{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"name":"python","version":"3.10.12","mimetype":"text/x-python","codemirror_mode":{"name":"ipython","version":3},"pygments_lexer":"ipython3","nbconvert_exporter":"python","file_extension":".py"},"kaggle":{"accelerator":"none","dataSources":[{"sourceId":60180,"databundleVersionId":6496577,"sourceType":"competition"}],"dockerImageVersionId":30627,"isInternetEnabled":true,"language":"python","sourceType":"notebook","isGpuEnabled":false}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"code","source":"# This Python 3 environment comes with many helpful analytics libraries installed\n# It is defined by the kaggle/python Docker image: https://github.com/kaggle/docker-python\n# For example, here's several helpful packages to load\n\nimport numpy as np # linear algebra\nimport pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)\n\n# Input data files are available in the read-only \"../input/\" directory\n# For example, running this (by clicking run or pressing Shift+Enter) will list all files under the input directory\n\nimport os\nfor dirname, _, filenames in os.walk('/kaggle/input'):\n    for filename in filenames:\n        print(os.path.join(dirname, filename))\n\n# You can write up to 20GB to the current directory (/kaggle/working/) that gets preserved as output when you create a version using \"Save & Run All\" \n# You can also write temporary files to /kaggle/temp/, but they won't be saved outside of the current session","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import tensorflow as tf\nfrom tensorflow.keras.layers import Input, Conv2D, MaxPooling2D, UpSampling2D\nfrom tensorflow.keras.models import Model\nfrom tensorflow.keras.optimizers import Adam\nfrom tensorflow.keras.callbacks import EarlyStopping\nfrom tensorflow.keras.layers import Dropout, BatchNormalization\nimport numpy as np\nimport zipfile\nimport os\nfrom tensorflow.keras.preprocessing.image import ImageDataGenerator\n\ndef preprocess_images(image_path, label_path):\n    image = tf.io.read_file(image_path)\n    image = tf.image.decode_png(image, channels=3)\n    image = tf.image.resize(image, [128, 128])\n    image = image / 255.0\n\n    label = tf.io.read_file(label_path)\n    label = tf.image.decode_png(label, channels=3)\n    label = tf.image.resize(label, [128, 128])\n    label = label / 255.0\n\n    return image, label\n\ndef paired_dataset_loader(noisy_dir, clean_dir, validation_split=0.2):\n    noisy_files = tf.data.Dataset.list_files(os.path.join(noisy_dir, '*.png'))\n    clean_files = noisy_files.map(lambda x: tf.strings.regex_replace(x, 'noisy', 'clean'))\n\n    dataset = tf.data.Dataset.zip((noisy_files, clean_files))\n    dataset = dataset.map(preprocess_images)\n    dataset = dataset.shuffle(buffer_size=1000)\n\n    total_size = len(os.listdir(noisy_dir))\n    val_size = int(total_size * validation_split)\n\n    val_dataset = dataset.take(val_size).batch(256)\n    train_dataset = dataset.skip(val_size).batch(256)\n\n    return train_dataset, val_dataset\n\ntrain_clean_dir = '/kaggle/input/fall-2023-cisc6000-deep-learning-hw2/train_clean_data/train_clean_data'\ntrain_noisy_dir = '/kaggle/input/fall-2023-cisc6000-deep-learning-hw2/train_noisy_data/train_noisy_data'\n\ntest_clean_dir = '/kaggle/input/fall-2023-cisc6000-deep-learning-hw2/test_noisy_data/test_noisy_data'\ntest_noisy_dir = '/kaggle/input/fall-2023-cisc6000-deep-learning-hw2/test_clean_data/test_clean_data'","metadata":{"execution":{"iopub.status.busy":"2023-12-21T12:32:38.834426Z","iopub.execute_input":"2023-12-21T12:32:38.834928Z","iopub.status.idle":"2023-12-21T12:32:38.851786Z","shell.execute_reply.started":"2023-12-21T12:32:38.834889Z","shell.execute_reply":"2023-12-21T12:32:38.850143Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_dataset, val_dataset = paired_dataset_loader(train_noisy_dir, train_clean_dir, validation_split=0.2)","metadata":{"execution":{"iopub.status.busy":"2023-12-21T12:32:41.440698Z","iopub.execute_input":"2023-12-21T12:32:41.441245Z","iopub.status.idle":"2023-12-21T12:32:49.619812Z","shell.execute_reply.started":"2023-12-21T12:32:41.441203Z","shell.execute_reply":"2023-12-21T12:32:49.618661Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"test_dataset, val_dataset = paired_dataset_loader(test_noisy_dir, test_clean_dir, validation_split = 0.2)","metadata":{"execution":{"iopub.status.busy":"2023-12-21T12:32:49.622322Z","iopub.execute_input":"2023-12-21T12:32:49.623206Z","iopub.status.idle":"2023-12-21T12:32:59.761864Z","shell.execute_reply.started":"2023-12-21T12:32:49.623132Z","shell.execute_reply":"2023-12-21T12:32:59.760328Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"input_img = Input(shape=(128, 128, 3))\n","metadata":{"execution":{"iopub.status.busy":"2023-12-20T00:16:21.350915Z","iopub.execute_input":"2023-12-20T00:16:21.351294Z","iopub.status.idle":"2023-12-20T00:16:21.374953Z","shell.execute_reply.started":"2023-12-20T00:16:21.351259Z","shell.execute_reply":"2023-12-20T00:16:21.374241Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"##Encoder\n##Original code had 0.57~ val, adding batchnorm/dropout to improve\nx = Conv2D(32, (3, 3), activation='relu', padding='same')(input_img)\nx = MaxPooling2D((2, 2), padding='same')(x)\nx = BatchNormalization()(x)  #Batch Normalization\nx = Conv2D(48, (3, 3), activation='relu', padding='same')(x)\nx = MaxPooling2D((2, 2), padding='same')(x)\nx = Dropout(0.2)(x)  #Add Dropout \nx = Conv2D(64, (3, 3), activation='relu', padding='same')(x)\nencoded = MaxPooling2D((2, 2), padding='same')(x)\n","metadata":{"execution":{"iopub.status.busy":"2023-12-20T00:16:21.375987Z","iopub.execute_input":"2023-12-20T00:16:21.376267Z","iopub.status.idle":"2023-12-20T00:16:21.47025Z","shell.execute_reply.started":"2023-12-20T00:16:21.376242Z","shell.execute_reply":"2023-12-20T00:16:21.469231Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"## Decoder\nx = Conv2D(64, (3, 3), activation='relu', padding='same')(encoded)\nx = UpSampling2D((2, 2))(x)\nx = Conv2D(48, (3, 3), activation='relu', padding='same')(x)\nx = UpSampling2D((2, 2))(x)\nx = Dropout(0.2)(x)  # Added Dropout\nx = Conv2D(32, (3, 3), activation='relu', padding='same')(x)\nx = UpSampling2D((2, 2))(x)\ndecoded = Conv2D(3, (3, 3), activation='sigmoid', padding='same')(x)\n","metadata":{"execution":{"iopub.status.busy":"2023-12-20T00:16:21.471378Z","iopub.execute_input":"2023-12-20T00:16:21.471674Z","iopub.status.idle":"2023-12-20T00:16:21.546342Z","shell.execute_reply.started":"2023-12-20T00:16:21.471647Z","shell.execute_reply":"2023-12-20T00:16:21.545555Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"autoencoder = Model(input_img, decoded)\nautoencoder.compile(optimizer='adam', loss='mean_squared_error')","metadata":{"execution":{"iopub.status.busy":"2023-12-20T00:16:21.547424Z","iopub.execute_input":"2023-12-20T00:16:21.547701Z","iopub.status.idle":"2023-12-20T00:16:21.571581Z","shell.execute_reply.started":"2023-12-20T00:16:21.547676Z","shell.execute_reply":"2023-12-20T00:16:21.570758Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"autoencoder.summary()","metadata":{"execution":{"iopub.status.busy":"2023-12-20T00:16:21.572786Z","iopub.execute_input":"2023-12-20T00:16:21.573209Z","iopub.status.idle":"2023-12-20T00:16:21.619118Z","shell.execute_reply.started":"2023-12-20T00:16:21.573173Z","shell.execute_reply":"2023-12-20T00:16:21.618154Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"encoder_input = autoencoder.input\nencoder_output = autoencoder.get_layer('max_pooling2d_2').output\nencoder_model = Model(inputs=encoder_input, outputs=encoder_output)","metadata":{"execution":{"iopub.status.busy":"2023-12-20T00:26:50.958715Z","iopub.execute_input":"2023-12-20T00:26:50.959471Z","iopub.status.idle":"2023-12-20T00:26:50.968962Z","shell.execute_reply.started":"2023-12-20T00:26:50.959433Z","shell.execute_reply":"2023-12-20T00:26:50.968047Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import time\n\nstart_time = time.time()","metadata":{"execution":{"iopub.status.busy":"2023-12-20T00:26:51.701451Z","iopub.execute_input":"2023-12-20T00:26:51.702341Z","iopub.status.idle":"2023-12-20T00:26:51.706766Z","shell.execute_reply.started":"2023-12-20T00:26:51.702303Z","shell.execute_reply":"2023-12-20T00:26:51.70547Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print(type(train_dataset))\nprint(type(val_dataset))\n","metadata":{"execution":{"iopub.status.busy":"2023-12-20T00:26:52.146444Z","iopub.execute_input":"2023-12-20T00:26:52.147163Z","iopub.status.idle":"2023-12-20T00:26:52.152153Z","shell.execute_reply.started":"2023-12-20T00:26:52.147125Z","shell.execute_reply":"2023-12-20T00:26:52.151138Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"start_time = time.time()\n\nhistory = autoencoder.fit(\n    train_dataset,\n    epochs=100,\n    validation_data=val_dataset  \n)\n\nend_time = time.time()","metadata":{"execution":{"iopub.status.busy":"2023-12-20T00:26:52.863327Z","iopub.execute_input":"2023-12-20T00:26:52.864245Z","iopub.status.idle":"2023-12-20T01:17:27.531901Z","shell.execute_reply.started":"2023-12-20T00:26:52.864209Z","shell.execute_reply":"2023-12-20T01:17:27.530965Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"training_time = end_time - start_time\n\ntraining_loss = history.history['loss']\nvalidation_loss = history.history['val_loss']\n\nlast_epoch_training_loss = training_loss[-1]\nlast_epoch_validation_loss = validation_loss[-1]\n\nprint(f\"Training Time is {training_time}\")\nprint(f\"Training Loss is {last_epoch_training_loss}\")\nprint(f\"Validation Loss is {last_epoch_validation_loss}\")","metadata":{"execution":{"iopub.status.busy":"2023-12-20T01:39:06.638482Z","iopub.execute_input":"2023-12-20T01:39:06.638909Z","iopub.status.idle":"2023-12-20T01:39:06.645479Z","shell.execute_reply.started":"2023-12-20T01:39:06.638875Z","shell.execute_reply":"2023-12-20T01:39:06.64443Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import matplotlib.pyplot as plt\nimport tensorflow as tf\nimport numpy as np\n\n# Up to first 10 images :10\nfor noisy_images, _ in train_dataset.take(1):\n    latent_vectors = encoder_model.predict(noisy_images[:10])\n    output_images = autoencoder.predict(noisy_images[:10])\n    break\n\n#Plotting\nfig, axes = plt.subplots(10, 3, figsize=(15, 30))\nfor i in range(10):\n    original_image = noisy_images[i].numpy()\n\n    output_image = output_images[i]\n\n    latent_vector = tf.squeeze(latent_vectors[i]).numpy()\n    latent_vector_processed = np.mean(latent_vector, axis=-1)\n\n    # Plot original image\n    axes[i, 0].imshow(original_image, cmap='gray')\n    axes[i, 0].set_title('Original Image')\n    axes[i, 0].axis('off')\n\n    # Plot processed latent vector\n    axes[i, 1].imshow(latent_vector_processed, cmap='gray')\n    axes[i, 1].set_title('Latent Vector')\n    axes[i, 1].axis('off')\n\n    # Plot output image\n    axes[i, 2].imshow(output_image, cmap='gray')\n    axes[i, 2].set_title('Output Image')\n    axes[i, 2].axis('off')\n\nplt.tight_layout()\nplt.show()\n\nplt.tight_layout()\nplt.savefig('output_plot.png')","metadata":{"execution":{"iopub.status.busy":"2023-12-20T01:40:54.917991Z","iopub.execute_input":"2023-12-20T01:40:54.918387Z","iopub.status.idle":"2023-12-20T01:41:05.477301Z","shell.execute_reply.started":"2023-12-20T01:40:54.918353Z","shell.execute_reply":"2023-12-20T01:41:05.476198Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"##2b\n\nfrom tensorflow.keras.models import Sequential\nfrom tensorflow.keras.layers import Conv2D, BatchNormalization, LeakyReLU, Flatten\n\n#Model_M\nmodel_M = Sequential([\n\n    Conv2D(64, kernel_size=(4, 4), padding='same', input_shape=(28, 28, 1), name='conv2d_5'),\n    LeakyReLU(name='leaky_re_lu_1'),\n\n    Conv2D(32, kernel_size=(4, 4), padding='same', name='conv2d_6'),\n    BatchNormalization(name='batch_normalization'),\n    LeakyReLU(name='leaky_re_lu_2'),\n\n    Conv2D(24, kernel_size=(4, 4), padding='same', name='conv2d_7'),\n    BatchNormalization(name='batch_normalization_1'),\n    LeakyReLU(name='leaky_re_lu_3'),\n\n\n    Conv2D(16, kernel_size=(4, 4), padding='same', name='conv2d_8'),\n    BatchNormalization(name='batch_normalization_2'),\n    LeakyReLU(name='leaky_re_lu_4'),\n\n    Conv2D(10, kernel_size=(3, 3), padding='same', name='conv2d_9'),\n    Flatten(name='flatten_1'),\n])\n\n# Load the weights\nweights_path = '/kaggle/input/fall-2023-cisc6000-deep-learning-hw2/M.weights.hdf5'\nmodel_M.load_weights(weights_path)\n\n\noptimizer = tf.keras.optimizers.Adam(learning_rate=0.001)\nmodel_M.compile(optimizer=optimizer, loss='categorical_crossentropy', metrics=['accuracy'])\nmodel_M.summary()\n","metadata":{"execution":{"iopub.status.busy":"2023-12-21T13:16:13.985701Z","iopub.execute_input":"2023-12-21T13:16:13.986245Z","iopub.status.idle":"2023-12-21T13:16:14.279933Z","shell.execute_reply.started":"2023-12-21T13:16:13.986195Z","shell.execute_reply":"2023-12-21T13:16:14.278547Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import os\nimport numpy as np\nfrom tensorflow.keras.preprocessing.image import load_img, img_to_array\n\ndef load_and_preprocess_images(directory, target_size=(28, 28)):\n    images = []\n    filenames = os.listdir(directory)\n    for filename in filenames:\n        img_path = os.path.join(directory, filename)\n        img = load_img(img_path, target_size=target_size, color_mode='grayscale')\n        img = img_to_array(img)\n        img /= 255.0  # Normalize the images\n        # No need to use np.expand_dims if img_to_array already creates a 3D array\n        images.append(img)\n    return np.array(images)\n\n#Load test_clean/noisy\ntest_clean_data = load_and_preprocess_images('/kaggle/input/fall-2023-cisc6000-deep-learning-hw2/train_clean_data/train_clean_data')\ntest_noisy_data = load_and_preprocess_images('/kaggle/input/fall-2023-cisc6000-deep-learning-hw2/train_noisy_data/train_noisy_data')","metadata":{"execution":{"iopub.status.busy":"2023-12-21T13:15:09.885196Z","iopub.execute_input":"2023-12-21T13:15:09.885723Z","iopub.status.idle":"2023-12-21T13:16:09.072427Z","shell.execute_reply.started":"2023-12-21T13:15:09.885682Z","shell.execute_reply":"2023-12-21T13:16:09.07073Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import pandas as pd\n\nlabels_df = pd.read_csv('/kaggle/input/fall-2023-cisc6000-deep-learning-hw2/labels.csv')\nlabels = labels_df.iloc[:, 1].values  \n\n# One-hot encode the labels\nlabels_encoded = to_categorical(labels, num_classes=10)\n\n# Predictions for clean and noisy data\npredictions_clean = model_M.predict(test_clean_data)\npredictions_noisy = model_M.predict(test_noisy_data)\n\n# Convert predictions to class labels\npredicted_classes_clean = np.argmax(predictions_clean, axis=1)\npredicted_classes_noisy = np.argmax(predictions_noisy, axis=1)\n\n# Function to calculate accuracy per class\ndef calculate_accuracy_per_class(predictions, true_labels):\n    accuracy_per_class = []\n    for class_id in range(10):  # For each class 0-9\n        class_indices = np.where(true_labels == class_id)[0]  # Indices of the current class\n        correct_preds = np.sum(predictions[class_indices] == class_id)\n        accuracy = correct_preds / len(class_indices)\n        accuracy_per_class.append(accuracy)\n    return accuracy_per_class\n\n# Calculate accuracy for each class\naccuracy_clean = calculate_accuracy_per_class(predicted_classes_clean, labels)\naccuracy_noisy = calculate_accuracy_per_class(predicted_classes_noisy, labels)\n\n# Print the results\nprint(\"Accuracy per class for Clean Data:\", accuracy_clean)\nprint(\"Accuracy per class for Noisy Data:\", accuracy_noisy)","metadata":{"execution":{"iopub.status.busy":"2023-12-21T13:16:21.0806Z","iopub.execute_input":"2023-12-21T13:16:21.08111Z","iopub.status.idle":"2023-12-21T13:17:42.279934Z","shell.execute_reply.started":"2023-12-21T13:16:21.081072Z","shell.execute_reply":"2023-12-21T13:17:42.278488Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]}]}