"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"import matplotlib.pyplot as plt\n",
"\n",
"# Visualize a sample from the training data\n",
"sample_index = 0\n",
"plt.imshow(X_train[sample_index, 0, :, :, 2], cmap='viridis')\n",
"plt.title('Sample Input')\n",
"plt.show()\n",
"\n",
"plt.imshow(y_train[sample_index, 0, :, :, 0], cmap='viridis')\n",
"plt.title('Sample Target')\n",
"plt.show()\n"
]
},
{
"cell_type": "code",
"execution_count": 31,
"metadata": {},
"outputs": [],
"source": [
"np.save(\"X_val.npy\", X_val)\n",
"np.save(\"Y_val.npy\", y_val)"
]
},
{
"cell_type": "code",
"execution_count": 23,
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"Model: \"smogseer\"\n",
"\n"
],
"text/plain": [
"\u001b[1mModel: \"smogseer\"\u001b[0m\n"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [
"┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓\n",
"┃ Layer (type) ┃ Output Shape ┃ Param # ┃\n",
"┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩\n",
"│ input_layer_1 (InputLayer) │ (None, 1, 291, 512, 6) │ 0 │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ batch_normalization_3 │ (None, 1, 291, 512, 6) │ 24 │\n",
"│ (BatchNormalization) │ │ │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ conv_lstm2d_2 (ConvLSTM2D) │ (None, 1, 291, 512, │ 12,736 │\n",
"│ │ 16) │ │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ batch_normalization_4 │ (None, 1, 291, 512, │ 64 │\n",
"│ (BatchNormalization) │ 16) │ │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ conv_lstm2d_3 (ConvLSTM2D) │ (None, 1, 291, 512, │ 55,424 │\n",
"│ │ 32) │ │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ batch_normalization_5 │ (None, 1, 291, 512, │ 128 │\n",
"│ (BatchNormalization) │ 32) │ │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ conv3d_1 (Conv3D) │ (None, 1, 291, 512, 1) │ 865 │\n",
"└─────────────────────────────────┴────────────────────────┴───────────────┘\n",
"\n"
],
"text/plain": [
"┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓\n",
"┃\u001b[1m \u001b[0m\u001b[1mLayer (type) \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mOutput Shape \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m Param #\u001b[0m\u001b[1m \u001b[0m┃\n",
"┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩\n",
"│ input_layer_1 (\u001b[38;5;33mInputLayer\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m1\u001b[0m, \u001b[38;5;34m291\u001b[0m, \u001b[38;5;34m512\u001b[0m, \u001b[38;5;34m6\u001b[0m) │ \u001b[38;5;34m0\u001b[0m │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ batch_normalization_3 │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m1\u001b[0m, \u001b[38;5;34m291\u001b[0m, \u001b[38;5;34m512\u001b[0m, \u001b[38;5;34m6\u001b[0m) │ \u001b[38;5;34m24\u001b[0m │\n",
"│ (\u001b[38;5;33mBatchNormalization\u001b[0m) │ │ │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ conv_lstm2d_2 (\u001b[38;5;33mConvLSTM2D\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m1\u001b[0m, \u001b[38;5;34m291\u001b[0m, \u001b[38;5;34m512\u001b[0m, │ \u001b[38;5;34m12,736\u001b[0m │\n",
"│ │ \u001b[38;5;34m16\u001b[0m) │ │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ batch_normalization_4 │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m1\u001b[0m, \u001b[38;5;34m291\u001b[0m, \u001b[38;5;34m512\u001b[0m, │ \u001b[38;5;34m64\u001b[0m │\n",
"│ (\u001b[38;5;33mBatchNormalization\u001b[0m) │ \u001b[38;5;34m16\u001b[0m) │ │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ conv_lstm2d_3 (\u001b[38;5;33mConvLSTM2D\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m1\u001b[0m, \u001b[38;5;34m291\u001b[0m, \u001b[38;5;34m512\u001b[0m, │ \u001b[38;5;34m55,424\u001b[0m │\n",
"│ │ \u001b[38;5;34m32\u001b[0m) │ │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ batch_normalization_5 │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m1\u001b[0m, \u001b[38;5;34m291\u001b[0m, \u001b[38;5;34m512\u001b[0m, │ \u001b[38;5;34m128\u001b[0m │\n",
"│ (\u001b[38;5;33mBatchNormalization\u001b[0m) │ \u001b[38;5;34m32\u001b[0m) │ │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ conv3d_1 (\u001b[38;5;33mConv3D\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m1\u001b[0m, \u001b[38;5;34m291\u001b[0m, \u001b[38;5;34m512\u001b[0m, \u001b[38;5;34m1\u001b[0m) │ \u001b[38;5;34m865\u001b[0m │\n",
"└─────────────────────────────────┴────────────────────────┴───────────────┘\n"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [
" Total params: 69,241 (270.47 KB)\n",
"\n"
],
"text/plain": [
"\u001b[1m Total params: \u001b[0m\u001b[38;5;34m69,241\u001b[0m (270.47 KB)\n"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [
" Trainable params: 69,133 (270.05 KB)\n",
"\n"
],
"text/plain": [
"\u001b[1m Trainable params: \u001b[0m\u001b[38;5;34m69,133\u001b[0m (270.05 KB)\n"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [
" Non-trainable params: 108 (432.00 B)\n",
"\n"
],
"text/plain": [
"\u001b[1m Non-trainable params: \u001b[0m\u001b[38;5;34m108\u001b[0m (432.00 B)\n"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"Epoch 1/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m150s\u001b[0m 496ms/step - loss: 0.7337 - mean_squared_error: 0.1448 - val_loss: 0.6862 - val_mean_squared_error: 0.1305 - learning_rate: 1.0000e-05\n",
"Epoch 2/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m137s\u001b[0m 469ms/step - loss: 0.7032 - mean_squared_error: 0.1331 - val_loss: 0.6964 - val_mean_squared_error: 0.1351 - learning_rate: 1.0000e-05\n",
"Epoch 3/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m138s\u001b[0m 473ms/step - loss: 0.6931 - mean_squared_error: 0.1289 - val_loss: 0.7002 - val_mean_squared_error: 0.1363 - learning_rate: 1.0000e-05\n",
"Epoch 4/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m138s\u001b[0m 472ms/step - loss: 0.6858 - mean_squared_error: 0.1247 - val_loss: 0.6941 - val_mean_squared_error: 0.1334 - learning_rate: 1.0000e-05\n",
"Epoch 5/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m138s\u001b[0m 473ms/step - loss: 0.6783 - mean_squared_error: 0.1212 - val_loss: 0.6801 - val_mean_squared_error: 0.1267 - learning_rate: 1.0000e-05\n",
"Epoch 6/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m139s\u001b[0m 478ms/step - loss: 0.6691 - mean_squared_error: 0.1169 - val_loss: 0.6731 - val_mean_squared_error: 0.1232 - learning_rate: 1.0000e-05\n",
"Epoch 7/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m141s\u001b[0m 483ms/step - loss: 0.6576 - mean_squared_error: 0.1122 - val_loss: 0.6551 - val_mean_squared_error: 0.1144 - learning_rate: 1.0000e-05\n",
"Epoch 8/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m147s\u001b[0m 503ms/step - loss: 0.6441 - mean_squared_error: 0.1054 - val_loss: 0.6403 - val_mean_squared_error: 0.1071 - learning_rate: 1.0000e-05\n",
"Epoch 9/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m149s\u001b[0m 510ms/step - loss: 0.6289 - mean_squared_error: 0.0970 - val_loss: 0.6240 - val_mean_squared_error: 0.0991 - learning_rate: 1.0000e-05\n",
"Epoch 10/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m144s\u001b[0m 492ms/step - loss: 0.6117 - mean_squared_error: 0.0892 - val_loss: 0.6035 - val_mean_squared_error: 0.0891 - learning_rate: 1.0000e-05\n",
"Epoch 11/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m142s\u001b[0m 485ms/step - loss: 0.5937 - mean_squared_error: 0.0801 - val_loss: 0.5833 - val_mean_squared_error: 0.0794 - learning_rate: 1.0000e-05\n",
"Epoch 12/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m137s\u001b[0m 469ms/step - loss: 0.5754 - mean_squared_error: 0.0714 - val_loss: 0.5642 - val_mean_squared_error: 0.0702 - learning_rate: 1.0000e-05\n",
"Epoch 13/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m138s\u001b[0m 472ms/step - loss: 0.5567 - mean_squared_error: 0.0626 - val_loss: 0.5496 - val_mean_squared_error: 0.0634 - learning_rate: 1.0000e-05\n",
"Epoch 14/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m138s\u001b[0m 473ms/step - loss: 0.5387 - mean_squared_error: 0.0540 - val_loss: 0.5301 - val_mean_squared_error: 0.0543 - learning_rate: 1.0000e-05\n",
"Epoch 15/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m137s\u001b[0m 469ms/step - loss: 0.5220 - mean_squared_error: 0.0461 - val_loss: 0.5103 - val_mean_squared_error: 0.0453 - learning_rate: 1.0000e-05\n",
"Epoch 16/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m142s\u001b[0m 486ms/step - loss: 0.5051 - mean_squared_error: 0.0391 - val_loss: 0.4963 - val_mean_squared_error: 0.0390 - learning_rate: 1.0000e-05\n",
"Epoch 17/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m147s\u001b[0m 503ms/step - loss: 0.4907 - mean_squared_error: 0.0324 - val_loss: 0.4815 - val_mean_squared_error: 0.0326 - learning_rate: 1.0000e-05\n",
"Epoch 18/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m147s\u001b[0m 502ms/step - loss: 0.4774 - mean_squared_error: 0.0263 - val_loss: 0.4664 - val_mean_squared_error: 0.0263 - learning_rate: 1.0000e-05\n",
"Epoch 19/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m146s\u001b[0m 501ms/step - loss: 0.4648 - mean_squared_error: 0.0214 - val_loss: 0.4584 - val_mean_squared_error: 0.0230 - learning_rate: 1.0000e-05\n",
"Epoch 20/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m142s\u001b[0m 485ms/step - loss: 0.4539 - mean_squared_error: 0.0169 - val_loss: 0.4452 - val_mean_squared_error: 0.0177 - learning_rate: 1.0000e-05\n",
"Epoch 21/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m139s\u001b[0m 475ms/step - loss: 0.4444 - mean_squared_error: 0.0133 - val_loss: 0.4366 - val_mean_squared_error: 0.0144 - learning_rate: 1.0000e-05\n",
"Epoch 22/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m138s\u001b[0m 472ms/step - loss: 0.4362 - mean_squared_error: 0.0103 - val_loss: 0.4267 - val_mean_squared_error: 0.0108 - learning_rate: 1.0000e-05\n",
"Epoch 23/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m138s\u001b[0m 473ms/step - loss: 0.4303 - mean_squared_error: 0.0077 - val_loss: 0.4232 - val_mean_squared_error: 0.0096 - learning_rate: 1.0000e-05\n",
"Epoch 24/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m139s\u001b[0m 475ms/step - loss: 0.4241 - mean_squared_error: 0.0058 - val_loss: 0.4167 - val_mean_squared_error: 0.0073 - learning_rate: 1.0000e-05\n",
"Epoch 25/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m145s\u001b[0m 497ms/step - loss: 0.4207 - mean_squared_error: 0.0044 - val_loss: 0.4112 - val_mean_squared_error: 0.0054 - learning_rate: 1.0000e-05\n",
"Epoch 26/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m147s\u001b[0m 502ms/step - loss: 0.4154 - mean_squared_error: 0.0034 - val_loss: 0.4105 - val_mean_squared_error: 0.0052 - learning_rate: 1.0000e-05\n",
"Epoch 27/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m146s\u001b[0m 499ms/step - loss: 0.4136 - mean_squared_error: 0.0026 - val_loss: 0.4066 - val_mean_squared_error: 0.0040 - learning_rate: 1.0000e-05\n",
"Epoch 28/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m146s\u001b[0m 500ms/step - loss: 0.4119 - mean_squared_error: 0.0020 - val_loss: 0.4046 - val_mean_squared_error: 0.0033 - learning_rate: 1.0000e-05\n",
"Epoch 29/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m139s\u001b[0m 476ms/step - loss: 0.4105 - mean_squared_error: 0.0016 - val_loss: 0.4034 - val_mean_squared_error: 0.0030 - learning_rate: 1.0000e-05\n",
"Epoch 30/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m138s\u001b[0m 474ms/step - loss: 0.4106 - mean_squared_error: 0.0013 - val_loss: 0.4030 - val_mean_squared_error: 0.0029 - learning_rate: 1.0000e-05\n",
"Epoch 31/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m138s\u001b[0m 472ms/step - loss: 0.4078 - mean_squared_error: 0.0012 - val_loss: 0.4021 - val_mean_squared_error: 0.0026 - learning_rate: 1.0000e-05\n",
"Epoch 32/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m138s\u001b[0m 471ms/step - loss: 0.4087 - mean_squared_error: 0.0011 - val_loss: 0.4015 - val_mean_squared_error: 0.0024 - learning_rate: 1.0000e-05\n",
"Epoch 33/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m139s\u001b[0m 475ms/step - loss: 0.4083 - mean_squared_error: 9.7096e-04 - val_loss: 0.4016 - val_mean_squared_error: 0.0024 - learning_rate: 1.0000e-05\n",
"Epoch 34/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m141s\u001b[0m 483ms/step - loss: 0.4075 - mean_squared_error: 9.5572e-04 - val_loss: 0.4013 - val_mean_squared_error: 0.0023 - learning_rate: 1.0000e-05\n",
"Epoch 35/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m139s\u001b[0m 475ms/step - loss: 0.4084 - mean_squared_error: 9.2558e-04 - val_loss: 0.4006 - val_mean_squared_error: 0.0021 - learning_rate: 1.0000e-05\n",
"Epoch 36/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m137s\u001b[0m 469ms/step - loss: 0.4088 - mean_squared_error: 8.9947e-04 - val_loss: 0.4008 - val_mean_squared_error: 0.0022 - learning_rate: 1.0000e-05\n",
"Epoch 37/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m139s\u001b[0m 477ms/step - loss: 0.4081 - mean_squared_error: 8.2714e-04 - val_loss: 0.4005 - val_mean_squared_error: 0.0021 - learning_rate: 1.0000e-05\n",
"Epoch 38/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m147s\u001b[0m 504ms/step - loss: 0.4093 - mean_squared_error: 8.5030e-04 - val_loss: 0.4005 - val_mean_squared_error: 0.0021 - learning_rate: 1.0000e-05\n",
"Epoch 39/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m146s\u001b[0m 500ms/step - loss: 0.4090 - mean_squared_error: 8.4955e-04 - val_loss: 0.4002 - val_mean_squared_error: 0.0020 - learning_rate: 1.0000e-05\n",
"Epoch 40/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m148s\u001b[0m 507ms/step - loss: 0.4085 - mean_squared_error: 8.1096e-04 - val_loss: 0.4000 - val_mean_squared_error: 0.0019 - learning_rate: 1.0000e-05\n",
"Epoch 41/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m140s\u001b[0m 481ms/step - loss: 0.4082 - mean_squared_error: 7.9257e-04 - val_loss: 0.4003 - val_mean_squared_error: 0.0020 - learning_rate: 1.0000e-05\n",
"Epoch 42/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m138s\u001b[0m 473ms/step - loss: 0.4087 - mean_squared_error: 7.7913e-04 - val_loss: 0.4006 - val_mean_squared_error: 0.0021 - learning_rate: 1.0000e-05\n",
"Epoch 43/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m139s\u001b[0m 474ms/step - loss: 0.4088 - mean_squared_error: 7.6459e-04 - val_loss: 0.3999 - val_mean_squared_error: 0.0019 - learning_rate: 1.0000e-05\n",
"Epoch 44/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m138s\u001b[0m 473ms/step - loss: 0.4083 - mean_squared_error: 7.5364e-04 - val_loss: 0.4001 - val_mean_squared_error: 0.0020 - learning_rate: 1.0000e-05\n",
"Epoch 45/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m138s\u001b[0m 473ms/step - loss: 0.4066 - mean_squared_error: 7.8711e-04 - val_loss: 0.3999 - val_mean_squared_error: 0.0019 - learning_rate: 1.0000e-05\n",
"Epoch 46/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m145s\u001b[0m 495ms/step - loss: 0.4093 - mean_squared_error: 7.6262e-04 - val_loss: 0.3998 - val_mean_squared_error: 0.0019 - learning_rate: 1.0000e-05\n",
"Epoch 47/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m147s\u001b[0m 504ms/step - loss: 0.4076 - mean_squared_error: 7.3529e-04 - val_loss: 0.3997 - val_mean_squared_error: 0.0019 - learning_rate: 1.0000e-05\n",
"Epoch 48/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m145s\u001b[0m 496ms/step - loss: 0.4094 - mean_squared_error: 6.9850e-04 - val_loss: 0.3998 - val_mean_squared_error: 0.0019 - learning_rate: 1.0000e-05\n",
"Epoch 49/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m147s\u001b[0m 503ms/step - loss: 0.4087 - mean_squared_error: 7.1750e-04 - val_loss: 0.3997 - val_mean_squared_error: 0.0019 - learning_rate: 1.0000e-05\n",
"Epoch 50/50\n",
"\u001b[1m292/292\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m138s\u001b[0m 472ms/step - loss: 0.4070 - mean_squared_error: 6.8971e-04 - val_loss: 0.3996 - val_mean_squared_error: 0.0018 - learning_rate: 1.0000e-05\n",
"Restoring model weights from the end of the best epoch: 50.\n"
]
}
],
"source": [
"# Define the model with correct input shape\n",
"inp = layers.Input(shape=(1, X_data_clean.shape[2], X_data_clean.shape[3], X_data_clean.shape[4]))\n",
"\n",
"x = layers.BatchNormalization()(inp)\n",
"x = layers.ConvLSTM2D(\n",
" filters=16,\n",
" kernel_size=(3, 3),\n",
" padding=\"same\",\n",
" return_sequences=True,\n",
" activation=\"tanh\",\n",
" recurrent_activation=\"sigmoid\",\n",
" kernel_initializer=\"glorot_uniform\"\n",
")(x)\n",
"x = layers.BatchNormalization()(x)\n",
"x = layers.ConvLSTM2D(\n",
" filters=32,\n",
" kernel_size=(3, 3),\n",
" padding=\"same\",\n",
" return_sequences=True,\n",
" activation=\"tanh\",\n",
" recurrent_activation=\"sigmoid\",\n",
" kernel_initializer=\"glorot_uniform\"\n",
")(x)\n",
"x = layers.BatchNormalization()(x)\n",
"x = layers.Conv3D(\n",
" filters=1, kernel_size=(3, 3, 3), activation=\"sigmoid\", padding=\"same\"\n",
")(x)\n",
"\n",
"model = keras.models.Model(inp, x, name=\"smogseer\")\n",
"\n",
"# Use a reduced learning rate and gradient clipping\n",
"optimizer = keras.optimizers.Adam(learning_rate=1e-5, clipnorm=1.0)\n",
"model.compile(\n",
" loss=keras.losses.binary_crossentropy,\n",
" optimizer=optimizer,\n",
" metrics=['mean_squared_error']\n",
")\n",
"\n",
"# Print the model summary\n",
"model.summary()\n",
"\n",
"# Data Generator Class\n",
"\n",
"class DataGenerator(Sequence):\n",
" def __init__(self, X_data, y_data, batch_size):\n",
" self.X_data = X_data\n",
" self.y_data = y_data\n",
" self.batch_size = batch_size\n",
" self.indices = np.arange(X_data.shape[0])\n",
" \n",
" def __len__(self):\n",
" return int(np.ceil(len(self.indices) / self.batch_size))\n",
" \n",
" def __getitem__(self, index):\n",
" batch_indices = self.indices[index * self.batch_size:(index + 1) * self.batch_size]\n",
" batch_X = self.X_data[batch_indices]\n",
" batch_y = self.y_data[batch_indices]\n",
" return batch_X, batch_y\n",
"\n",
" def on_epoch_end(self):\n",
" np.random.shuffle(self.indices)\n",
"\n",
"batch_size = 1\n",
"train_generator = DataGenerator(X_train, y_train, batch_size)\n",
"val_generator = DataGenerator(X_val, y_val, batch_size)\n",
"\n",
"# Define callbacks for monitoring and adjusting learning rate\n",
"callbacks = [\n",
" keras.callbacks.ReduceLROnPlateau(\n",
" monitor='val_loss', factor=0.1, patience=10, verbose=1, min_lr=1e-7\n",
" ),\n",
" keras.callbacks.EarlyStopping(\n",
" monitor='val_loss', patience=15, verbose=1, restore_best_weights=True\n",
" ),\n",
" keras.callbacks.TensorBoard(log_dir='./logs')\n",
"]\n",
"\n",
"# Train the model using data generators\n",
"history = model.fit(train_generator, epochs=50, validation_data=val_generator, callbacks=callbacks)\n",
"# Save the model\n",
"model.save('smogseer.keras')\n"
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"\u001b[1m3/3\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m7s\u001b[0m 747ms/step\n",
"\u001b[1m3/3\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 460ms/step - loss: 0.3992 - mean_squared_error: 0.0016\n",
"Validation Loss: 0.39958614110946655\n",
"Validation Accuracy: 0.0018263210076838732\n"
]
}
],
"source": [
"## LOAD CHECKPOINTS IF NEEDED\n",
"X_val = np.load('X_val.npy')\n",
"y_val = np.load('Y_val.npy')\n",
"\n",
"# Load the model\n",
"model = load_model('smogseer50.keras')\n",
"\n",
"# Run predictions on validation data\n",
"predictions = model.predict(X_val)\n",
"\n",
"# Evaluate the model on validation data\n",
"val_loss, val_accuracy = model.evaluate(X_val, y_val)\n",
"print(f\"Validation Loss: {val_loss}\")\n",
"print(f\"Validation Accuracy: {val_accuracy}\")"
]
},
{
"cell_type": "code",
"execution_count": 28,
"metadata": {},
"outputs": [],
"source": [
"\n",
"# Plot comparisons and training history\n",
"\n",
"def plot_comparison(y_true, y_pred, index, save_path):\n",
" \"\"\"\n",
" Plots the ground truth and the predicted output for a given index.\n",
" \n",
" Parameters:\n",
" - y_true: Ground truth data\n",
" - y_pred: Predicted data\n",
" - index: Index of the sample to plot\n",
" - save_path: Path to save the plot\n",
" \"\"\"\n",
" fig, axes = plt.subplots(1, 2, figsize=(12, 6))\n",
"\n",
" # Plot ground truth\n",
" ax = axes[0]\n",
" ax.imshow(y_true[index, 0, :, :, 0], cmap='viridis')\n",
" ax.set_title('Ground Truth')\n",
" ax.axis('off')\n",
"\n",
" # Plot prediction\n",
" ax = axes[1]\n",
" ax.imshow(y_pred[index, 0, :, :, 0], cmap='viridis')\n",
" ax.set_title('Prediction')\n",
" ax.axis('off')\n",
"\n",
" plt.tight_layout()\n",
" plt.savefig(save_path)\n",
" plt.close()\n",
"\n",
"# Visualize a few samples\n",
"num_samples_to_plot = 5\n",
"for i in range(num_samples_to_plot):\n",
" plot_comparison(y_val, predictions, i, f'comparison_plot_{i}.png')\n",
"\n",
"# Plot training history\n",
"def plot_training_history(history, save_path):\n",
" \"\"\"\n",
" Plots the training and validation loss and accuracy over epochs.\n",
"\n",
" Parameters:\n",
" - history: Keras History object\n",
" - save_path: Path to save the plot\n",
" \"\"\"\n",
" fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(14, 6))\n",
"\n",
" # Plot loss\n",
" ax1.plot(history.history['loss'], label='Training Loss')\n",
" ax1.plot(history.history['val_loss'], label='Validation Loss')\n",
" ax1.set_title('Loss over epochs')\n",
" ax1.set_xlabel('Epoch')\n",
" ax1.set_ylabel('Loss')\n",
" ax1.legend()\n",
"\n",
" # Plot accuracy\n",
" ax2.plot(history.history['mean_squared_error'], label='Training MSE')\n",
" ax2.plot(history.history['val_mean_squared_error'], label='Validation MSE')\n",
" ax2.set_title('MSE over epochs')\n",
" ax2.set_xlabel('Epoch')\n",
" ax2.set_ylabel('MSE')\n",
" ax2.legend()\n",
"\n",
" plt.tight_layout()\n",
" plt.savefig(save_path)\n",
" plt.close()\n",
"\n",
"# Plot training history\n",
"plot_training_history(history, 'training_history_epoch50.png')"
]
},
{
"cell_type": "code",
"execution_count": 24,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"(291, 512)\n",
"(291, 512)\n",
"(291, 512)\n",
"(291, 512)\n",
"(291, 512)\n"
]
}
],
"source": [
"# Visualize a few samples\n",
"num_timesteps_to_plot = 5\n",
"for i in range(num_timesteps_to_plot):\n",
" print(predictions[i, 0, :, :, 0].shape)"
]
},
{
"cell_type": "code",
"execution_count": 25,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Target data range: 0.99998665 1.6277676e-06\n"
]
}
],
"source": [
"print(\"Target data range: \", predictions.max(), predictions.min())"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "venv",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.9.12"
}
},
"nbformat": 4,
"nbformat_minor": 2
}