare the following three
@@ -246,10 +238,10 @@ XEs = [model.fit_transform(Wage)
anova_lm(*[sm.OLS(y, X_).fit() for X_ in XEs])
```
As an alternative to using hypothesis tests and ANOVA, we could choose
the polynomial degree using cross-validation, as discussed in Chapter 5.
the polynomial degree using cross-validation, as discussed in Chapter~\ref{Ch5:resample}.
Next we consider the task of predicting whether an individual earns
more than $250,000 per year. We proceed much as before, except
@@ -267,8 +259,8 @@ B = glm.fit()
summarize(B)
```
Once again, we make predictions using the `get_prediction()` method.
```{python}
@@ -277,7 +269,7 @@ preds = B.get_prediction(newX)
bands = preds.conf_int(alpha=0.05)
```
We now plot the estimated relationship.
```{python}
@@ -308,7 +300,7 @@ value do not cover each other up. This type of plot is often called a
*rug plot*.
In order to fit a step function, as discussed in
Section 7.2, we first use the `pd.qcut()`
Section~\ref{Ch7:sec:scolstep-function}, we first use the `pd.qcut()`
function to discretize `age` based on quantiles. Then we use `pd.get_dummies()` to create the
columns of the model matrix for this categorical variable. Note that this function will
include *all* columns for a given categorical, rather than the usual approach which drops one
@@ -319,8 +311,8 @@ cut_age = pd.qcut(age, 4)
summarize(sm.OLS(y, pd.get_dummies(cut_age)).fit())
```
Here `pd.qcut()` automatically picked the cutpoints based on the quantiles 25%, 50% and 75%, which results in four regions. We could also have specified our own
quantiles directly instead of the argument `4`. For cuts not based
on quantiles we would use the `pd.cut()` function.
@@ -340,7 +332,7 @@ evaluation functions are in the `scipy.interpolate` package;
we have simply wrapped them as transforms
similar to `Poly()` and `PCA()`.
In Section 7.4, we saw
In Section~\ref{Ch7:sec:scolr-splin}, we saw
that regression splines can be fit by constructing an appropriate
matrix of basis functions. The `BSpline()` function generates the
entire matrix of basis functions for splines with the specified set of
@@ -355,7 +347,7 @@ bs_age.shape
```
This results in a seven-column matrix, which is what is expected for a cubic-spline basis with 3 interior knots.
We can form this same matrix using the `bs()` object,
which facilitates adding this to a model-matrix builder (as in `poly()` versus its workhorse `Poly()`) described in Section 7.8.1.
which facilitates adding this to a model-matrix builder (as in `poly()` versus its workhorse `Poly()`) described in Section~\ref{Ch7-nonlin-lab:polynomial-regression-and-step-functions}.
We now fit a cubic spline model to the `Wage` data.
@@ -377,7 +369,7 @@ M = sm.OLS(y, Xbs).fit()
summarize(M)
```
Notice that there are 6 spline coefficients rather than 7. This is because, by default,
`bs()` assumes `intercept=False`, since we typically have an overall intercept in the model.
So it generates the spline basis with the given knots, and then discards one of the basis functions to account for the intercept.
@@ -435,7 +427,7 @@ deciding bin membership.
In order to fit a natural spline, we use the `NaturalSpline()`
transform with the corresponding helper `ns()`. Here we fit a natural spline with five
degrees of freedom (excluding the intercept) and plot the results.
@@ -453,7 +445,7 @@ plot_wage_fit(age_df,
'Natural spline, df=5');
```
## Smoothing Splines and GAMs
A smoothing spline is a special case of a GAM with squared-error loss
and a single feature. To fit GAMs in `Python` we will use the
@@ -464,7 +456,7 @@ of a model matrix with a particular smoothing operation:
`s` for smoothing spline; `l` for linear, and `f` for factor or categorical variables.
The argument `0` passed to `s` below indicates that this smoother will
apply to the first column of a feature matrix. Below, we pass it a
matrix with a single column: `X_age`. The argument `lam` is the penalty parameter $\lambda$ as discussed in Section 7.5.2.
matrix with a single column: `X_age`. The argument `lam` is the penalty parameter $\lambda$ as discussed in Section~\ref{Ch7:sec5.2}.
```{python}
X_age = np.asarray(age).reshape((-1,1))
@@ -472,7 +464,7 @@ gam = LinearGAM(s_gam(0, lam=0.6))
gam.fit(X_age, y)
```
The `pygam` library generally expects a matrix of features so we reshape `age` to be a matrix (a two-dimensional array) instead
of a vector (i.e. a one-dimensional array). The `-1` in the call to the `reshape()` method tells `numpy` to impute the
size of that dimension based on the remaining entries of the shape tuple.
@@ -495,7 +487,7 @@ ax.set_ylabel('Wage', fontsize=20);
ax.legend(title='$\lambda$');
```
The `pygam` package can perform a search for an optimal smoothing parameter.
```{python}
@@ -508,7 +500,7 @@ ax.legend()
fig
```
Alternatively, we can fix the degrees of freedom of the smoothing
spline using a function included in the `ISLP.pygam` package. Below we
find a value of $\lambda$ that gives us roughly four degrees of
@@ -523,8 +515,8 @@ age_term.lam = lam_4
degrees_of_freedom(X_age, age_term)
```
Let’s vary the degrees of freedom in a similar plot to above. We choose the degrees of freedom
as the desired degrees of freedom plus one to account for the fact that these smoothing
splines always have an intercept term. Hence, a value of one for `df` is just a linear fit.
@@ -554,7 +546,7 @@ The strength of generalized additive models lies in their ability to fit multiva
We now fit a GAM by hand to predict
`wage` using natural spline functions of `year` and `age`,
treating `education` as a qualitative predictor, as in (7.16).
treating `education` as a qualitative predictor, as in (\ref{Ch7:nsmod}).
Since this is just a big linear regression model
using an appropriate choice of basis functions, we can simply do this
using the `sm.OLS()` function.
@@ -636,10 +628,10 @@ ax.set_ylabel('Effect on wage')
ax.set_title('Partial dependence of year on wage', fontsize=20);
```
We now fit the model (7.16) using smoothing splines rather
We now fit the model (\ref{Ch7:nsmod}) using smoothing splines rather
than natural splines. All of the
terms in (7.16) are fit simultaneously, taking each other
terms in (\ref{Ch7:nsmod}) are fit simultaneously, taking each other
into account to explain the response. The `pygam` package only works with matrices, so we must convert
the categorical series `education` to its array representation, which can be found
with the `cat.codes` attribute of `education`. As `year` only has 7 unique values, we
@@ -728,7 +720,7 @@ gam_linear = LinearGAM(age_term +
gam_linear.fit(Xgam, y)
```
Notice our use of `age_term` in the expressions above. We do this because
earlier we set the value for `lam` in this term to achieve four degrees of freedom.
@@ -775,7 +767,7 @@ We can make predictions from `gam` objects, just like from
Yhat = gam_full.predict(Xgam)
```
In order to fit a logistic regression GAM, we use `LogisticGAM()`
from `pygam`.
@@ -786,7 +778,7 @@ gam_logit = LogisticGAM(age_term +
gam_logit.fit(Xgam, high_earn)
```
```{python}
fig, ax = subplots(figsize=(8, 8))
@@ -838,8 +830,8 @@ gam_logit_ = LogisticGAM(age_term +
gam_logit_.fit(Xgam_, high_earn_)
```
Let’s look at the effect of `education`, `year` and `age` on high earner status now that we’ve
removed those observations.
@@ -872,7 +864,7 @@ ax.set_ylabel('Effect on wage')
ax.set_title('Partial dependence of high earner status on age', fontsize=20);
```
## Local Regression
We illustrate the use of local regression using the `lowess()`

665
Ch07-nonlin-lab.ipynb
View File

File diff suppressed because one or more lines are too long

104
Ch08-baggboost-lab.Rmd
View File

@@ -1,23 +1,15 @@
---
jupyter:
jupytext:
cell_metadata_filter: -all
formats: Rmd,ipynb
main_language: python
text_representation:
extension: .Rmd
format_name: rmarkdown
format_version: '1.2'
jupytext_version: 1.14.7
---
# Chapter 8
# Tree-Based Methods
<a target="_blank" href="https://colab.research.google.com/github/intro-stat-learning/ISLP_labs/blob/v2.2/Ch08-baggboost-lab.ipynb">
<img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/>
</a>
[![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/intro-stat-learning/ISLP_labs/v2.2?labpath=Ch08-baggboost-lab.ipynb)
# Lab: Tree-Based Methods
We import some of our usual libraries at this top
level.
@@ -46,10 +38,10 @@ from sklearn.ensemble import \
from ISLP.bart import BART
```
## Fitting Classification Trees
We first use classification trees to analyze the `Carseats` data set.
In these data, `Sales` is a continuous variable, and so we begin
@@ -65,7 +57,7 @@ High = np.where(Carseats.Sales > 8,
"No")
```
We now use `DecisionTreeClassifier()` to fit a classification tree in
order to predict `High` using all variables but `Sales`.
To do so, we must form a model matrix as we did when fitting regression
@@ -93,13 +85,13 @@ clf = DTC(criterion='entropy',
clf.fit(X, High)
```
In our discussion of qualitative features in Section 3.3,
In our discussion of qualitative features in Section~\ref{ch3:sec3},
we noted that for a linear regression model such a feature could be
represented by including a matrix of dummy variables (one-hot-encoding) in the model
matrix, using the formula notation of `statsmodels`.
As mentioned in Section 8.1, there is a more
As mentioned in Section~\ref{Ch8:decison.tree.sec}, there is a more
natural way to handle qualitative features when building a decision
tree, that does not require such dummy variables; each split amounts to partitioning the levels into two groups.
However,
@@ -110,8 +102,8 @@ advantage of this approach; instead it simply treats the one-hot-encoded levels
accuracy_score(High, clf.predict(X))
```
With only the default arguments, the training error rate is
21%.
For classification trees, we can
@@ -129,8 +121,8 @@ resid_dev = np.sum(log_loss(High, clf.predict_proba(X)))
resid_dev
```
This is closely related to the *entropy*, defined in (8.7).
This is closely related to the *entropy*, defined in (\ref{Ch8:eq:cross-entropy}).
A small deviance indicates a
tree that provides a good fit to the (training) data.
@@ -161,13 +153,13 @@ print(export_text(clf,
show_weights=True))
```
In order to properly evaluate the performance of a classification tree
on these data, we must estimate the test error rather than simply
computing the training error. We split the observations into a
training set and a test set, build the tree using the training set,
and evaluate its performance on the test data. This pattern is
similar to that in Chapter 6, with the linear models
similar to that in Chapter~\ref{Ch6:varselect}, with the linear models
replaced here by decision trees --- the code for validation
is almost identical. This approach leads to correct predictions
for 68.5% of the locations in the test data set.
@@ -264,8 +256,8 @@ confusion = confusion_table(best_.predict(X_test),
confusion
```
Now 72.0% of the test observations are correctly classified, which is slightly worse than the error for the full tree (with 35 leaves). So cross-validation has not helped us much here; it only pruned off 5 leaves, at a cost of a slightly worse error. These results would change if we were to change the random number seeds above; even though cross-validation gives an unbiased approach to model selection, it does have variance.
@@ -283,7 +275,7 @@ feature_names = list(D.columns)
X = np.asarray(D)
```
First, we split the data into training and test sets, and fit the tree
to the training data. Here we use 30% of the data for the test set.
@@ -298,7 +290,7 @@ to the training data. Here we use 30% of the data for the test set.
random_state=0)
```
Having formed our training and test data sets, we fit the regression tree.
```{python}
@@ -310,7 +302,7 @@ plot_tree(reg,
ax=ax);
```
The variable `lstat` measures the percentage of individuals with
lower socioeconomic status. The tree indicates that lower
values of `lstat` correspond to more expensive houses.
@@ -334,7 +326,7 @@ grid = skm.GridSearchCV(reg,
G = grid.fit(X_train, y_train)
```
In keeping with the cross-validation results, we use the pruned tree
to make predictions on the test set.
@@ -343,8 +335,8 @@ best_ = grid.best_estimator_
np.mean((y_test - best_.predict(X_test))**2)
```
In other words, the test set MSE associated with the regression tree
is 28.07. The square root of
the MSE is therefore around
@@ -367,7 +359,7 @@ plot_tree(G.best_estimator_,
## Bagging and Random Forests
Here we apply bagging and random forests to the `Boston` data, using
the `RandomForestRegressor()` from the `sklearn.ensemble` package. Recall
@@ -380,8 +372,8 @@ bag_boston = RF(max_features=X_train.shape[1], random_state=0)
bag_boston.fit(X_train, y_train)
```
The argument `max_features` indicates that all 12 predictors should
be considered for each split of the tree --- in other words, that
bagging should be done. How well does this bagged model perform on
@@ -394,7 +386,7 @@ ax.scatter(y_hat_bag, y_test)
np.mean((y_test - y_hat_bag)**2)
```
The test set MSE associated with the bagged regression tree is
14.63, about half that obtained using an optimally-pruned single
tree. We could change the number of trees grown from the default of
@@ -425,8 +417,8 @@ y_hat_RF = RF_boston.predict(X_test)
np.mean((y_test - y_hat_RF)**2)
```
The test set MSE is 20.04;
this indicates that random forests did somewhat worse than bagging
in this case. Extracting the `feature_importances_` values from the fitted model, we can view the
@@ -440,7 +432,7 @@ feature_imp.sort_values(by='importance', ascending=False)
```
This
is a relative measure of the total decrease in node impurity that results from
splits over that variable, averaged over all trees (this was plotted in Figure 8.9 for a model fit to the `Heart` data).
splits over that variable, averaged over all trees (this was plotted in Figure~\ref{Ch8:fig:varimp} for a model fit to the `Heart` data).
The results indicate that across all of the trees considered in the
random forest, the wealth level of the community (`lstat`) and the
@@ -450,7 +442,7 @@ house size (`rm`) are by far the two most important variables.
## Boosting
Here we use `GradientBoostingRegressor()` from `sklearn.ensemble`
to fit boosted regression trees to the `Boston` data
@@ -469,7 +461,7 @@ boost_boston = GBR(n_estimators=5000,
boost_boston.fit(X_train, y_train)
```
We can see how the training error decreases with the `train_score_` attribute.
To get an idea of how the test error decreases we can use the
`staged_predict()` method to get the predicted values along the path.
@@ -492,7 +484,7 @@ ax.plot(plot_idx,
ax.legend();
```
We now use the boosted model to predict `medv` on the test set:
```{python}
@@ -500,11 +492,11 @@ y_hat_boost = boost_boston.predict(X_test);
np.mean((y_test - y_hat_boost)**2)
```
The test MSE obtained is 14.48,
similar to the test MSE for bagging. If we want to, we can
perform boosting with a different value of the shrinkage parameter
$\lambda$ in (8.10). The default value is 0.001, but
$\lambda$ in (\ref{Ch8:alphaboost}). The default value is 0.001, but
this is easily modified. Here we take $\lambda=0.2$.
```{python}
@@ -518,8 +510,8 @@ y_hat_boost = boost_boston.predict(X_test);
np.mean((y_test - y_hat_boost)**2)
```
In this case, using $\lambda=0.2$ leads to a almost the same test MSE
as when using $\lambda=0.001$.
@@ -527,7 +519,7 @@ as when using $\lambda=0.001$.
## Bayesian Additive Regression Trees
In this section we demonstrate a `Python` implementation of BART found in the
`ISLP.bart` package. We fit a model
@@ -540,8 +532,8 @@ bart_boston = BART(random_state=0, burnin=5, ndraw=15)
bart_boston.fit(X_train, y_train)
```
On this data set, with this split into test and training, we see that the test error of BART is similar to that of random forest.
```{python}
@@ -549,8 +541,8 @@ yhat_test = bart_boston.predict(X_test.astype(np.float32))
np.mean((y_test - yhat_test)**2)
```
We can check how many times each variable appeared in the collection of trees.
This gives a summary similar to the variable importance plot for boosting and random forests.

2166
Ch08-baggboost-lab.ipynb
View File

File diff suppressed because one or more lines are too long

96
Ch09-svm-lab.Rmd
View File

@@ -1,23 +1,15 @@
---
jupyter:
jupytext:
cell_metadata_filter: -all
formats: Rmd,ipynb
main_language: python
text_representation:
extension: .Rmd
format_name: rmarkdown
format_version: '1.2'
jupytext_version: 1.14.7
---
# Support Vector Machines
<a target="_blank" href="https://colab.research.google.com/github/intro-stat-learning/ISLP_labs/blob/v2.2/Ch09-svm-lab.ipynb">
<img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/>
</a>
[![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/intro-stat-learning/ISLP_labs/v2.2?labpath=Ch09-svm-lab.ipynb)
# Chapter 9
# Lab: Support Vector Machines
In this lab, we use the `sklearn.svm` library to demonstrate the support
vector classifier and the support vector machine.
@@ -39,7 +31,7 @@ from ISLP.svm import plot as plot_svm
from sklearn.metrics import RocCurveDisplay
```
We will use the function `RocCurveDisplay.from_estimator()` to
produce several ROC plots, using a shorthand `roc_curve`.
@@ -84,8 +76,8 @@ svm_linear = SVC(C=10, kernel='linear')
svm_linear.fit(X, y)
```
The support vector classifier with two features can
be visualized by plotting values of its *decision function*.
We have included a function for this in the `ISLP` package (inspired by a similar
@@ -99,7 +91,7 @@ plot_svm(X,
ax=ax)
```
The decision
boundary between the two classes is linear (because we used the
argument `kernel='linear'`). The support vectors are marked with `+`
@@ -126,8 +118,8 @@ coefficients of the linear decision boundary as follows:
svm_linear.coef_
```
Since the support vector machine is an estimator in `sklearn`, we
can use the usual machinery to tune it.
@@ -144,8 +136,8 @@ grid.fit(X, y)
grid.best_params_
```
We can easily access the cross-validation errors for each of these models
in `grid.cv_results_`. This prints out a lot of detail, so we
extract the accuracy results only.
@@ -166,7 +158,7 @@ y_test = np.array([-1]*10+[1]*10)
X_test[y_test==1] += 1
```
Now we predict the class labels of these test observations. Here we
use the best model selected by cross-validation in order to make the
predictions.
@@ -177,7 +169,7 @@ y_test_hat = best_.predict(X_test)
confusion_table(y_test_hat, y_test)
```
Thus, with this value of `C`,
70% of the test
observations are correctly classified. What if we had instead used
@@ -190,7 +182,7 @@ y_test_hat = svm_.predict(X_test)
confusion_table(y_test_hat, y_test)
```
In this case 60% of test observations are correctly classified.
We now consider a situation in which the two classes are linearly
@@ -205,7 +197,7 @@ fig, ax = subplots(figsize=(8,8))
ax.scatter(X[:,0], X[:,1], c=y, cmap=cm.coolwarm);
```
Now the observations are just barely linearly separable.
```{python}
@@ -214,7 +206,7 @@ y_hat = svm_.predict(X)
confusion_table(y_hat, y)
```
We fit the
support vector classifier and plot the resulting hyperplane, using a
very large value of `C` so that no observations are
@@ -240,7 +232,7 @@ y_hat = svm_.predict(X)
confusion_table(y_hat, y)
```
Using `C=0.1`, we again do not misclassify any training observations, but we
also obtain a much wider margin and make use of twelve support
vectors. These jointly define the orientation of the decision boundary, and since there are more of them, it is more stable. It seems possible that this model will perform better on test
@@ -254,7 +246,7 @@ plot_svm(X,
ax=ax)
```
## Support Vector Machine
In order to fit an SVM using a non-linear kernel, we once again use
@@ -264,9 +256,9 @@ kernel we use `kernel="poly"`, and to fit an SVM with a
radial kernel we use
`kernel="rbf"`. In the former case we also use the
`degree` argument to specify a degree for the polynomial kernel
(this is $d$ in (9.22)), and in the latter case we use
(this is $d$ in (\ref{Ch9:eq:polyd})), and in the latter case we use
`gamma` to specify a value of $\gamma$ for the radial basis
kernel (9.24).
kernel (\ref{Ch9:eq:radial}).
We first generate some data with a non-linear class boundary, as follows:
@@ -277,7 +269,7 @@ X[100:150] -= 2
y = np.array([1]*150+[2]*50)
```
Plotting the data makes it clear that the class boundary is indeed non-linear.
```{python}
@@ -285,11 +277,11 @@ fig, ax = subplots(figsize=(8,8))
ax.scatter(X[:,0],
X[:,1],
c=y,
cmap=cm.coolwarm)
cmap=cm.coolwarm);
```
The data is randomly split into training and testing groups. We then
fit the training data using the `SVC()` estimator with a
radial kernel and $\gamma=1$:
@@ -306,7 +298,7 @@ svm_rbf = SVC(kernel="rbf", gamma=1, C=1)
svm_rbf.fit(X_train, y_train)
```
The plot shows that the resulting SVM has a decidedly non-linear
boundary.
@@ -318,7 +310,7 @@ plot_svm(X_train,
ax=ax)
```
We can see from the figure that there are a fair number of training
errors in this SVM fit. If we increase the value of `C`, we
can reduce the number of training errors. However, this comes at the
@@ -335,7 +327,7 @@ plot_svm(X_train,
ax=ax)
```
We can perform cross-validation using `skm.GridSearchCV()` to select the
best choice of $\gamma$ and `C` for an SVM with a radial
kernel:
@@ -354,7 +346,7 @@ grid.fit(X_train, y_train)
grid.best_params_
```
The best choice of parameters under five-fold CV is achieved at `C=1`
and `gamma=0.5`, though several other values also achieve the same
value.
@@ -371,7 +363,7 @@ y_hat_test = best_svm.predict(X_test)
confusion_table(y_hat_test, y_test)
```
With these parameters, 12% of test
observations are misclassified by this SVM.
@@ -386,7 +378,7 @@ classifier, the fitted value for an observation $X= (X_1, X_2, \ldots,
X_p)^T$ takes the form $\hat{\beta}_0 + \hat{\beta}_1 X_1 +
\hat{\beta}_2 X_2 + \ldots + \hat{\beta}_p X_p$. For an SVM with a
non-linear kernel, the equation that yields the fitted value is given
in (9.23). The sign of the fitted value
in (\ref{Ch9:eq:svmip}). The sign of the fitted value
determines on which side of the decision boundary the observation
lies. Therefore, the relationship between the fitted value and the
class prediction for a given observation is simple: if the fitted
@@ -431,7 +423,7 @@ roc_curve(svm_flex,
ax=ax);
```
However, these ROC curves are all on the training data. We are really
more interested in the level of prediction accuracy on the test
data. When we compute the ROC curves on the test data, the model with
@@ -447,7 +439,7 @@ roc_curve(svm_flex,
fig;
```
Let’s look at our tuned SVM.
```{python}
@@ -466,7 +458,7 @@ for (X_, y_, c, name) in zip(
color=c)
```
## SVM with Multiple Classes
If the response is a factor containing more than two levels, then the
@@ -485,7 +477,7 @@ fig, ax = subplots(figsize=(8,8))
ax.scatter(X[:,0], X[:,1], c=y, cmap=cm.coolwarm);
```
We now fit an SVM to the data:
```{python}
@@ -521,7 +513,7 @@ Khan = load_data('Khan')
Khan['xtrain'].shape, Khan['xtest'].shape
```
This data set consists of expression measurements for 2,308
genes. The training and test sets consist of 63 and 20
observations, respectively.
@@ -540,7 +532,7 @@ confusion_table(khan_linear.predict(Khan['xtrain']),
Khan['ytrain'])
```
We see that there are *no* training
errors. In fact, this is not surprising, because the large number of
variables relative to the number of observations implies that it is
@@ -553,7 +545,7 @@ confusion_table(khan_linear.predict(Khan['xtest']),
Khan['ytest'])
```
We see that using `C=10` yields two test set errors on these data.

1286
Ch09-svm-lab.ipynb
View File

File diff suppressed because one or more lines are too long

65
Ch10-deeplearning-lab.Rmd
View File

@@ -1,13 +1,18 @@
# Deep Learning
<a target="_blank" href="https://colab.research.google.com/github/intro-stat-learning/ISLP_labs/blob/v2.2/Ch10-deeplearning-lab.ipynb">
<img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/>
</a>
[![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/intro-stat-learning/ISLP_labs/v2.2?labpath=Ch10-deeplearning-lab.ipynb)
# Chapter 10
# Lab: Deep Learning
In this section we demonstrate how to fit the examples discussed
in the text. We use the `Python` `torch` package, along with the
in the text. We use the `Python`{} `torch` package, along with the
`pytorch_lightning` package which provides utilities to simplify
fitting and evaluating models. This code can be impressively fast
with certain special processors, such as Apple’s new M1 chip. The package is well-structured, flexible, and will feel comfortable
to `Python` users. A good companion is the site
to `Python`{} users. A good companion is the site
[pytorch.org/tutorials](https://pytorch.org/tutorials/beginner/basics/intro.html).
Much of our code is adapted from there, as well as the `pytorch_lightning` documentation. {The precise URLs at the time of writing are <https://pytorch.org/tutorials/beginner/basics/intro.html> and <https://pytorch-lightning.readthedocs.io/en/latest/>.}
@@ -50,7 +55,13 @@ the `torchmetrics` package has utilities to compute
various metrics to evaluate performance when fitting
a model. The `torchinfo` package provides a useful
summary of the layers of a model. We use the `read_image()`
function when loading test images in Section 10.9.4.
function when loading test images in Section~\ref{Ch13-deeplearning-lab:using-pretrained-cnn-models}.
If you have not already installed the packages `torchvision`
and `torchinfo` you can install them by running
`pip install torchinfo torchvision`.
We can now import from `torchinfo`.
```{python}
from torchmetrics import (MeanAbsoluteError,
@@ -150,7 +161,7 @@ import json
## Single Layer Network on Hitters Data
We start by fitting the models in Section 10.6 on the `Hitters` data.
We start by fitting the models in Section~\ref{Ch13:sec:when-use-deep} on the `Hitters` data.
```{python}
Hitters = load_data('Hitters').dropna()
@@ -204,7 +215,7 @@ np.abs(Yhat_test - Y_test).mean()
Next we fit the lasso using `sklearn`. We are using
mean absolute error to select and evaluate a model, rather than mean squared error.
The specialized solver we used in Section 6.5.2 uses only mean squared error. So here, with a bit more work, we create a cross-validation grid and perform the cross-validation directly.
The specialized solver we used in Section~\ref{Ch6-varselect-lab:lab-2-ridge-regression-and-the-lasso} uses only mean squared error. So here, with a bit more work, we create a cross-validation grid and perform the cross-validation directly.
We encode a pipeline with two steps: we first normalize the features using a `StandardScaler()` transform,
and then fit the lasso without further normalization.
@@ -426,7 +437,7 @@ hit_module = SimpleModule.regression(hit_model,
```
By using the `SimpleModule.regression()` method, we indicate that we will use squared-error loss as in
(10.23).
(\ref{Ch13:eq:4}).
We have also asked for mean absolute error to be tracked as well
in the metrics that are logged.
@@ -463,7 +474,7 @@ hit_trainer = Trainer(deterministic=True,
hit_trainer.fit(hit_module, datamodule=hit_dm)
```
At each step of SGD, the algorithm randomly selects 32 training observations for
the computation of the gradient. Recall from Section 10.7
the computation of the gradient. Recall from Section~\ref{Ch13:sec:fitt-neur-netw}
that an epoch amounts to the number of SGD steps required to process $n$
observations. Since the training set has
$n=175$, and we specified a `batch_size` of 32 in the construction of `hit_dm`, an epoch is $175/32=5.5$ SGD steps.
@@ -701,12 +712,13 @@ mnist_logger = CSVLogger('logs', name='MNIST')
```
Now we are ready to go. The final step is to supply training data, and fit the model.
Now we are ready to go. The final step is to supply training data, and fit the model. We disable the progress bar below to avoid lengthy output in the browser when running.
```{python}
mnist_trainer = Trainer(deterministic=True,
max_epochs=30,
logger=mnist_logger,
enable_progress_bar=False,
callbacks=[ErrorTracker()])
mnist_trainer.fit(mnist_module,
datamodule=mnist_dm)
@@ -751,8 +763,8 @@ mnist_trainer.test(mnist_module,
datamodule=mnist_dm)
```
Table 10.1 also reports the error rates resulting from LDA (Chapter 4) and multiclass logistic
regression. For LDA we refer the reader to Section 4.7.3.
Table~\ref{Ch13:tab:mnist} also reports the error rates resulting from LDA (Chapter~\ref{Ch4:classification}) and multiclass logistic
regression. For LDA we refer the reader to Section~\ref{Ch4-classification-lab:linear-discriminant-analysis}.
Although we could use the `sklearn` function `LogisticRegression()` to fit
multiclass logistic regression, we are set up here to fit such a model
with `torch`.
@@ -776,6 +788,7 @@ mlr_logger = CSVLogger('logs', name='MNIST_MLR')
```{python}
mlr_trainer = Trainer(deterministic=True,
max_epochs=30,
enable_progress_bar=False,
callbacks=[ErrorTracker()])
mlr_trainer.fit(mlr_module, datamodule=mnist_dm)
```
@@ -856,7 +869,7 @@ for idx, (X_ ,Y_) in enumerate(cifar_dm.train_dataloader()):
Before we start, we look at some of the training images; similar code produced
Figure 10.5 on page 406. The example below also illustrates
Figure~\ref{Ch13:fig:cifar100} on page \pageref{Ch13:fig:cifar100}. The example below also illustrates
that `TensorDataset` objects can be indexed with integers --- we are choosing
random images from the training data by indexing `cifar_train`. In order to display correctly,
we must reorder the dimensions by a call to `np.transpose()`.
@@ -879,7 +892,7 @@ for i in range(5):
Here the `imshow()` method recognizes from the shape of its argument that it is a 3-dimensional array, with the last dimension indexing the three RGB color channels.
We specify a moderately-sized CNN for
demonstration purposes, similar in structure to Figure 10.8.
demonstration purposes, similar in structure to Figure~\ref{Ch13:fig:DeepCNN}.
We use several layers, each consisting of convolution, ReLU, and max-pooling steps.
We first define a module that defines one of these layers. As in our
previous examples, we overwrite the `__init__()` and `forward()` methods
@@ -995,6 +1008,7 @@ cifar_logger = CSVLogger('logs', name='CIFAR100')
cifar_trainer = Trainer(deterministic=True,
max_epochs=30,
logger=cifar_logger,
enable_progress_bar=False,
callbacks=[ErrorTracker()])
cifar_trainer.fit(cifar_module,
datamodule=cifar_dm)
@@ -1051,6 +1065,7 @@ try:
cifar_module.metrics[name] = metric.to('mps')
cifar_trainer_mps = Trainer(accelerator='mps',
deterministic=True,
enable_progress_bar=False,
max_epochs=30)
cifar_trainer_mps.fit(cifar_module,
datamodule=cifar_dm)
@@ -1066,7 +1081,7 @@ clauses; if it works, we get the speedup, if it fails, nothing happens.
## Using Pretrained CNN Models
We now show how to use a CNN pretrained on the `imagenet` database to classify natural
images, and demonstrate how we produced Figure 10.10.
images, and demonstrate how we produced Figure~\ref{Ch13:fig:homeimages}.
We copied six JPEG images from a digital photo album into the
directory `book_images`. These images are available
from the data section of <www.statlearning.com>, the ISLP book website. Download `book_images.zip`; when
@@ -1175,7 +1190,7 @@ del(cifar_test,
## IMDB Document Classification
We now implement models for sentiment classification (Section 10.4) on the `IMDB`
We now implement models for sentiment classification (Section~\ref{Ch13:sec:docum-class}) on the `IMDB`
dataset. As mentioned above code block~8, we are using
a preprocessed version of the `IMDB` dataset found in the
`keras` package. As `keras` uses `tensorflow`, a different
@@ -1299,6 +1314,7 @@ imdb_logger = CSVLogger('logs', name='IMDB')
imdb_trainer = Trainer(deterministic=True,
max_epochs=30,
logger=imdb_logger,
enable_progress_bar=False,
callbacks=[ErrorTracker()])
imdb_trainer.fit(imdb_module,
datamodule=imdb_dm)
@@ -1328,7 +1344,7 @@ matrix that is recognized by `sklearn.`
```
Similar to what we did in
Section 10.9.1,
Section~\ref{Ch13-deeplearning-lab:single-layer-network-on-hitters-data},
we construct a series of 50 values for the lasso reguralization parameter $\lambda$.
```{python}
@@ -1436,16 +1452,16 @@ del(imdb_model,
## Recurrent Neural Networks
In this lab we fit the models illustrated in
Section 10.5.
Section~\ref{Ch13:sec:recurr-neur-netw}.
### Sequential Models for Document Classification
Here we fit a simple LSTM RNN for sentiment prediction to
the `IMDb` movie-review data, as discussed in Section 10.5.1.
the `IMDb` movie-review data, as discussed in Section~\ref{Ch13:sec:sequ-models-docum}.
For an RNN we use the sequence of words in a document, taking their
order into account. We loaded the preprocessed
data at the beginning of
Section 10.9.5.
Section~\ref{Ch13-deeplearning-lab:imdb-document-classification}.
A script that details the preprocessing can be found in the
`ISLP` library. Notably, since more than 90% of the documents
had fewer than 500 words, we set the document length to 500. For
@@ -1519,6 +1535,7 @@ lstm_logger = CSVLogger('logs', name='IMDB_LSTM')
lstm_trainer = Trainer(deterministic=True,
max_epochs=20,
logger=lstm_logger,
enable_progress_bar=False,
callbacks=[ErrorTracker()])
lstm_trainer.fit(lstm_module,
datamodule=imdb_seq_dm)
@@ -1559,7 +1576,7 @@ del(lstm_model,
### Time Series Prediction
We now show how to fit the models in Section 10.5.2
We now show how to fit the models in Section~\ref{Ch13:sec:time-seri-pred}
for time series prediction.
We first load and standardize the data.
@@ -1750,6 +1767,7 @@ The results on the test data are very similar to the linear AR model.
```{python}
nyse_trainer = Trainer(deterministic=True,
max_epochs=200,
enable_progress_bar=False,
callbacks=[ErrorTracker()])
nyse_trainer.fit(nyse_module,
datamodule=nyse_dm)
@@ -1820,8 +1838,9 @@ and evaluate the test error. We see the test $R^2$ is a slight improvement over
```{python}
nl_trainer = Trainer(deterministic=True,
max_epochs=20,
callbacks=[ErrorTracker()])
max_epochs=20,
enable_progress_bar=False,
callbacks=[ErrorTracker()])
nl_trainer.fit(nl_module, datamodule=day_dm)
nl_trainer.test(nl_module, datamodule=day_dm)
```

1594
Ch10-deeplearning-lab.ipynb
View File

File diff suppressed because one or more lines are too long

118
Ch11-surv-lab.Rmd
View File

@@ -1,27 +1,19 @@
---
jupyter:
jupytext:
cell_metadata_filter: -all
formats: Rmd,ipynb
main_language: python
text_representation:
extension: .Rmd
format_name: rmarkdown
format_version: '1.2'
jupytext_version: 1.14.7
---
# Chapter 11
# Survival Analysis
<a target="_blank" href="https://colab.research.google.com/github/intro-stat-learning/ISLP_labs/blob/v2.2/Ch11-surv-lab.ipynb">
<img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/>
</a>
[![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/intro-stat-learning/ISLP_labs/v2.2?labpath=Ch11-surv-lab.ipynb)
# Lab: Survival Analysis
In this lab, we perform survival analyses on three separate data
sets. In Section 11.8.1 we analyze the `BrainCancer`
data that was first described in Section 11.3. In Section 11.8.2, we examine the `Publication`
data from Section 11.5.4. Finally, Section 11.8.3 explores
sets. In Section~\ref{brain.cancer.sec} we analyze the `BrainCancer`
data that was first described in Section~\ref{sec:KM}. In Section~\ref{time.to.pub.sec}, we examine the `Publication`
data from Section~\ref{sec:pub}. Finally, Section~\ref{call.center.sec} explores
a simulated call-center data set.
We begin by importing some of our libraries at this top
@@ -37,7 +29,7 @@ from ISLP.models import ModelSpec as MS
from ISLP import load_data
```
We also collect the new imports
needed for this lab.
@@ -61,7 +53,7 @@ BrainCancer = load_data('BrainCancer')
BrainCancer.columns
```
The rows index the 88 patients, while the 8 columns contain the predictors and outcome variables.
We first briefly examine the data.
@@ -69,20 +61,20 @@ We first briefly examine the data.
BrainCancer['sex'].value_counts()
```
```{python}
BrainCancer['diagnosis'].value_counts()
```
```{python}
BrainCancer['status'].value_counts()
```
Before beginning an analysis, it is important to know how the
`status` variable has been coded. Most software
uses the convention that a `status` of 1 indicates an
@@ -91,7 +83,7 @@ observation. But some scientists might use the opposite coding. For
the `BrainCancer` data set 35 patients died before the end of
the study, so we are using the conventional coding.
To begin the analysis, we re-create the Kaplan-Meier survival curve shown in Figure 11.2. The main
To begin the analysis, we re-create the Kaplan-Meier survival curve shown in Figure~\ref{fig:survbrain}. The main
package we will use for survival analysis
is `lifelines`.
The variable `time` corresponds to $y_i$, the time to the $i$th event (either censoring or
@@ -109,9 +101,9 @@ km_brain = km.fit(BrainCancer['time'], BrainCancer['status'])
km_brain.plot(label='Kaplan Meier estimate', ax=ax)
```
Next we create Kaplan-Meier survival curves that are stratified by
`sex`, in order to reproduce Figure 11.3.
`sex`, in order to reproduce Figure~\ref{fig:survbrain2}.
We do this using the `groupby()` method of a dataframe.
This method returns a generator that can
be iterated over in the `for` loop. In this case,
@@ -138,8 +130,8 @@ for sex, df in BrainCancer.groupby('sex'):
km_sex.plot(label='Sex=%s' % sex, ax=ax)
```
As discussed in Section 11.4, we can perform a
As discussed in Section~\ref{sec:logrank}, we can perform a
log-rank test to compare the survival of males to females. We use
the `logrank_test()` function from the `lifelines.statistics` module.
The first two arguments are the event times, with the second
@@ -152,8 +144,8 @@ logrank_test(by_sex['Male']['time'],
by_sex['Female']['status'])
```
The resulting $p$-value is $0.23$, indicating no evidence of a
difference in survival between the two sexes.
@@ -172,7 +164,7 @@ cox_fit = coxph().fit(model_df,
cox_fit.summary[['coef', 'se(coef)', 'p']]
```
The first argument to `fit` should be a data frame containing
at least the event time (the second argument `time` in this case),
as well as an optional censoring variable (the argument `status` in this case).
@@ -186,7 +178,7 @@ with no features as follows:
cox_fit.log_likelihood_ratio_test()
```
Regardless of which test we use, we see that there is no clear
evidence for a difference in survival between males and females. As
we learned in this chapter, the score test from the Cox model is
@@ -206,7 +198,7 @@ fit_all = coxph().fit(all_df,
fit_all.summary[['coef', 'se(coef)', 'p']]
```
The `diagnosis` variable has been coded so that the baseline
corresponds to HG glioma. The results indicate that the risk associated with HG glioma
is more than eight times (i.e. $e^{2.15}=8.62$) the risk associated
@@ -233,7 +225,7 @@ def representative(series):
modal_data = cleaned.apply(representative, axis=0)
```
We make four
copies of the column means and assign the `diagnosis` column to be the four different
diagnoses.
@@ -245,7 +237,7 @@ modal_df['diagnosis'] = levels
modal_df
```
We then construct the model matrix based on the model specification `all_MS` used to fit
the model, and name the rows according to the levels of `diagnosis`.
@@ -272,12 +264,12 @@ fig, ax = subplots(figsize=(8, 8))
predicted_survival.plot(ax=ax);
```
## Publication Data
The `Publication` data presented in Section 11.5.4 can be
The `Publication` data presented in Section~\ref{sec:pub} can be
found in the `ISLP` package.
We first reproduce Figure 11.5 by plotting the Kaplan-Meier curves
We first reproduce Figure~\ref{fig:lauersurv} by plotting the Kaplan-Meier curves
stratified on the `posres` variable, which records whether the
study had a positive or negative result.
@@ -291,7 +283,7 @@ for result, df in Publication.groupby('posres'):
km_result.plot(label='Result=%d' % result, ax=ax)
```
As discussed previously, the $p$-values from fitting Cox’s
proportional hazards model to the `posres` variable are quite
large, providing no evidence of a difference in time-to-publication
@@ -308,8 +300,8 @@ posres_fit = coxph().fit(posres_df,
posres_fit.summary[['coef', 'se(coef)', 'p']]
```
However, the results change dramatically when we include other
predictors in the model. Here we exclude the funding mechanism
variable.
@@ -322,7 +314,7 @@ coxph().fit(model.fit_transform(Publication),
'status').summary[['coef', 'se(coef)', 'p']]
```
We see that there are a number of statistically significant variables,
including whether the trial focused on a clinical endpoint, the impact
of the study, and whether the study had positive or negative results.
@@ -332,7 +324,7 @@ of the study, and whether the study had positive or negative results.
In this section, we will simulate survival data using the relationship
between cumulative hazard and
the survival function explored in Exercise 8.
the survival function explored in Exercise \ref{ex:all3}.
Our simulated data will represent the observed
wait times (in seconds) for 2,000 customers who have phoned a call
center. In this context, censoring occurs if a customer hangs up
@@ -372,7 +364,7 @@ model = MS(['Operators',
intercept=False)
X = model.fit_transform(D)
```
It is worthwhile to take a peek at the model matrix `X`, so
that we can be sure that we understand how the variables have been coded. By default,
the levels of categorical variables are sorted and, as usual, the first column of the one-hot encoding
@@ -382,7 +374,7 @@ of the variable is dropped.
X[:5]
```
Next, we specify the coefficients and the hazard function.
```{python}
@@ -403,7 +395,7 @@ that the risk of a call being answered at Center B is 0.74 times the
risk that it will be answered at Center A; in other words, the wait
times are a bit longer at Center B.
Recall from Section 2.3.7 the use of `lambda`
Recall from Section~\ref{Ch2-statlearn-lab:loading-data} the use of `lambda`
for creating short functions on the fly.
We use the function
`sim_time()` from the `ISLP.survival` package. This function
@@ -431,7 +423,7 @@ W = np.array([sim_time(l, cum_hazard, rng)
D['Wait time'] = np.clip(W, 0, 1000)
```
We now simulate our censoring variable, for which we assume
90% of calls were answered (`Failed==1`) before the
customer hung up (`Failed==0`).
@@ -443,13 +435,13 @@ D['Failed'] = rng.choice([1, 0],
D[:5]
```
```{python}
D['Failed'].mean()
```
We now plot Kaplan-Meier survival curves. First, we stratify by `Center`.
```{python}
@@ -462,7 +454,7 @@ for center, df in D.groupby('Center'):
ax.set_title("Probability of Still Being on Hold")
```
Next, we stratify by `Time`.
```{python}
@@ -475,7 +467,7 @@ for time, df in D.groupby('Time'):
ax.set_title("Probability of Still Being on Hold")
```
It seems that calls at Call Center B take longer to be answered than
calls at Centers A and C. Similarly, it appears that wait times are
longest in the morning and shortest in the evening hours. We can use a
@@ -488,8 +480,8 @@ multivariate_logrank_test(D['Wait time'],
D['Failed'])
```
Next, we consider the effect of `Time`.
```{python}
@@ -498,8 +490,8 @@ multivariate_logrank_test(D['Wait time'],
D['Failed'])
```
As in the case of a categorical variable with 2 levels, these
results are similar to the likelihood ratio test
from the Cox proportional hazards model. First, we
@@ -514,8 +506,8 @@ F = coxph().fit(X, 'Wait time', 'Failed')
F.log_likelihood_ratio_test()
```
Next, we look at the results for `Time`.
```{python}
@@ -527,8 +519,8 @@ F = coxph().fit(X, 'Wait time', 'Failed')
F.log_likelihood_ratio_test()
```
We find that differences between centers are highly significant, as
are differences between times of day.
@@ -544,8 +536,8 @@ fit_queuing = coxph().fit(
fit_queuing.summary[['coef', 'se(coef)', 'p']]
```
The $p$-values for Center B and evening time
are very small. It is also clear that the
hazard --- that is, the instantaneous risk that a call will be

471
Ch11-surv-lab.ipynb
View File

File diff suppressed because one or more lines are too long

122
Ch12-unsup-lab.Rmd
View File

@@ -1,20 +1,12 @@
---
jupyter:
jupytext:
cell_metadata_filter: -all
formats: Rmd,ipynb
main_language: python
text_representation:
extension: .Rmd
format_name: rmarkdown
format_version: '1.2'
jupytext_version: 1.14.7
---
# Unsupervised Learning
<a target="_blank" href="https://colab.research.google.com/github/intro-stat-learning/ISLP_labs/blob/v2.2/Ch12-unsup-lab.ipynb">
<img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/>
</a>
[![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/intro-stat-learning/ISLP_labs/v2.2?labpath=Ch12-unsup-lab.ipynb)
# Chapter 12
# Lab: Unsupervised Learning
In this lab we demonstrate PCA and clustering on several datasets.
As in other labs, we import some of our libraries at this top
level. This makes the code more readable, as scanning the first few
@@ -44,7 +36,7 @@ from scipy.cluster.hierarchy import \
from ISLP.cluster import compute_linkage
```
## Principal Components Analysis
In this lab, we perform PCA on `USArrests`, a data set in the
`R` computing environment.
@@ -58,22 +50,22 @@ USArrests = get_rdataset('USArrests').data
USArrests
```
The columns of the data set contain the four variables.
```{python}
USArrests.columns
```
We first briefly examine the data. We notice that the variables have vastly different means.
```{python}
USArrests.mean()
```
Dataframes have several useful methods for computing
column-wise summaries. We can also examine the
variance of the four variables using the `var()` method.
@@ -82,7 +74,7 @@ variance of the four variables using the `var()` method.
USArrests.var()
```
Not surprisingly, the variables also have vastly different variances.
The `UrbanPop` variable measures the percentage of the population
in each state living in an urban area, which is not a comparable
@@ -132,7 +124,7 @@ of the variables. In this case, since we centered and scaled the data with
pcaUS.mean_
```
The scores can be computed using the `transform()` method
of `pcaUS` after it has been fit.
@@ -150,7 +142,7 @@ principal component loading vector.
pcaUS.components_
```
The `biplot` is a common visualization method used with
PCA. It is not built in as a standard
part of `sklearn`, though there are python
@@ -170,7 +162,7 @@ for k in range(pcaUS.components_.shape[1]):
USArrests.columns[k])
```
Notice that this figure is a reflection of Figure 12.1 through the $y$-axis. Recall that the
Notice that this figure is a reflection of Figure~\ref{Ch10:fig:USArrests:obs} through the $y$-axis. Recall that the
principal components are only unique up to a sign change, so we can
reproduce that figure by flipping the
signs of the second set of scores and loadings.
@@ -191,14 +183,14 @@ for k in range(pcaUS.components_.shape[1]):
USArrests.columns[k])
```
The standard deviations of the principal component scores are as follows:
```{python}
scores.std(0, ddof=1)
```
The variance of each score can be extracted directly from the `pcaUS` object via
the `explained_variance_` attribute.
@@ -220,7 +212,7 @@ We can plot the PVE explained by each component, as well as the cumulative PVE.
plot the proportion of variance explained.
```{python}
# %%capture
%%capture
fig, axes = plt.subplots(1, 2, figsize=(15, 6))
ticks = np.arange(pcaUS.n_components_)+1
ax = axes[0]
@@ -247,7 +239,7 @@ ax.set_xticks(ticks)
fig
```
The result is similar to that shown in Figure 12.3. Note
The result is similar to that shown in Figure~\ref{Ch10:fig:USArrests:scree}. Note
that the method `cumsum()` computes the cumulative sum of
the elements of a numeric vector. For instance:
@@ -259,15 +251,15 @@ np.cumsum(a)
## Matrix Completion
We now re-create the analysis carried out on the `USArrests` data in
Section 12.3.
Section~\ref{Ch10:sec:princ-comp-with}.
We saw in Section 12.2.2 that solving the optimization
problem (12.6) on a centered data matrix $\bf X$ is
We saw in Section~\ref{ch10:sec2.2} that solving the optimization
problem~(\ref{Ch10:eq:mc2}) on a centered data matrix $\bf X$ is
equivalent to computing the first $M$ principal
components of the data. We use our scaled
and centered `USArrests` data as $\bf X$ below. The *singular value decomposition*
(SVD) is a general algorithm for solving
(12.6).
(\ref{Ch10:eq:mc2}).
```{python}
X = USArrests_scaled
@@ -320,14 +312,14 @@ Xna = X.copy()
Xna[r_idx, c_idx] = np.nan
```
Here the array `r_idx`
contains 20 integers from 0 to 49; this represents the states (rows of `X`) that are selected to contain missing values. And `c_idx` contains
20 integers from 0 to 3, representing the features (columns in `X`) that contain the missing values for each of the selected states.
We now write some code to implement Algorithm 12.1.
We now write some code to implement Algorithm~\ref{Ch10:alg:hardimpute}.
We first write a function that takes in a matrix, and returns an approximation to the matrix using the `svd()` function.
This will be needed in Step 2 of Algorithm 12.1.
This will be needed in Step 2 of Algorithm~\ref{Ch10:alg:hardimpute}.
```{python}
def low_rank(X, M=1):
@@ -336,7 +328,7 @@ def low_rank(X, M=1):
return L.dot(V[:M])
```
To conduct Step 1 of the algorithm, we initialize `Xhat` --- this is $\tilde{\bf X}$ in Algorithm 12.1 --- by replacing
To conduct Step 1 of the algorithm, we initialize `Xhat` --- this is $\tilde{\bf X}$ in Algorithm~\ref{Ch10:alg:hardimpute} --- by replacing
the missing values with the column means of the non-missing entries. These are stored in
`Xbar` below after running `np.nanmean()` over the row axis.
We make a copy so that when we assign values to `Xhat` below we do not also overwrite the
@@ -348,7 +340,7 @@ Xbar = np.nanmean(Xhat, axis=0)
Xhat[r_idx, c_idx] = Xbar[c_idx]
```
Before we begin Step 2, we set ourselves up to measure the progress of our
iterations:
@@ -366,11 +358,11 @@ a given element is `True` if the corresponding matrix element is missing. The no
because it allows us to access both the missing and non-missing entries. We store the mean of the squared non-missing elements in `mss0`.
We store the mean squared error of the non-missing elements of the old version of `Xhat` in `mssold` (which currently
agrees with `mss0`). We plan to store the mean squared error of the non-missing elements of the current version of `Xhat` in `mss`, and will then
iterate Step 2 of Algorithm 12.1 until the *relative error*, defined as
iterate Step 2 of Algorithm~\ref{Ch10:alg:hardimpute} until the *relative error*, defined as
`(mssold - mss) / mss0`, falls below `thresh = 1e-7`.
{Algorithm 12.1 tells us to iterate Step 2 until (12.14) is no longer decreasing. Determining whether (12.14) is decreasing requires us only to keep track of `mssold - mss`. However, in practice, we keep track of `(mssold - mss) / mss0` instead: this makes it so that the number of iterations required for Algorithm 12.1 to converge does not depend on whether we multiplied the raw data $\bf X$ by a constant factor.}
{Algorithm~\ref{Ch10:alg:hardimpute} tells us to iterate Step 2 until \eqref{Ch10:eq:mc6} is no longer decreasing. Determining whether \eqref{Ch10:eq:mc6} is decreasing requires us only to keep track of `mssold - mss`. However, in practice, we keep track of `(mssold - mss) / mss0` instead: this makes it so that the number of iterations required for Algorithm~\ref{Ch10:alg:hardimpute} to converge does not depend on whether we multiplied the raw data $\bf X$ by a constant factor.}
In Step 2(a) of Algorithm 12.1, we approximate `Xhat` using `low_rank()`; we call this `Xapp`. In Step 2(b), we use `Xapp` to update the estimates for elements in `Xhat` that are missing in `Xna`. Finally, in Step 2(c), we compute the relative error. These three steps are contained in the following `while` loop:
In Step 2(a) of Algorithm~\ref{Ch10:alg:hardimpute}, we approximate `Xhat` using `low_rank()`; we call this `Xapp`. In Step 2(b), we use `Xapp` to update the estimates for elements in `Xhat` that are missing in `Xna`. Finally, in Step 2(c), we compute the relative error. These three steps are contained in the following `while` loop:
```{python}
while rel_err > thresh:
@@ -387,7 +379,7 @@ while rel_err > thresh:
.format(count, mss, rel_err))
```
We see that after eight iterations, the relative error has fallen below `thresh = 1e-7`, and so the algorithm terminates. When this happens, the mean squared error of the non-missing elements equals 0.381.
Finally, we compute the correlation between the 20 imputed values
@@ -397,9 +389,9 @@ and the actual values:
np.corrcoef(Xapp[ismiss], X[ismiss])[0,1]
```
In this lab, we implemented Algorithm 12.1 ourselves for didactic purposes. However, a reader who wishes to apply matrix completion to their data might look to more specialized `Python` implementations.
In this lab, we implemented Algorithm~\ref{Ch10:alg:hardimpute} ourselves for didactic purposes. However, a reader who wishes to apply matrix completion to their data might look to more specialized `Python`{} implementations.
## Clustering
@@ -444,7 +436,7 @@ ax.scatter(X[:,0], X[:,1], c=kmeans.labels_)
ax.set_title("K-Means Clustering Results with K=2");
```
Here the observations can be easily plotted because they are
two-dimensional. If there were more than two variables then we could
instead perform PCA and plot the first two principal component score
@@ -470,7 +462,7 @@ We have used the `n_init` argument to run the $K$-means with 20
initial cluster assignments (the default is 10). If a
value of `n_init` greater than one is used, then $K$-means
clustering will be performed using multiple random assignments in
Step 1 of Algorithm 12.2, and the `KMeans()`
Step 1 of Algorithm~\ref{Ch10:alg:km}, and the `KMeans()`
function will report only the best results. Here we compare using
`n_init=1` to `n_init=20`.
@@ -486,7 +478,7 @@ kmeans1.inertia_, kmeans20.inertia_
```
Note that `kmeans.inertia_` is the total within-cluster sum
of squares, which we seek to minimize by performing $K$-means
clustering (12.17).
clustering \eqref{Ch10:eq:kmeans}.
We *strongly* recommend always running $K$-means clustering with
a large value of `n_init`, such as 20 or 50, since otherwise an
@@ -519,7 +511,7 @@ hc_comp = HClust(distance_threshold=0,
hc_comp.fit(X)
```
This computes the entire dendrogram.
We could just as easily perform hierarchical clustering with average or single linkage instead:
@@ -534,7 +526,7 @@ hc_sing = HClust(distance_threshold=0,
hc_sing.fit(X);
```
To use a precomputed distance matrix, we provide an additional
argument `metric="precomputed"`. In the code below, the first four lines computes the $50\times 50$ pairwise-distance matrix.
@@ -550,7 +542,7 @@ hc_sing_pre = HClust(distance_threshold=0,
hc_sing_pre.fit(D)
```
We use
`dendrogram()` from `scipy.cluster.hierarchy` to plot the dendrogram. However,
`dendrogram()` expects a so-called *linkage-matrix representation*
@@ -573,7 +565,7 @@ dendrogram(linkage_comp,
**cargs);
```
We may want to color branches of the tree above
and below a cut-threshold differently. This can be achieved
by changing the `color_threshold`. Let’s cut the tree at a height of 4,
@@ -587,7 +579,7 @@ dendrogram(linkage_comp,
above_threshold_color='black');
```
To determine the cluster labels for each observation associated with a
given cut of the dendrogram, we can use the `cut_tree()`
function from `scipy.cluster.hierarchy`:
@@ -607,7 +599,7 @@ or `height` to `cut_tree()`.
cut_tree(linkage_comp, height=5)
```
To scale the variables before performing hierarchical clustering of
the observations, we use `StandardScaler()` as in our PCA example:
@@ -651,7 +643,7 @@ dendrogram(linkage_cor, ax=ax, **cargs)
ax.set_title("Complete Linkage with Correlation-Based Dissimilarity");
```
## NCI60 Data Example
Unsupervised techniques are often used in the analysis of genomic
@@ -666,7 +658,7 @@ nci_labs = NCI60['labels']
nci_data = NCI60['data']
```
Each cell line is labeled with a cancer type. We do not make use of
the cancer types in performing PCA and clustering, as these are
unsupervised techniques. But after performing PCA and clustering, we
@@ -679,8 +671,8 @@ The data has 64 rows and 6830 columns.
nci_data.shape
```
We begin by examining the cancer types for the cell lines.
@@ -688,7 +680,7 @@ We begin by examining the cancer types for the cell lines.
nci_labs.value_counts()
```
### PCA on the NCI60 Data
@@ -703,7 +695,7 @@ nci_pca = PCA()
nci_scores = nci_pca.fit_transform(nci_scaled)
```
We now plot the first few principal component score vectors, in order
to visualize the data. The observations (cell lines) corresponding to
a given cancer type will be plotted in the same color, so that we can
@@ -739,7 +731,7 @@ to have pretty similar gene expression levels.
We can also plot the percent variance
explained by the principal components as well as the cumulative percent variance explained.
This is similar to the plots we made earlier for the `USArrests` data.
@@ -798,7 +790,7 @@ def plot_nci(linkage, ax, cut=-np.inf):
return hc
```
Let’s plot our results.
```{python}
@@ -830,7 +822,7 @@ pd.crosstab(nci_labs['label'],
pd.Series(comp_cut.reshape(-1), name='Complete'))
```
There are some clear patterns. All the leukemia cell lines fall in
one cluster, while the breast cancer cell lines are spread out over
@@ -844,7 +836,7 @@ plot_nci('Complete', ax, cut=140)
ax.axhline(140, c='r', linewidth=4);
```
The `axhline()` function draws a horizontal line line on top of any
existing set of axes. The argument `140` plots a horizontal
line at height 140 on the dendrogram; this is a height that
@@ -852,7 +844,7 @@ results in four distinct clusters. It is easy to verify that the
resulting clusters are the same as the ones we obtained in
`comp_cut`.
We claimed earlier in Section 12.4.2 that
We claimed earlier in Section~\ref{Ch10:subsec:hc} that
$K$-means clustering and hierarchical clustering with the dendrogram
cut to obtain the same number of clusters can yield very different
results. How do these `NCI60` hierarchical clustering results compare
@@ -866,7 +858,7 @@ pd.crosstab(pd.Series(comp_cut, name='HClust'),
pd.Series(nci_kmeans.labels_, name='K-means'))
```
We see that the four clusters obtained using hierarchical clustering
and $K$-means clustering are somewhat different. First we note
that the labels in the two clusterings are arbitrary. That is, swapping

1978
Ch12-unsup-lab.ipynb
View File

File diff suppressed because one or more lines are too long

114
Ch13-multiple-lab.Rmd
View File

@@ -1,22 +1,14 @@
---
jupyter:
jupytext:
cell_metadata_filter: -all
formats: Rmd,ipynb
main_language: python
text_representation:
extension: .Rmd
format_name: rmarkdown
format_version: '1.2'
jupytext_version: 1.14.7
---
# Multiple Testing
<a target="_blank" href="https://colab.research.google.com/github/intro-stat-learning/ISLP_labs/blob/v2.2/Ch13-multiple-lab.ipynb">
<img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/>
</a>
[![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/intro-stat-learning/ISLP_labs/v2.2?labpath=Ch13-multiple-lab.ipynb)
# Chapter 13
# Lab: Multiple Testing
We include our usual imports seen in earlier labs.
@@ -28,7 +20,7 @@ import statsmodels.api as sm
from ISLP import load_data
```
We also collect the new imports
needed for this lab.
@@ -60,7 +52,7 @@ true_mean = np.array([0.5]*50 + [0]*50)
X += true_mean[None,:]
```
To begin, we use `ttest_1samp()` from the
`scipy.stats` module to test $H_{0}: \mu_1=0$, the null
hypothesis that the first variable has mean zero.
@@ -70,7 +62,7 @@ result = ttest_1samp(X[:,0], 0)
result.pvalue
```
The $p$-value comes out to 0.931, which is not low enough to
reject the null hypothesis at level $\alpha=0.05$. In this case,
$\mu_1=0.5$, so the null hypothesis is false. Therefore, we have made
@@ -97,7 +89,7 @@ truth = pd.Categorical(true_mean == 0,
```
Since this is a simulated data set, we can create a $2 \times 2$ table
similar to Table 13.2.
similar to Table~\ref{Ch12:tab-hypotheses}.
```{python}
pd.crosstab(decision,
@@ -108,7 +100,7 @@ pd.crosstab(decision,
```
Therefore, at level $\alpha=0.05$, we reject 15 of the 50 false
null hypotheses, and we incorrectly reject 5 of the true null
hypotheses. Using the notation from Section 13.3, we have
hypotheses. Using the notation from Section~\ref{sec:fwer}, we have
$V=5$, $S=15$, $U=45$ and $W=35$.
We have set $\alpha=0.05$, which means that we expect to reject around
5% of the true null hypotheses. This is in line with the $2 \times 2$
@@ -146,12 +138,12 @@ pd.crosstab(decision,
## Family-Wise Error Rate
Recall from (13.5) that if the null hypothesis is true
Recall from \eqref{eq:FWER.indep} that if the null hypothesis is true
for each of $m$ independent hypothesis tests, then the FWER is equal
to $1-(1-\alpha)^m$. We can use this expression to compute the FWER
for $m=1,\ldots, 500$ and $\alpha=0.05$, $0.01$, and $0.001$.
We plot the FWER for these values of $\alpha$ in order to
reproduce Figure 13.2.
reproduce Figure~\ref{Ch12:fwer}.
```{python}
m = np.linspace(1, 501)
@@ -167,7 +159,7 @@ ax.legend()
ax.axhline(0.05, c='k', ls='--');
```
As discussed previously, even for moderate values of $m$ such as $50$,
the FWER exceeds $0.05$ unless $\alpha$ is set to a very low value,
such as $0.001$. Of course, the problem with setting $\alpha$ to such
@@ -189,7 +181,7 @@ for i in range(5):
fund_mini_pvals
```
The $p$-values are low for Managers One and Three, and high for the
other three managers. However, we cannot simply reject $H_{0,1}$ and
$H_{0,3}$, since this would fail to account for the multiple testing
@@ -219,8 +211,8 @@ reject, bonf = mult_test(fund_mini_pvals, method = "bonferroni")[:2]
reject
```
The $p$-values `bonf` are simply the `fund_mini_pvalues` multiplied by 5 and truncated to be less than
or equal to 1.
@@ -228,7 +220,7 @@ or equal to 1.
bonf, np.minimum(fund_mini_pvals * 5, 1)
```
Therefore, using Bonferroni’s method, we are able to reject the null hypothesis only for Manager
One while controlling FWER at $0.05$.
@@ -240,8 +232,8 @@ hypotheses for Managers One and Three at a FWER of $0.05$.
mult_test(fund_mini_pvals, method = "holm", alpha=0.05)[:2]
```
As discussed previously, Manager One seems to perform particularly
well, whereas Manager Two has poor performance.
@@ -250,8 +242,8 @@ well, whereas Manager Two has poor performance.
fund_mini.mean()
```
Is there evidence of a meaningful difference in performance between
these two managers? We can check this by performing a paired $t$-test using the `ttest_rel()` function
from `scipy.stats`:
@@ -261,7 +253,7 @@ ttest_rel(fund_mini['Manager1'],
fund_mini['Manager2']).pvalue
```
The test results in a $p$-value of 0.038,
suggesting a statistically significant difference.
@@ -269,7 +261,7 @@ However, we decided to perform this test only after examining the data
and noting that Managers One and Two had the highest and lowest mean
performances. In a sense, this means that we have implicitly
performed ${5 \choose 2} = 5(5-1)/2=10$ hypothesis tests, rather than
just one, as discussed in Section 13.3.2. Hence, we use the
just one, as discussed in Section~\ref{tukey.sec}. Hence, we use the
`pairwise_tukeyhsd()` function from
`statsmodels.stats.multicomp` to apply Tukey’s method
in order to adjust for multiple testing. This function takes
@@ -286,8 +278,8 @@ tukey = pairwise_tukeyhsd(returns, managers)
print(tukey.summary())
```
The `pairwise_tukeyhsd()` function provides confidence intervals
for the difference between each pair of managers (`lower` and
`upper`), as well as a $p$-value. All of these quantities have
@@ -317,7 +309,7 @@ for i, manager in enumerate(Fund.columns):
fund_pvalues[i] = ttest_1samp(Fund[manager], 0).pvalue
```
There are far too many managers to consider trying to control the FWER.
Instead, we focus on controlling the FDR: that is, the expected fraction of rejected null hypotheses that are actually false positives.
The `multipletests()` function (abbreviated `mult_test()`) can be used to carry out the Benjamini--Hochberg procedure.
@@ -327,7 +319,7 @@ fund_qvalues = mult_test(fund_pvalues, method = "fdr_bh")[1]
fund_qvalues[:10]
```
The *q-values* output by the
Benjamini--Hochberg procedure can be interpreted as the smallest FDR
threshold at which we would reject a particular null hypothesis. For
@@ -354,9 +346,9 @@ null hypotheses!
(fund_pvalues <= 0.1 / 2000).sum()
```
Figure 13.6 displays the ordered
Figure~\ref{Ch12:fig:BonferroniBenjamini} displays the ordered
$p$-values, $p_{(1)} \leq p_{(2)} \leq \cdots \leq p_{(2000)}$, for
the `Fund` dataset, as well as the threshold for rejection by the
Benjamini--Hochberg procedure. Recall that the Benjamini--Hochberg
@@ -384,8 +376,8 @@ else:
sorted_set_ = []
```
We now reproduce the middle panel of Figure 13.6.
We now reproduce the middle panel of Figure~\ref{Ch12:fig:BonferroniBenjamini}.
```{python}
fig, ax = plt.subplots()
@@ -399,12 +391,12 @@ ax.scatter(sorted_set_+1, sorted_[sorted_set_], c='r', s=20)
ax.axline((0, 0), (1,q/m), c='k', ls='--', linewidth=3);
```
## A Re-Sampling Approach
Here, we implement the re-sampling approach to hypothesis testing
using the `Khan` dataset, which we investigated in
Section 13.5. First, we merge the training and
Section~\ref{sec:permutations}. First, we merge the training and
testing data, which results in observations on 83 patients for
2,308 genes.
@@ -415,8 +407,8 @@ D['Y'] = pd.concat([Khan['ytrain'], Khan['ytest']])
D['Y'].value_counts()
```
There are four classes of cancer. For each gene, we compare the mean
expression in the second class (rhabdomyosarcoma) to the mean
expression in the fourth class (Burkitt’s lymphoma). Performing a
@@ -436,8 +428,8 @@ observedT, pvalue = ttest_ind(D2[gene_11],
observedT, pvalue
```
However, this $p$-value relies on the assumption that under the null
hypothesis of no difference between the two groups, the test statistic
follows a $t$-distribution with $29+25-2=52$ degrees of freedom.
@@ -462,15 +454,15 @@ for b in range(B):
D_null[n_:],
equal_var=True)
Tnull[b] = ttest_.statistic
(np.abs(Tnull) > np.abs(observedT)).mean()
(np.abs(Tnull) < np.abs(observedT)).mean()
```
This fraction, 0.0398,
is our re-sampling-based $p$-value.
It is almost identical to the $p$-value of 0.0412 obtained using the theoretical null distribution.
We can plot a histogram of the re-sampling-based test statistics in order to reproduce Figure 13.7.
We can plot a histogram of the re-sampling-based test statistics in order to reproduce Figure~\ref{Ch12:fig-permp-1}.
```{python}
fig, ax = plt.subplots(figsize=(8,8))
@@ -493,7 +485,7 @@ ax.set_xlabel("Null Distribution of Test Statistic");
The re-sampling-based null distribution is almost identical to the theoretical null distribution, which is displayed in red.
Finally, we implement the plug-in re-sampling FDR approach outlined in
Algorithm 13.4. Depending on the speed of your
Algorithm~\ref{Ch12:alg-plugin-fdr}. Depending on the speed of your
computer, calculating the FDR for all 2,308 genes in the `Khan`
dataset may take a while. Hence, we will illustrate the approach on a
random subset of 100 genes. For each gene, we first compute the
@@ -522,11 +514,11 @@ for j in range(m):
Tnull_vals[j,b] = ttest_.statistic
```
Next, we compute the number of rejected null hypotheses $R$, the
estimated number of false positives $\widehat{V}$, and the estimated
FDR, for a range of threshold values $c$ in
Algorithm 13.4. The threshold values are chosen
Algorithm~\ref{Ch12:alg-plugin-fdr}. The threshold values are chosen
using the absolute values of the test statistics from the 100 genes.
```{python}
@@ -540,7 +532,7 @@ for j in range(m):
FDRs[j] = V / R
```
Now, for any given FDR, we can find the genes that will be
rejected. For example, with FDR controlled at 0.1, we reject 15 of the
100 null hypotheses. On average, we would expect about one or two of
@@ -556,7 +548,7 @@ the genes whose estimated FDR is less than 0.1.
sorted(idx[np.abs(T_vals) >= cutoffs[FDRs < 0.1].min()])
```
At an FDR threshold of 0.2, more genes are selected, at the cost of having a higher expected
proportion of false discoveries.
@@ -564,9 +556,9 @@ proportion of false discoveries.
sorted(idx[np.abs(T_vals) >= cutoffs[FDRs < 0.2].min()])
```
The next line generates Figure 13.11, which is similar
to Figure 13.9,
The next line generates Figure~\ref{fig:labfdr}, which is similar
to Figure~\ref{Ch12:fig-plugin-fdr},
except that it is based on only a subset of the genes.
```{python}

407
Ch13-multiple-lab.ipynb
View File

File diff suppressed because one or more lines are too long

28
requirements.txt
View File

@@ -1,16 +1,16 @@
numpy==1.24.2
scipy==1.11.1
pandas==1.5.3
lxml==4.9.3
scikit-learn==1.3.0
joblib==1.3.1
statsmodels==0.14.0
lifelines==0.27.7
pygam==0.9.0
numpy==1.26.4
scipy==1.11.4
pandas==2.2.2
lxml==5.2.2
scikit-learn==1.5.0
joblib==1.4.2
statsmodels==0.14.2
lifelines==0.28.0
pygam==0.9.1
l0bnb==1.0.0
torch==2.0.1
torchvision==0.15.2
pytorch-lightning==2.0.6
torch==2.3.0
torchvision==0.18.0
pytorch-lightning==2.2.5
torchinfo==1.8.0
torchmetrics==1.0.1
ISLP==0.3.22
torchmetrics==1.4.0.post0
ISLP==0.4.0