daviddoji/pbdl-book
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

cleanup of airfoil case

Browse Source
This commit is contained in:
NT
2021-02-18 14:46:37 +08:00
parent 0625cb6b0b
commit 56dc78c32c
3 changed files with 85 additions and 87 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
diffphys-code-ns.ipynb
View File

@@ -640,7 +640,7 @@
"name": "python", "name": "python",
"nbconvert_exporter": "python", "nbconvert_exporter": "python",
"pygments_lexer": "ipython3", "pygments_lexer": "ipython3",
"version": "3.7.6" "version": "3.8.5"
} }
}, },
"nbformat": 4, "nbformat": 4,

120
overview-ns-forw-v2.ipynb
View File

File diff suppressed because one or more lines are too long

50
supervised-airfoils.ipynb
View File

@@ -135,9 +135,9 @@
" c = np.concatenate(c,axis=1)\n", " c = np.concatenate(c,axis=1)\n",
" display(Image.fromarray( cm.magma(c, bytes=True) ))\n", " display(Image.fromarray( cm.magma(c, bytes=True) ))\n",
"\n", "\n",
"num=72\n", "NUM=72\n",
"print(\"\\nHere are all 3 inputs are shown at the top (mask,in x, in y) \\nSide by side with the 3 output channels (p,vx,vy) at the bottom:\")\n", "print(\"\\nHere are all 3 inputs are shown at the top (mask,in x, in y) \\nSide by side with the 3 output channels (p,vx,vy) at the bottom:\")\n",
"showSbs(npfile[\"inputs\"][num],npfile[\"targets\"][num])\n" "showSbs(npfile[\"inputs\"][NUM],npfile[\"targets\"][NUM])\n"
] ]
}, },
{ {
@@ -173,14 +173,14 @@
} }
], ],
"source": [ "source": [
"# some global training parameters\n", "# some global training constants\n",
"\n", "\n",
"# number of training epochs\n", "# number of training epochs\n",
"epochs = 100\n", "EPOCHS = 100\n",
"# batch size\n", "# batch size\n",
"batch_size = 10\n", "BATCH_SIZE = 10\n",
"# learning rate\n", "# learning rate\n",
"lr = 0.00002\n", "LR = 0.00002\n",
"\n", "\n",
"class DfpDataset():\n", "class DfpDataset():\n",
" def __init__(self, inputs,targets): \n", " def __init__(self, inputs,targets): \n",
@@ -196,8 +196,8 @@
"tdata = DfpDataset(npfile[\"inputs\"],npfile[\"targets\"])\n", "tdata = DfpDataset(npfile[\"inputs\"],npfile[\"targets\"])\n",
"vdata = DfpDataset(npfile[\"vinputs\"],npfile[\"vtargets\"])\n", "vdata = DfpDataset(npfile[\"vinputs\"],npfile[\"vtargets\"])\n",
"\n", "\n",
"trainLoader = torch.utils.data.DataLoader(tdata, batch_size=batch_size, shuffle=True , drop_last=True) \n", "trainLoader = torch.utils.data.DataLoader(tdata, batch_size=BATCH_SIZE, shuffle=True , drop_last=True) \n",
"valiLoader = torch.utils.data.DataLoader(vdata, batch_size=batch_size, shuffle=False, drop_last=True) \n", "valiLoader = torch.utils.data.DataLoader(vdata, batch_size=BATCH_SIZE, shuffle=False, drop_last=True) \n",
"\n", "\n",
"print(\"Training & validation batches: {} , {}\".format(len(trainLoader),len(valiLoader) ))" "print(\"Training & validation batches: {} , {}\".format(len(trainLoader),len(valiLoader) ))"
] ]
@@ -326,9 +326,9 @@
"id": "QAl3VgKVQSI3" "id": "QAl3VgKVQSI3"
}, },
"source": [ "source": [
"Initialize net...\n", "Next, we can initialize an instance of the `DfpNet`.\n",
"\n", "\n",
"The `expo` parameter here controls the exponent for the feature maps of our Unet: this directly scales the network size (3 gives a model with ca. 150k parameters). This is relatively small for a generative model for $3 \\times 128^2 = \\text{ca. }49k$ outputs, but yields fast training times and prevents overfitting given the relatively small data set we're using here. Hence it's a good starting point." "Here, the `EXPO` parameter here controls the exponent for the feature maps of our Unet: this directly scales the network size (3 gives a model with ca. 150k parameters). This is relatively small for a generative model for $3 \\times 128^2 = \\text{ca. }49k$ outputs, but yields fast training times and prevents overfitting given the relatively small data set we're using here. Hence it's a good starting point."
] ]
}, },
{ {
@@ -426,10 +426,10 @@
], ],
"source": [ "source": [
"# channel exponent to control network size\n", "# channel exponent to control network size\n",
"expo = 3\n", "EXPO = 3\n",
"\n", "\n",
"# setup network\n", "# setup network\n",
"net = DfpNet(channelExponent=expo)\n", "net = DfpNet(channelExponent=EXPO)\n",
"#print(net) # to double check the details...\n", "#print(net) # to double check the details...\n",
"\n", "\n",
"model_parameters = filter(lambda p: p.requires_grad, net.parameters())\n", "model_parameters = filter(lambda p: p.requires_grad, net.parameters())\n",
@@ -441,10 +441,10 @@
"net.apply(weights_init)\n", "net.apply(weights_init)\n",
"\n", "\n",
"criterionL1 = nn.L1Loss()\n", "criterionL1 = nn.L1Loss()\n",
"optimizerG = optim.Adam(net.parameters(), lr=lr, betas=(0.5, 0.999), weight_decay=0.0)\n", "optimizerG = optim.Adam(net.parameters(), lr=LR, betas=(0.5, 0.999), weight_decay=0.0)\n",
"\n", "\n",
"targets = torch.autograd.Variable(torch.FloatTensor(batch_size, 3, 128, 128))\n", "targets = torch.autograd.Variable(torch.FloatTensor(BATCH_SIZE, 3, 128, 128))\n",
"inputs = torch.autograd.Variable(torch.FloatTensor(batch_size, 3, 128, 128))\n" "inputs = torch.autograd.Variable(torch.FloatTensor(BATCH_SIZE, 3, 128, 128))\n"
] ]
}, },
{ {
@@ -455,7 +455,7 @@
"source": [ "source": [
"## Training\n", "## Training\n",
"\n", "\n",
"Finally, we can train" "Finally, we can train the model. This step can take a while, as we'll go over all 320 samples 100 times, and continually evaluate the validation samples to keep track of how well we're doing."
] ]
}, },
{ {
@@ -584,11 +584,11 @@
"\n", "\n",
"if os.path.isfile(\"model\"):\n", "if os.path.isfile(\"model\"):\n",
" print(\"Found existing model, loading & skipping training\")\n", " print(\"Found existing model, loading & skipping training\")\n",
" net.load_state_dict(torch.load(doLoad)) # optionally, load existing model\n", " net.load_state_dict(torch.load(\"model\")) # optionally, load existing model\n",
"\n", "\n",
"else:\n", "else:\n",
" print(\"Training from scratch\")\n", " print(\"Training from scratch\")\n",
" for epoch in range(epochs):\n", " for epoch in range(EPOCHS):\n",
" net.train()\n", " net.train()\n",
" L1_accum = 0.0\n", " L1_accum = 0.0\n",
" for i, traindata in enumerate(trainLoader, 0):\n", " for i, traindata in enumerate(trainLoader, 0):\n",
@@ -622,7 +622,7 @@
" history_L1.append( L1_accum / len(trainLoader) )\n", " history_L1.append( L1_accum / len(trainLoader) )\n",
" history_L1val.append( L1val_accum / len(valiLoader) )\n", " history_L1val.append( L1val_accum / len(valiLoader) )\n",
"\n", "\n",
" if i<3 or i%20==0:\n", " if epoch<3 or epoch%20==0:\n",
" print( \"Epoch: {}, L1 train: {:7.5f}, L1 vali: {:7.5f}\".format(epoch, history_L1[-1], history_L1val[-1]) )\n", " print( \"Epoch: {}, L1 train: {:7.5f}, L1 vali: {:7.5f}\".format(epoch, history_L1[-1], history_L1val[-1]) )\n",
"\n", "\n",
" torch.save(net.state_dict(), \"model\" )\n", " torch.save(net.state_dict(), \"model\" )\n",
@@ -635,11 +635,9 @@
"id": "4KuUpJsSL3Jv" "id": "4KuUpJsSL3Jv"
}, },
"source": [ "source": [
"Yay, the model is trained...! \n", "Yay, the model is trained...! The losses nicely went down in terms of absolute values: With the standard settings from an initial value of around 0.2 for the validation loss, to ca. 0.02 after 100 epochs. \n",
"\n", "\n",
"The losses nicely went down in terms of absolute values, let's look at the graphs.\n", "Let's look at the graphs to get some intuition for how the trained progressed over time. This is typically important to identify longer-term trends in the training. In practice it's tricky to spot whether the overall trend of 100 or so noisy numbers in a command line log is going slightly up or down - this is much easier to spot in a visualization."
"\n",
"This is typically important to identify longer-term trends in the data. In practice it's tricky to spot whether the overall trend of 100 or so noisy numbers is going slightly up or down."
] ]
}, },
{ {
@@ -912,11 +910,11 @@
"\n", "\n",
"## Nex steps\n", "## Nex steps\n",
"\n", "\n",
"* Experiment with learning rate, dropout, and model size to reduce the error on the test set. How low can you get it with the given training data?\n", "* Experiment with learning rate, dropout, and model size to reduce the error on the test set. How small can you make it with the given training data?\n",
"\n", "\n",
"* As you'll see, it's a bit limited here what you can get out of this dataset, head over to [the main github repo of this project](https://github.com/thunil/Deep-Flow-Prediction) to download larger data sets, or generate own data\n", "* As you'll see, it's a bit limited here what you can get out of this dataset, head over to [the main github repo of this project](https://github.com/thunil/Deep-Flow-Prediction) to download larger data sets, or generate own data\n",
"\n", "\n",
"**TODO us: provide data with \"errors\" (nan & huge neg number in 1 cell), filter out to make model train...**\n" "**(TODO, us: for exercise, provide data with \"errors\" (nan & huge neg number in 1 cell), filter out to make model train...)**\n"
] ]
}, },
{ {
@@ -947,7 +945,7 @@
"name": "python", "name": "python",
"nbconvert_exporter": "python", "nbconvert_exporter": "python",
"pygments_lexer": "ipython3", "pygments_lexer": "ipython3",
"version": "3.7.6" "version": "3.8.5"
} }
}, },
"nbformat": 4, "nbformat": 4,