\n
๐ก Model Architecture:
\n \n1. **Input Layer**: Accepts input images of shape (224, 224, 3), which corresponds to images with a height and width of 224 pixels and three color channels (RGB).\n2. **Resizing Layer: Resizes the input images to a smaller size of (64, 64). This reduction in image dimensions may help speed up training.\n3. **Normalization Layer**: Normalizes pixel values to have zero mean and unit variance, which aids in stabilizing and speeding up the training process.\n4. **Convolutional Layers**: Utilizes two convolutional layers:\n - The first convolutional layer has 64 filters, a kernel size of 3x3, and ReLU activation.\n - The second convolutional layer has 128 filters, a kernel size of 3x3, and ReLU activation.\n7. **MaxPooling Layer**: Performs max pooling with a pool size of 2x2, reducing the spatial dimensions of the feature maps.\n8. **Dropout Layer**: Introduces a dropout rate of 20% to prevent overfitting by randomly deactivating a fraction of neurons during training.\n9. **Flatten Layer**: Flattens the 2D feature maps into a 1D vector to prepare for the fully connected layers.\n10. **RandomFourierFeatures Layer**: Incorporates random Fourier features with 5 components, which can approximate non-linear mappings efficiently for the data.\n11. **Compilation**: Compiles the model using the AdamW optimizer with a learning rate of 0.01, categorical cross-entropy loss function (suitable for multi-class classification), and accuracy as the evaluation metric.","metadata":{}},{"cell_type":"code","source":"from tensorflow.keras.models import Sequential\nfrom tensorflow.keras.layers import Conv2D, MaxPooling2D,Flatten, Dense, Dropout,Normalization,Resizing,InputLayer\nfrom tensorflow.keras.layers.experimental import RandomFourierFeatures\nfrom tensorflow.keras.optimizers import Adam,Adafactor,AdamW,Lion\nfrom tensorflow.keras.optimizers.experimental import Adadelta,Adagrad,Adamax,RMSprop,SGD,Nadam,Ftrl\n\n\n# Load pre-trained VGG16 model without top layers\nbase_model = VGG16(weights='imagenet', include_top=False, input_shape=(224, 224, 3))\n\n# Freeze the convolutional layers of VGG16\nfor layer in base_model.layers:\n layer.trainable = False\n\n# Create a new Sequential model\nmodel = Sequential()\nmodel.add(Input(shape=(224, 224, 3)))\nmodel.add(Resizing(224, 224))\nmodel.add(Normalization())\nmodel.add(base_model)\nmodel.add(Flatten())\nmodel.add(Dense(256, activation='relu'))\nmodel.add(Dropout(0.3))\nmodel.add(Dense(5, activation='softmax'))\n# Learning rate schedule for AdamW optimizer\ninitial_learning_rate = 0.001\nlr_schedule = ExponentialDecay(\n initial_learning_rate, decay_steps=10000, decay_rate=0.9, staircase=True\n)\n\n# Compile the model\nmodel.compile(optimizer=AdamW(learning_rate=lr_schedule), loss='categorical_crossentropy', metrics=['accuracy'])\nmodel.summary()\n","metadata":{"id":"oa1b9vgEn6kG","outputId":"9e90d311-00b1-420e-ca68-5998e7dea087","execution":{"iopub.status.busy":"2023-12-27T09:26:37.012656Z","iopub.execute_input":"2023-12-27T09:26:37.013376Z","iopub.status.idle":"2023-12-27T09:26:37.503210Z","shell.execute_reply.started":"2023-12-27T09:26:37.013345Z","shell.execute_reply":"2023-12-27T09:26:37.502349Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"\n\nhist = model.fit(x_train_norm, y_train_encoded, batch_size=32,\n steps_per_epoch=len(x_train_norm) / 32,\n epochs=20,\n validation_data=(x_val_norm, y_val_encoded),\n class_weight=class_weight_dict\n )\n# Evaluate on test set\ntest_loss, test_acc = model.evaluate(x_test_norm, y_test_encoded)\nprint(f'Test accuracy: {test_acc}')","metadata":{"id":"AB8UY1DgseoL","outputId":"c1694b58-9705-4b2e-9618-745560f9c867","execution":{"iopub.status.busy":"2023-12-27T09:28:22.802384Z","iopub.execute_input":"2023-12-27T09:28:22.802749Z","iopub.status.idle":"2023-12-27T09:28:39.423787Z","shell.execute_reply.started":"2023-12-27T09:28:22.802718Z","shell.execute_reply":"2023-12-27T09:28:39.422873Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"acc = hist.history['accuracy']\nval_acc = hist.history['val_accuracy']\nepochs = range(1, len(acc) + 1)\n\nplt.plot(epochs, acc, '-', label='Training Accuracy')\nplt.plot(epochs, val_acc, ':', label='Validation Accuracy')\nplt.title('Training and Validation Accuracy')\nplt.xlabel('Epoch')\nplt.ylabel('Accuracy')\nplt.legend(loc='lower right')\nplt.plot()\nplt.show()","metadata":{"id":"6lgzf5RIn-m-","outputId":"6ebd621c-9c91-4a5d-ec3b-3989e7badaf9","execution":{"iopub.status.busy":"2023-12-27T09:28:44.840458Z","iopub.execute_input":"2023-12-27T09:28:44.840797Z","iopub.status.idle":"2023-12-27T09:28:45.180066Z","shell.execute_reply.started":"2023-12-27T09:28:44.840771Z","shell.execute_reply":"2023-12-27T09:28:45.179143Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"#
๐งช Testing","metadata":{}},{"cell_type":"markdown","source":"
\n For predicting on single tets files this code will be used. For each image file path obtained, the code performs the following operations:\n\n- Loads the image using image.load_img() from the Keras preprocessing module, resizing it to the required dimensions of 224x224 pixels.\n- Converts the image to a format compatible with the model for prediction (image.img_to_array() and np.expand_dims()).\n- Utilizes the model for prediction on the image\n- Retrieves class labels corresponding to the categories ('belly pain', 'burping', 'discomfort', 'hungry', 'tired').\n- Prints the original file path and its predicted category, appending this information to the results list.\n
","metadata":{}},{"cell_type":"code","source":"import glob\nimport os\n\ndef get_png_files(directory):\n folders = ['belly_pain', 'burping', 'discomfort', 'hungry', 'tired'] \n png_files = []\n\n for folder_name in folders:\n folder_path = os.path.join(directory, folder_name)\n if os.path.exists(folder_path):\n png_files.extend(glob.glob(os.path.join(folder_path, '*.png')))\n else:\n print(f\"Folder '{folder_name}' does not exist in '{directory}'.\")\n\n return png_files\n\ndirectory_path = '/kaggle/working/' \npng_files_list = get_png_files(directory_path)\n\nresults = []\nfor file_path in png_files_list:\n x = image.load_img(file_path, target_size=(224, 224))\n\n x = image.img_to_array(x)\n x = np.expand_dims(x, axis=0)\n\n y = model.predict(x)\n\n class_labels = ['belly pain','burping','discomfort','hungry','tired']\n\n # for i, label in enumerate(class_labels):\n # print(f'{label}: {y[0][i]}')\n\n results.append(f\"Original:{file_path.split('/')[3]} Predicted: {class_labels[np.argmax(y)]}\")\n \nprint('\\n')\n\nfor i in results:\n print(i)\n print('\\n')","metadata":{"execution":{"iopub.status.busy":"2023-12-27T09:29:04.059133Z","iopub.execute_input":"2023-12-27T09:29:04.059847Z","iopub.status.idle":"2023-12-27T09:29:04.801211Z","shell.execute_reply.started":"2023-12-27T09:29:04.059814Z","shell.execute_reply":"2023-12-27T09:29:04.800316Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]}]}
\ No newline at end of file