updated graphs for BNN code, removed phiflow warnings via json-cleanup script
This commit is contained in:
NT committed
1 parent
410454d156
commit
e24d1b3ecc
4 files changed
+1411
-1447
No files matched your search
+115
-130
@@ -11,8 +11,7 @@
|
||||
"## Overview\n",
|
||||
"\n",
|
||||
"We are now considering the same setup as in the notebook {doc}`supervised-airfoils`: A turbulent airflow around wing profiles, for which we'd like to know the average motion\n",
|
||||
"and pressure distribution around this airfoil for different Reynolds numbers and angles of attack. In the earlier notebook, we tackled this by completely bypassing any physical solver and instead training a neural network that learns the quantities of interest. Now, we want to extend this approach to the variational Bayesian Neural Networks (BNNs) of the previous section. In contrast to traditional networks, that learn a single point estimate for each weight value, BNNs aim at learning a _distribution_ over each weight parameter (e.g. a Gaussian with mean $\\mu$ and variance $\\sigma^{2}$). During a forward-pass, each parameter in the network is then sampled from its corresponding approximate posterior distribution $q_{\\phi}(\\theta)$. In that sense, the network parameters themselves are _random variables_ and each forward pass becomes _stochastic_, because for a given input the predictions will vary with every forward-pass. This allows to assess how _uncertain_ the network is: If the predictions vary a lot, we think that the network is uncertain about its output. ",
|
||||
"[[run in colab]](https://colab.research.google.com/github/tum-pbs/pbdl-book/blob/main/bayesian-code.ipynb)"
|
||||
"and pressure distribution around this airfoil for different Reynolds numbers and angles of attack. In the earlier notebook, we tackled this by completely bypassing any physical solver and instead training a neural network that learns the quantities of interest. Now, we want to extend this approach to the variational Bayesian Neural Networks (BNNs) of the previous section. In contrast to traditional networks, that learn a single point estimate for each weight value, BNNs aim at learning a _distribution_ over each weight parameter (e.g. a Gaussian with mean $\\mu$ and variance $\\sigma^{2}$). During a forward-pass, each parameter in the network is then sampled from its corresponding approximate posterior distribution $q_{\\phi}(\\theta)$. In that sense, the network parameters themselves are _random variables_ and each forward pass becomes _stochastic_, because for a given input the predictions will vary with every forward-pass. This allows to assess how _uncertain_ the network is: If the predictions vary a lot, we think that the network is uncertain about its output. [[run in colab]](https://colab.research.google.com/github/tum-pbs/pbdl-book/blob/main/bayesian-code.ipynb)"
|
||||
]
|
||||
},
|
||||
{
|
||||
@@ -33,7 +32,7 @@
|
||||
"base_uri": "https://localhost:8080/"
|
||||
},
|
||||
"id": "JwZudtWauiGa",
|
||||
"outputId": "d0f589f8-959e-40e1-d1c8-8fe06c33f36b"
|
||||
"outputId": "4b8b6f0c-ad97-4527-869c-622ad11c292b"
|
||||
},
|
||||
"outputs": [
|
||||
{
|
||||
@@ -103,10 +102,10 @@
|
||||
"metadata": {
|
||||
"colab": {
|
||||
"base_uri": "https://localhost:8080/",
|
||||
"height": 403
|
||||
"height": 383
|
||||
},
|
||||
"id": "1y1zHmKlAWNs",
|
||||
"outputId": "90ecb091-a733-4c2a-fd56-56447d32d414"
|
||||
"outputId": "2a3223d3-5708-48ee-8186-78d6c403145e"
|
||||
},
|
||||
"outputs": [
|
||||
{
|
||||
@@ -126,8 +125,7 @@
|
||||
]
|
||||
},
|
||||
"metadata": {
|
||||
"needs_background": "light",
|
||||
"tags": []
|
||||
"needs_background": "light"
|
||||
},
|
||||
"output_type": "display_data"
|
||||
}
|
||||
@@ -179,7 +177,7 @@
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 35,
|
||||
"execution_count": 3,
|
||||
"metadata": {
|
||||
"id": "9-8O6qJ-G_Qt"
|
||||
},
|
||||
@@ -236,7 +234,7 @@
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 36,
|
||||
"execution_count": 4,
|
||||
"metadata": {
|
||||
"id": "pGoGJwqdG_Qv"
|
||||
},
|
||||
@@ -307,7 +305,7 @@
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 37,
|
||||
"execution_count": 5,
|
||||
"metadata": {
|
||||
"id": "H23a0JAJclEN"
|
||||
},
|
||||
@@ -347,14 +345,14 @@
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 38,
|
||||
"execution_count": 6,
|
||||
"metadata": {
|
||||
"colab": {
|
||||
"base_uri": "https://localhost:8080/",
|
||||
"height": 316
|
||||
"height": 296
|
||||
},
|
||||
"id": "tDXZO0FN02do",
|
||||
"outputId": "d409105b-4d5e-4064-c2cf-c8a8fe2a6712"
|
||||
"outputId": "f19c0484-180c-4780-b7a0-cbc95d2c0fa9"
|
||||
},
|
||||
"outputs": [
|
||||
{
|
||||
@@ -363,10 +361,8 @@
|
||||
"Text(0, 0.5, 'Learning Rate')"
|
||||
]
|
||||
},
|
||||
"execution_count": 38,
|
||||
"metadata": {
|
||||
"tags": []
|
||||
},
|
||||
"execution_count": 6,
|
||||
"metadata": {},
|
||||
"output_type": "execute_result"
|
||||
},
|
||||
{
|
||||
@@ -377,8 +373,7 @@
|
||||
]
|
||||
},
|
||||
"metadata": {
|
||||
"needs_background": "light",
|
||||
"tags": []
|
||||
"needs_background": "light"
|
||||
},
|
||||
"output_type": "display_data"
|
||||
}
|
||||
@@ -401,21 +396,19 @@
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 39,
|
||||
"execution_count": 7,
|
||||
"metadata": {
|
||||
"colab": {
|
||||
"base_uri": "https://localhost:8080/"
|
||||
},
|
||||
"id": "LGaH2ZW73M1b",
|
||||
"outputId": "7c5d2487-88cd-4350-c0bc-b3f477a58cf9"
|
||||
"outputId": "13c2ada4-79bf-42f7-f2f9-4355c0173fb0"
|
||||
},
|
||||
"outputs": [
|
||||
{
|
||||
"name": "stderr",
|
||||
"output_type": "stream",
|
||||
"text": [
|
||||
" warnings.warn('`layer.add_variable` is deprecated and '\n"
|
||||
]
|
||||
"text": []
|
||||
},
|
||||
{
|
||||
"name": "stdout",
|
||||
@@ -450,27 +443,27 @@
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 40,
|
||||
"execution_count": 8,
|
||||
"metadata": {
|
||||
"colab": {
|
||||
"base_uri": "https://localhost:8080/"
|
||||
},
|
||||
"id": "VEQuKBegcd1K",
|
||||
"outputId": "4030fbb2-04aa-44aa-ff6a-230788669f4f"
|
||||
"outputId": "52d03473-3c95-4df6-9657-9d64c41348c8"
|
||||
},
|
||||
"outputs": [
|
||||
{
|
||||
"name": "stdout",
|
||||
"output_type": "stream",
|
||||
"text": [
|
||||
"Epoch 0/120, total loss: 4.274, KL loss: 4.118, MAE loss: 0.1559, MAE loss vali: 0.0773\n",
|
||||
"Epoch 1/120, total loss: 4.151, KL loss: 4.086, MAE loss: 0.0655, MAE loss vali: 0.0546\n",
|
||||
"Epoch 2/120, total loss: 4.102, KL loss: 4.050, MAE loss: 0.0526, MAE loss vali: 0.0473\n",
|
||||
"Epoch 20/120, total loss: 3.312, KL loss: 3.284, MAE loss: 0.0285, MAE loss vali: 0.0244\n",
|
||||
"Epoch 40/120, total loss: 2.445, KL loss: 2.421, MAE loss: 0.0246, MAE loss vali: 0.0231\n",
|
||||
"Epoch 60/120, total loss: 1.653, KL loss: 1.629, MAE loss: 0.0237, MAE loss vali: 0.0221\n",
|
||||
"Epoch 80/120, total loss: 1.130, KL loss: 1.108, MAE loss: 0.0224, MAE loss vali: 0.0221\n",
|
||||
"Epoch 100/120, total loss: 0.811, KL loss: 0.789, MAE loss: 0.0222, MAE loss vali: 0.0223\n"
|
||||
"Epoch 0/120, total loss: 4.265, KL loss: 4.118, MAE loss: 0.1464, MAE loss vali: 0.0872\n",
|
||||
"Epoch 1/120, total loss: 4.159, KL loss: 4.089, MAE loss: 0.0706, MAE loss vali: 0.0691\n",
|
||||
"Epoch 2/120, total loss: 4.115, KL loss: 4.054, MAE loss: 0.0610, MAE loss vali: 0.0589\n",
|
||||
"Epoch 20/120, total loss: 3.344, KL loss: 3.315, MAE loss: 0.0291, MAE loss vali: 0.0271\n",
|
||||
"Epoch 40/120, total loss: 2.495, KL loss: 2.471, MAE loss: 0.0245, MAE loss vali: 0.0242\n",
|
||||
"Epoch 60/120, total loss: 1.712, KL loss: 1.689, MAE loss: 0.0228, MAE loss vali: 0.0208\n",
|
||||
"Epoch 80/120, total loss: 1.190, KL loss: 1.169, MAE loss: 0.0212, MAE loss vali: 0.0200\n",
|
||||
"Epoch 100/120, total loss: 0.869, KL loss: 0.848, MAE loss: 0.0208, MAE loss vali: 0.0203\n"
|
||||
]
|
||||
}
|
||||
],
|
||||
@@ -536,38 +529,35 @@
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 41,
|
||||
"execution_count": 9,
|
||||
"metadata": {
|
||||
"colab": {
|
||||
"base_uri": "https://localhost:8080/",
|
||||
"height": 261
|
||||
"height": 233
|
||||
},
|
||||
"id": "3niTOL6CcrHo",
|
||||
"outputId": "6f499bed-11de-49c5-cb53-2068e5c31849"
|
||||
"outputId": "24acd9cc-c0aa-4848-bd19-e0721eb4b998"
|
||||
},
|
||||
"outputs": [
|
||||
{
|
||||
"data": {
|
||||
"text/plain": [
|
||||
"<matplotlib.legend.Legend at 0x7efd05319e50>"
|
||||
"<matplotlib.legend.Legend at 0x7f6285c60490>"
|
||||
]
|
||||
},
|
||||
"execution_count": 41,
|
||||
"metadata": {
|
||||
"tags": []
|
||||
},
|
||||
"execution_count": 9,
|
||||
"metadata": {},
|
||||
"output_type": "execute_result"
|
||||
},
|
||||
{
|
||||
"data": {
|
||||
"image/png": "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 truncated
|
||||
"image/png": "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 truncated
|
||||
"text/plain": [
|
||||
"<Figure size 1440x288 with 3 Axes>"
|
||||
]
|
||||
},
|
||||
"metadata": {
|
||||
"needs_background": "light",
|
||||
"tags": []
|
||||
"needs_background": "light"
|
||||
},
|
||||
"output_type": "display_data"
|
||||
}
|
||||
@@ -600,13 +590,13 @@
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 42,
|
||||
"execution_count": 10,
|
||||
"metadata": {
|
||||
"colab": {
|
||||
"base_uri": "https://localhost:8080/"
|
||||
},
|
||||
"id": "LusTjxWRdRjn",
|
||||
"outputId": "dd208675-b851-43ca-ef4f-13f7abc8ffbe"
|
||||
"outputId": "e4c401d3-c55f-49d2-b780-460bbdf47d0a"
|
||||
},
|
||||
"outputs": [
|
||||
{
|
||||
@@ -614,7 +604,7 @@
|
||||
"output_type": "stream",
|
||||
"text": [
|
||||
"There are 6 entries in model.losses\n",
|
||||
"[<tf.Tensor: shape=(), dtype=float32, numpy=0.033721145>, <tf.Tensor: shape=(), dtype=float32, numpy=0.062639065>, <tf.Tensor: shape=(), dtype=float32, numpy=0.26068285>, <tf.Tensor: shape=(), dtype=float32, numpy=0.08385982>, <tf.Tensor: shape=(), dtype=float32, numpy=0.06156125>, <tf.Tensor: shape=(), dtype=float32, numpy=0.056031853>]\n"
|
||||
"[<tf.Tensor: shape=(), dtype=float32, numpy=0.03536303>, <tf.Tensor: shape=(), dtype=float32, numpy=0.06506929>, <tf.Tensor: shape=(), dtype=float32, numpy=0.2647468>, <tf.Tensor: shape=(), dtype=float32, numpy=0.09337218>, <tf.Tensor: shape=(), dtype=float32, numpy=0.0795429>, <tf.Tensor: shape=(), dtype=float32, numpy=0.075103864>]\n"
|
||||
]
|
||||
}
|
||||
],
|
||||
@@ -643,7 +633,7 @@
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 43,
|
||||
"execution_count": 11,
|
||||
"metadata": {
|
||||
"id": "Z-gyY4LKG_Q7"
|
||||
},
|
||||
@@ -668,26 +658,25 @@
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 44,
|
||||
"execution_count": 12,
|
||||
"metadata": {
|
||||
"colab": {
|
||||
"base_uri": "https://localhost:8080/",
|
||||
"height": 341
|
||||
"height": 321
|
||||
},
|
||||
"id": "WSEXdGc7G_Q8",
|
||||
"outputId": "a517ad8d-ac77-481b-f8c5-8e628143c5f8"
|
||||
"outputId": "f0f35cfa-c9d9-408c-8a12-8b2540fe8df0"
|
||||
},
|
||||
"outputs": [
|
||||
{
|
||||
"data": {
|
||||
"image/png": "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 truncated
|
||||
"image/png": "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 truncated
|
||||
"text/plain": [
|
||||
"<Figure size 1152x360 with 2 Axes>"
|
||||
]
|
||||
},
|
||||
"metadata": {
|
||||
"needs_background": "light",
|
||||
"tags": []
|
||||
"needs_background": "light"
|
||||
},
|
||||
"output_type": "display_data"
|
||||
}
|
||||
@@ -711,26 +700,25 @@
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 45,
|
||||
"execution_count": 13,
|
||||
"metadata": {
|
||||
"colab": {
|
||||
"base_uri": "https://localhost:8080/",
|
||||
"height": 157
|
||||
"height": 132
|
||||
},
|
||||
"id": "gAcKTBCGmohC",
|
||||
"outputId": "6f954e63-68fa-4c29-fefb-a14ec40c16ea"
|
||||
"outputId": "9fe549dd-c74e-49fa-9c3f-3c1a260bdb00"
|
||||
},
|
||||
"outputs": [
|
||||
{
|
||||
"data": {
|
||||
"image/png": "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 truncated
|
||||
"image/png": "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 truncated
|
||||
"text/plain": [
|
||||
"<Figure size 1440x288 with 8 Axes>"
|
||||
]
|
||||
},
|
||||
"metadata": {
|
||||
"needs_background": "light",
|
||||
"tags": []
|
||||
"needs_background": "light"
|
||||
},
|
||||
"output_type": "display_data"
|
||||
}
|
||||
@@ -775,13 +763,13 @@
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 46,
|
||||
"execution_count": 14,
|
||||
"metadata": {
|
||||
"colab": {
|
||||
"base_uri": "https://localhost:8080/"
|
||||
},
|
||||
"id": "vQRbK594S2Ak",
|
||||
"outputId": "71eaae33-e9a1-4f5a-bdfc-a6bdd8719fed"
|
||||
"outputId": "7e3ea005-6d48-4308-8e53-002173f4ebc2"
|
||||
},
|
||||
"outputs": [
|
||||
{
|
||||
@@ -791,8 +779,8 @@
|
||||
"\n",
|
||||
"Average pixel prediction on validation set: \n",
|
||||
" pressure: 0.025 +- 0.009, \n",
|
||||
" ux: 0.459 +- 0.02,\n",
|
||||
" uy: 0.079 +- 0.016\n"
|
||||
" ux: 0.471 +- 0.019,\n",
|
||||
" uy: 0.081 +- 0.016\n"
|
||||
]
|
||||
}
|
||||
],
|
||||
@@ -827,19 +815,20 @@
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 47,
|
||||
"execution_count": 15,
|
||||
"metadata": {
|
||||
"colab": {
|
||||
"base_uri": "https://localhost:8080/"
|
||||
},
|
||||
"id": "ClH53HqtOw3L",
|
||||
"outputId": "d0241d6b-c3b5-4fe7-80e4-c6314a028b4a"
|
||||
"outputId": "10ad4fab-48ed-405b-9148-304e4624d29f"
|
||||
},
|
||||
"outputs": [
|
||||
{
|
||||
"name": "stdout",
|
||||
"output_type": "stream",
|
||||
"text": [
|
||||
"Downloading test data, this should be fast...\n",
|
||||
"Loaded 10/10 test samples\n"
|
||||
]
|
||||
}
|
||||
@@ -866,13 +855,13 @@
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 48,
|
||||
"execution_count": 16,
|
||||
"metadata": {
|
||||
"colab": {
|
||||
"base_uri": "https://localhost:8080/"
|
||||
},
|
||||
"id": "TP5D9rhkNlsf",
|
||||
"outputId": "dd20559d-3980-444a-f79e-0dfb6af798fd"
|
||||
"outputId": "5a316880-f5a9-4c4b-8962-59bc6350c788"
|
||||
},
|
||||
"outputs": [
|
||||
{
|
||||
@@ -880,7 +869,7 @@
|
||||
"output_type": "stream",
|
||||
"text": [
|
||||
"\n",
|
||||
"Average test error: 0.02462211523168565\n"
|
||||
"Average test error: 0.023046530292471824\n"
|
||||
]
|
||||
}
|
||||
],
|
||||
@@ -901,13 +890,13 @@
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 49,
|
||||
"execution_count": 17,
|
||||
"metadata": {
|
||||
"colab": {
|
||||
"base_uri": "https://localhost:8080/"
|
||||
},
|
||||
"id": "kLmWz_kPQ_RE",
|
||||
"outputId": "b6a110a4-1e3e-4e26-859f-a1cd67398b48"
|
||||
"outputId": "6656bda9-532c-4dce-a4b9-1c4b4c70ecbd"
|
||||
},
|
||||
"outputs": [
|
||||
{
|
||||
@@ -916,9 +905,9 @@
|
||||
"text": [
|
||||
"\n",
|
||||
"Average pixel prediction on test set: \n",
|
||||
" pressure: 0.029 +- 0.01, \n",
|
||||
" ux: 0.452 +- 0.026,\n",
|
||||
" uy: 0.084 +- 0.02\n"
|
||||
" pressure: 0.03 +- 0.012, \n",
|
||||
" ux: 0.466 +- 0.024,\n",
|
||||
" uy: 0.091 +- 0.02\n"
|
||||
]
|
||||
}
|
||||
],
|
||||
@@ -929,61 +918,56 @@
|
||||
"print(\"\\nAverage pixel prediction on test set: \\n pressure: {} +- {}, \\n ux: {} +- {},\\n uy: {} +- {}\".format(np.round(preds_test_mean_global[0],3),np.round(uncertainty_test_total[0],3),np.round(preds_test_mean_global[1],3),np.round(uncertainty_test_total[1],3),np.round(preds_test_mean_global[2],3),np.round(uncertainty_test_total[2],3)))\n"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"This is reassuring: The uncertainties on the OOD test set with new shapes are at least slightly higher than on the validation set. \n",
|
||||
"\n",
|
||||
"### Visualizations\n",
|
||||
"\n",
|
||||
"The following graph visualizes these measurements: it shows the mean absolute errors for validation and test sets side by side, together with the uncertainties of the predictions as error bars:"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 50,
|
||||
"execution_count": 18,
|
||||
"metadata": {
|
||||
"colab": {
|
||||
"base_uri": "https://localhost:8080/",
|
||||
"height": 361
|
||||
"height": 298
|
||||
},
|
||||
"id": "Sn8g_ACGsVgo",
|
||||
"outputId": "fd7015b9-457d-4235-a326-26436474050f"
|
||||
"outputId": "26c88b29-1eb9-4316-d5d5-1ab08b3f08f1"
|
||||
},
|
||||
"outputs": [
|
||||
{
|
||||
"data": {
|
||||
"image/png": "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 truncated
|
||||
"text/plain": [
|
||||
"<matplotlib.legend.Legend at 0x7efcf8c54150>"
|
||||
]
|
||||
},
|
||||
"execution_count": 50,
|
||||
"metadata": {
|
||||
"tags": []
|
||||
},
|
||||
"output_type": "execute_result"
|
||||
},
|
||||
{
|
||||
"data": {
|
||||
"image/png": "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 truncated
|
||||
"text/plain": [
|
||||
"<Figure size 648x360 with 3 Axes>"
|
||||
"<Figure size 432x288 with 1 Axes>"
|
||||
]
|
||||
},
|
||||
"metadata": {
|
||||
"needs_background": "light",
|
||||
"tags": []
|
||||
"needs_background": "light"
|
||||
},
|
||||
"output_type": "display_data"
|
||||
}
|
||||
],
|
||||
"source": [
|
||||
"# plot average pixel predictions with uncertainty\n",
|
||||
"max_unc = max(max(uncertainty_test_total),max(uncertainty_total))\n",
|
||||
"fig,axs=plt.subplots(nrows=1,ncols=3,figsize=(9,5))\n",
|
||||
"# plot per channel MAE with uncertainty\n",
|
||||
"val_loss_c, test_loss_c = [], []\n",
|
||||
"for channel in range(3):\n",
|
||||
" title = 'P' if channel ==0 else ('u_x' if channel ==1 else 'u_y')\n",
|
||||
" axs[channel].errorbar(0.5,preds_mean_global[channel],uncertainty_total[channel],linestyle='none',marker='o',markersize=8.5,label='val',linewidth=3.)\n",
|
||||
" axs[channel].errorbar(1.5,preds_test_mean_global[channel],uncertainty_test_total[channel],linestyle='none',marker='o',markersize=8.5,label='test',linewidth=3.)\n",
|
||||
" axs[channel].set_title(title)\n",
|
||||
" axs[channel].tick_params(\n",
|
||||
" axis='x', # changes apply to the x-axis\n",
|
||||
" which='both', # both major and minor ticks are affected\n",
|
||||
" bottom=False, # ticks along the bottom edge are off\n",
|
||||
" top=False, # ticks along the top edge are off\n",
|
||||
" labelbottom=False)\n",
|
||||
" axs[channel].set_ylim(preds_mean_global[channel]-1.5*max_unc,preds_mean_global[channel]+1.5*max_unc)\n",
|
||||
"axs[2].legend(loc='center')"
|
||||
" val_loss_c.append( tf.reduce_mean(mae(preds_mean[...,channel], y_val[...,channel])) )\n",
|
||||
" test_loss_c.append( tf.reduce_mean(mae(preds_test_mean[...,channel], y_test[...,channel])) )\n",
|
||||
"\n",
|
||||
"fig, ax = plt.subplots()\n",
|
||||
"ind = np.arange(len(val_loss_c)); width=0.3\n",
|
||||
"bars1 = ax.bar(ind - width/2, val_loss_c, width, yerr=uncertainty_total, capsize=4, label=\"validation\")\n",
|
||||
"bars2 = ax.bar(ind + width/2, test_loss_c, width, yerr=uncertainty_test_total, capsize=4, label=\"test\")\n",
|
||||
"ax.set_ylabel(\"MAE & Uncertainty\")\n",
|
||||
"ax.set_xticks(ind); ax.set_xticklabels(('P', 'u_x', 'u_y'))\n",
|
||||
"ax.legend(); plt.tight_layout()"
|
||||
]
|
||||
},
|
||||
{
|
||||
@@ -992,32 +976,34 @@
|
||||
"id": "icgfvAIqoMpE"
|
||||
},
|
||||
"source": [
|
||||
"This is reassuring: The uncertainties on the OOD test set with new shapes are at least slightly higher than on the validation set. The mean error is also larger for most cases.\n",
|
||||
"In general it is hard to obtain a calibrated uncertainty estimate, but since we are dealing with a fairly simple problem here, it seems that the BNN is able to estimate it reasonably well."
|
||||
"The mean error is clearly larger, and the slightly larger uncertainties of the predictions are likewise visible via the error bars.\n",
|
||||
"\n",
|
||||
"In general it is hard to obtain a calibrated uncertainty estimate, but since we are dealing with a fairly simple problem here, the BNN is able to estimate the uncertainty reasonably well.\n",
|
||||
"\n",
|
||||
"The next graph shows the differences of the BNN predictions for a single case of the test set (using the same style as for the validation sample above):"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 51,
|
||||
"execution_count": 19,
|
||||
"metadata": {
|
||||
"colab": {
|
||||
"base_uri": "https://localhost:8080/",
|
||||
"height": 157
|
||||
"height": 132
|
||||
},
|
||||
"id": "oP1HOciVjh-o",
|
||||
"outputId": "f3486c62-7641-47c5-d95f-575d53329607"
|
||||
"outputId": "932ab62b-f7cc-4626-8848-2d8f4328e532"
|
||||
},
|
||||
"outputs": [
|
||||
{
|
||||
"data": {
|
||||
"image/png": "iVBORw0KGgoAAAANSUhEUgAABG4AAACVCAYAAAAJ17yjAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAgAElEQVR4nOy9e7Q1y1Ef9que2fs8vnuli4Qe6AFiGQuLEAOOzWMZG2EgGGzFghhiwEaylx0gwQRiYmIvQ+wANsZekAQCImCMDBaPBIgFiQlgHubtADaPxIBxJJCQrh5X9+p+j3PO3jNd+aO6uqt7embPPt/Z53z3amqts+bMTE9Pz0zt7qpf/6qamBmLLLLIIossssgiiyyyyCKLLLLIIos8eOJuugGLLLLIIossssgiiyyyyCKLLLLIIovUZQFuFllkkUUWWWSRRRZZZJFFFllkkUUeUFmAm0UWWWSRRRZZZJFFFllkkUUWWWSRB1QW4GaRRRZZZJFFFllkkUUWWWSRRRZZ5AGVBbhZZJFFFllkkUUWWWSRRRZZZJFFFnlAZQFuFllkkUUWWWSRRRZZZJFFFllkkUUeUFmAm0UWOYAQ0bcR0VfcdDsWeWrJojeLXFYW3VnkMrLozSKXlUV3FrlpIaK/TUTfcdPteFCFiF5NRD990+1QIaIPIqJfJCK66bbsEiJiIvqA8P9riOhLr7j++G2I6IiIfoOInrPruqc8cENEd8yfJ6Izs/9Z19SGlxPRm6/jXotcnRDRG42+vC0YIQ9dw33fh4heT0RvCR3DSw59z0WuTm5Qb/4UEf00ET1BRI8S0bcQ0cOHvu8iVyc3qDsfS0S/FnTnMSL6fiJ64aHvu8jVyE3pTdGGb7WG7CJPDbnBPuflwSa3NvqrDn3fRfYXIvoJInqciI4OWP950IF3EtH3EdH7HOJeTzchoo8mop8loncT0buI6GeI6I+Ec/cFyhDRS0Kffsf0Dz9IRJ9wdU8wkC8H8A+ZmQ94jysXZv5cZv7yA9Z/AeBbAfy3u8o+5YEbZn5I/wD8LoBXmGP/dE4dRNQetpWLPMDyiqA7fwjAHwbwt8oCB9APD+CHAPynV1zvItcnN6E3zwTwFQBeAOBlAF4I4B9c8T0WObzchO78vwA+kZkfgejPvwPwjVd8j0UOKzehN1rvRwP4fYeoe5FrkZvSnbdYG52ZX3uAeyxyHxImDv8YAAbwnxzwVp8fdPClAB4B8LWVtiy+mBEiegaAHwTwdQCeBbH5/g6Aiyu+1SPh23wIgB8B8P1E9OorvgcCWPexAP73A9T9dNCd1wF41S4A9SkP3IwJEX04Ef1cmGF8KxF9PRGtzXkmov+SiP4dxIgFEf31UPYtRPSXC5rUERH9QyL63YBKvoaITojoFoB/DuAFBrV8wY089CKXFmb+Pch3/GBgVD/+NBH9m6BTP0tEf1CvJ6IPI6JfJqLbRPTdAI4n7vU2Zv4GAP/3QR9qkYPLNevN65j5h5j5HjM/DuCbAfzRQz7fIoeTG+hz3mIO9QAW5sRTUK5Tb0L5FuI4/NVDPdMi1yPXrTuLPCXkswH8PIBvA/AqIPo7TxDRB2shInoOCXPruWF/1F+aEmZ+F4DvRdLBNxLRlxDRrwK4S0QtEX1k0L0niOhXiOjlph3vT0Q/GXTwRwC895W9iQdPXgoAzPydzNwz8xkz/zAz/yoRvQzAawB8VPA7nwAAIno2CaP/SSL6V9gDcGfmR5n5fwTwtwH8fSJyoc4XENH3EtE7iOgNRPQF5vgZET1L6wh9xDuJaFW5xScA+GVmPjfl30hEX0xEv0rCKvpuIjo25/8KEf02Cdvo9da/LvsvCtEvQTffHvTzlUT0yUT0W6GOv2mun8QJrJAJCyWiH6BhtM+rw7k/QEQ/Eu71m0T06aaOyW/DzG8G8DiAj5z6Tk9b4AZimH4R5Ef9UQA+DsB/UZR5JYCPAPBBRPQnAfzXAD4eYtC+vCj7VZAf0YeG8y8E8GXMfBfAJyGfWXgLFnlKCRG9GMAnA/jX5rDVjw+D0Ng+B8CzAXwTgNeHAW4NQZC/HYKK/69Y2DTvEXLDevPHAfw/9/0Qi9yIXLfuENH7BuPuDMAXA/jqq32iRa5DbqDP+SIA/5KZf/VKH2SRa5cb0J3nkkx0voGIvpZkonORB0s+G8A/DX+fSETPC2Eb3wfgM0y5Twfwk8z89hn+0qgQ0XtD9Mbq4GcA+FMQJs7zAPwfEHbxsyBj1fdSyv3xOgC/BPHtvhwBbHqaym8B6InotUT0SUT0XnqCmf8tgM8F8HPB73wknPqfAZwDeB8Afyn87SvfB+C5AD4wgDc/AOBXIH7vxwH4QiL6xODr/hzyfuAzAfxvzLyt1PsfAvjNyvFPB/AnAbw/gD8I4NUAQER/AsDfC+ffB8DvAPiu4trYf4X950MA5RcC+DLIBOefB/AfQZhlX0pE7x/KzsEJBsLMMbIHwKcBeBTAvwj9249AdPS5AP4cgG8gIm3bnG/zbyHMp8kGPG3+ALwRwMePnPtCAN9v9hnAnzD73wrg75n9DwhlPgAAAbgL4PeZ8x8F4A3h/5cDePNNP//ydyl9uQPgCUiH8A0ATkb04xsBfHlx/W8C+BiIA/0WAGTO/SyAr9hx/zbc5yU3/S6Wv6eO3oRynwBB5l960+9j+XvK6c6zAHwJgI+86fex/D3YegPgxQB+G8Azzb0+4Kbfx/L3lNCd50OcKQdxyP4lgG+66fex/GXf6KMBbAG8d9j/DQBfFP7/eAD/3pT9GQCfHf4f9ZdG7vMTAO4FHfw9CEj0HKOff8mU/RIA315c/39BAJr3BdABuGXOvQ7Ad9z0uzzgN3oZhA315vDsrwfwvHDu1QB+2pRtwvf8A+bY37VlirpfEr5bWxw/Dsf/KAQU+d3i/N8A8I/D/38ZwI+F/wnAmwD88ZH7fTOAryqOvRHAnzf7Xw3gNeH/fwTgq825h8LzvSTsl/3XyyETU03YfziU+QhT5pcAvHKkfTWc4APC/9+Goq+DkDneDuCjw/5/BuCnijLfBOC/m/ttwm/jy6Z04ukQE1YVInopgK+BxPOeQpzkXyqKvcn8/wIAvzhy7jmhjl+ilAibIB9ikae2vJKZf3TknNWB94PEHlq6+BqiNwzg9zj86oL8ztU2c5EHTG5Mb4joIyHGyp9l5t/ar9mLPAByo30OM7+LiF4L4FeI6IXM3O3R9kVuTm5Cb/4HAP89M7/7Mg1e5IGRa9cdZn4UMhMNAG8gor8OydfxOfs2fpGDyasA/DAzvzPsvy4c+1oAPw7glIg+AsDbINEG3x/KTflLY/IFzPwtI+dKHfw0InqFObYK7XkBgMdZIh1UfgcCMD8thYVZ82pAwnAAfAekX/6MSvHnQHxd+z4v44vowgXvgrBkXqChWEEaAD8V/v9eAF9Hkr/mpZAcnj+FujwOAVNKedT8fw/ynRG2v6wnmPkOET0W2vfGcLjUvceYuQ//n4Xt28z5MwgANBcnqAoRPRPAPwPwt5hZE0S/H4CPKN5VC2Epzv02D0MAzlF5OodKfSMEPf79zPwMAH8TArZYsQPQWwG8yOzbjuCdkI/9HzDzI+HvmSw0qbKeRZ4+Yr/rmwB8pfn+jzDzKTN/J0R3XkiULW/3vtfa0kUeJDmY3gQq++shM1T/4spbvshNy3X1OS2EyvuM+2/yIg+AHEpvPg7APyBZxU6N658jos+82uYvcoNyXX0O4+ntczylhIhOICEoH2N+318E4EOI6EOC8/s9EIDgMwD8IDPfDpdP+UuXkVIHv73QwVvM/FXhvu9VhNy9x9jazPwbEOaH5h4qfc93QFg59ntc5v18CoRJ8puQ7/GG4ns8zMyfHNr0OIAfhrBNPhPAdxXgrpVfRcjbM1PeAgFDAADhuz8bwtpSuR//ew5OMJAQPvY6AD/OzP+LOfUmSDihfVcPMfPnYf63eRkkLG1Uns6d6MMAngRwJ6CUn7ej/PcA+ItE9DIiOgUQ12tnZg+heH0tpcRcLySiTwxF3gbg2QGBW+TpKd8M4HOJ6CNI5BbJ8swPQ2I8OwBfQEQrIvpUAB8+VRlJ8i3NHH5EJhnXIk8ruTK9IUkU+EMA/ioz/8C1tH6Rm5Sr1J1PJaIPJCIXcgV8DYB/zZIocpGnl1zlWPVSSLz9h4Y/AHgF0sz7Ik8vuco+52OJ6P1CPS+G5In8Z9fyFIvMkVdCcnx8ENLv+2UQtsRnhzKvgzjknxX+Vxn1l65AvgPAK4joE4moIaJjkqSzL2Lm34Ewff4OEa1JVrt7xXR1T10hSXT714joRWH/xRAQ7edDkbcBeBGFhLoBbPs+AH+biE5Jcqu8ao/7PY+IPh8S2vM3gu/7rwDcJkkgfRK+yQdTWJI8yOsgOvNnketJKT8C4A/t4e98J0TPPpRkpaW/C+AXmPmNc59ph+yLE6h8JYBbAP6r4vgPAngpEf2F0EeuiOiPENHL5nwbInohJJT95zEhT2fg5osh6N9tyGD03VOFmfmfA/ifIHS830Z6cbrs2pfocSJ6EsCPAvjAcO1vQBTs/yPJTr2sKvU0E2b+RQB/BcDXQ+h+v41AX2TmDYBPDfvvggx037ejyjNI3DkgiO/ZRNlFnqJyxXrz1yB0y39EKZv9kpz4aSpXrDsvhIB+twH8GoTO/CmHafkiNylXqTfM/HaWlUYeDaEvAPBOZl7Gq6ehXHGf82GQHDh3w/bXAHzBYVq+yCXkVZA8Jb9b/Ma/HsBnEVHLzL8A+X4vgKxGBmCWv3RpYeY3AfgzEPbDOyAshv8GyV/9TEjelXdBAIZ/cr/3fIDlNuRZf4GI7kLe869DbEEA+DHIAhWPEpGGu30+JBToUQg75x/PuM8Tof5fgyQv/zRm/lYggkF/GgLsvQESgfItACxR4fUAfj+AR5l5lC3CzG8Lbf4zM9qEEN75pZBwrLdCVmH6c3OunSl74QRGPgOy8tPLine truncated
|
||||
"image/png": "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 truncated
|
||||
"text/plain": [
|
||||
"<Figure size 1440x288 with 8 Axes>"
|
||||
]
|
||||
},
|
||||
"metadata": {
|
||||
"needs_background": "light",
|
||||
"tags": []
|
||||
"needs_background": "light"
|
||||
},
|
||||
"output_type": "display_data"
|
||||
}
|
||||
@@ -1033,19 +1019,19 @@
|
||||
"id": "6gCw2W4z6bYS"
|
||||
},
|
||||
"source": [
|
||||
"We can visualize several shapes from the test set together with the corresponding marginalized prediction and uncertainty."
|
||||
"We can also visualize several shapes from the test set together with the corresponding marginalized prediction and uncertainty."
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 52,
|
||||
"execution_count": 20,
|
||||
"metadata": {
|
||||
"colab": {
|
||||
"base_uri": "https://localhost:8080/",
|
||||
"height": 590
|
||||
"height": 570
|
||||
},
|
||||
"id": "iJ2y_HmDzkp6",
|
||||
"outputId": "a54e614a-a338-4075-be5e-d86bbf2f020e"
|
||||
"outputId": "6b22347d-9d76-4f6c-e06d-04703cbe1f8b"
|
||||
},
|
||||
"outputs": [
|
||||
{
|
||||
@@ -1054,22 +1040,19 @@
|
||||
"Text(0.5, 1.0, 'Std. Dev')"
|
||||
]
|
||||
},
|
||||
"execution_count": 52,
|
||||
"metadata": {
|
||||
"tags": []
|
||||
},
|
||||
"execution_count": 20,
|
||||
"metadata": {},
|
||||
"output_type": "execute_result"
|
||||
},
|
||||
{
|
||||
"data": {
|
||||
"image/png": "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 truncated
|
||||
"image/png": "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 truncated
|
||||
"text/plain": [
|
||||
"<Figure size 648x648 with 9 Axes>"
|
||||
]
|
||||
},
|
||||
"metadata": {
|
||||
"needs_background": "light",
|
||||
"tags": []
|
||||
"needs_background": "light"
|
||||
},
|
||||
"output_type": "display_data"
|
||||
}
|
||||
@@ -1093,7 +1076,7 @@
|
||||
"id": "3to3b5OS6p_9"
|
||||
},
|
||||
"source": [
|
||||
"As we can see, the shapes from the test set differ quite a bit from another. Nevertheless, the uncertainty estimate is reasonably distributed. This is, for every shape the bulk of uncertainty is located around the shape where one expects the most turbulent flow. "
|
||||
"As we can see, the shapes from the test set differ quite a bit from another. Nevertheless, the uncertainty estimate is reasonably distributed. It is especially high in the boundary layer around the airfoil, and in the regions of the low pressure pocket."
|
||||
]
|
||||
},
|
||||
{
|
||||
@@ -1102,6 +1085,8 @@
|
||||
"id": "trAG3xoQG_Q-"
|
||||
},
|
||||
"source": [
|
||||
"### Discussion\n",
|
||||
"\n",
|
||||
"Despite these promising results, there are still several issues with Bayesian Neural Nets, limiting their use in many practical applications. One serious drawback is the need for additional scaling of the KL-loss and the fact that there is no convincing argument on why it is necessary yet (read e.g. [here](http://proceedings.mlr.press/v119/wenzel20a/wenzel20a.pdf) or [here](https://arxiv.org/abs/2008.05912)).\n",
|
||||
"Furthermore, some people think that assuming independent normal distributions as variational approximations to the posterior is an oversimplification since in practice the weights are actually highly correlated ([paper](https://arxiv.org/abs/1909.00719)). Other people instead argue that this might not be an issue, as long as the networks in use are deep enough ([paper](https://arxiv.org/abs/2002.03704)). On top of that, there is research on different (e.g. heavy-tailed) priors other than normals and many other aspects of BNNs."
|
||||
]
|
||||
@@ -1148,5 +1133,5 @@
|
||||
}
|
||||
},
|
||||
"nbformat": 4,
|
||||
"nbformat_minor": 0
|
||||
"nbformat_minor": 1
|
||||
}
|
||||
@@ -14,11 +14,12 @@ if len(sys.argv)>1:
|
||||
|
||||
fileList = [
|
||||
"diffphys-code-burgers.ipynb", "diffphys-code-ns.ipynb", "diffphys-code-sol.ipynb", "physicalloss-code.ipynb", # TF
|
||||
"bayesian-code.ipynb", "supervised-airfoils.ipynb" # pytorch
|
||||
"bayesian-code.ipynb", "supervised-airfoils.ipynb", # pytorch
|
||||
"reinflearn-code.ipynb", # phiflow
|
||||
]
|
||||
|
||||
#fileList = [ "diffphys-code-burgers.ipynb"] # debug
|
||||
#fileList = [ "diffphys-code-ns.ipynb"] # debug
|
||||
#fileList = [ "diffphys-code-burgers.ipynb"] # debug, only 1 file
|
||||
#fileList = [ "diffphys-code-ns.ipynb"] # debug, only 1 file
|
||||
|
||||
|
||||
# main
|
||||
@@ -28,8 +29,7 @@ for fnOut in fileList:
|
||||
fn0 = fnOut[:-5] + "bak"
|
||||
fn = fn0 + "0"; cnt = 0
|
||||
while os.path.isfile(fn):
|
||||
#print("Error: "+fn+" already exists!")
|
||||
#exit(1)
|
||||
#print("Error: "+fn+" already exists!"); exit(1)
|
||||
print("Warning: "+fn+" already exists!")
|
||||
fn = fn0 + format(cnt); cnt=cnt+1
|
||||
|
||||
@@ -52,6 +52,7 @@ for fnOut in fileList:
|
||||
res.append( re.compile(r"DeprecationWarning:") )
|
||||
res.append( re.compile(r"InsecureRequestWarning") ) # for https download
|
||||
res.append( re.compile(r"Building wheel") ) # phiflow install, also gives weird unicode characters
|
||||
res.append( re.compile(r"warnings.warn") ) # phiflow warnings
|
||||
# remove all "warnings.warn" from phiflow?
|
||||
|
||||
# shorten data line: "0.008612174447657694, 0.02584669669548606, 0.043136357266407785"
|
||||
|
||||
+1
-1
@@ -5,7 +5,7 @@ echo WARNING - still requires one manual quit of first pdf/latex pass, use shift
|
||||
echo
|
||||
|
||||
# do clean git checkout for changes from json-cleanup-for-pdf.py?
|
||||
# git checkout diffphys-code-burgers.ipynb diffphys-code-ns.ipynb diffphys-code-sol.ipynb physicalloss-code.ipynb bayesian-code.ipynb supervised-airfoils.ipynb
|
||||
# git checkout diffphys-code-burgers.ipynb diffphys-code-ns.ipynb diffphys-code-sol.ipynb physicalloss-code.ipynb bayesian-code.ipynb supervised-airfoils.ipynb reinflearn-code.ipynb
|
||||
|
||||
# warning - modifies notebooks!
|
||||
python3.7 json-cleanup-for-pdf.py
|
||||
|
||||
+1289
-1311
File diff suppressed because it is too large.
Load diff
Reference in new issue
Block a user