{"metadata":{"kernelspec":{"display_name":"Python 3","language":"python","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"},"widgets":{"application/vnd.jupyter.widget-state+json":{"state":{"108585220d384f1d87f3c8444587ae5b":{"model_module":"@jupyter-widgets/controls","model_module_version":"1.5.0","model_name":"IntProgressModel","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"IntProgressModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"ProgressView","bar_style":"success","description":"100%","description_tooltip":null,"layout":"IPY_MODEL_e4855ad98fad40deb899d57d401ae8c3","max":50210,"min":0,"orientation":"horizontal","style":"IPY_MODEL_f0b119ea39cc420ea11296b456d078d8","value":50210}},"12ab2aed987f4a2e9bed87f8df6fbace":{"model_module":"@jupyter-widgets/controls","model_module_version":"1.5.0","model_name":"HBoxModel","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HBoxModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HBoxView","box_style":"","children":["IPY_MODEL_b2e73348f1084afe92c3f03035d8606b","IPY_MODEL_b57249febd6c457baa5d4ba55f379ec1"],"layout":"IPY_MODEL_c9970f6bf5c54ff8808f97fdeb286846"}},"3540aa1072634a95b56adbc63b871c44":{"model_module":"@jupyter-widgets/base","model_module_version":"1.2.0","model_name":"LayoutModel","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"570a0b20e27c418687a1a1b252721597":{"model_module":"@jupyter-widgets/controls","model_module_version":"1.5.0","model_name":"DescriptionStyleModel","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"62710fc6ed3346d5894184aa95d3a0dd":{"model_module":"@jupyter-widgets/base","model_module_version":"1.2.0","model_name":"LayoutModel","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"75b62e2e75604016b1b25b7dfafd055e":{"model_module":"@jupyter-widgets/controls","model_module_version":"1.5.0","model_name":"ProgressStyleModel","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"ProgressStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","bar_color":null,"description_width":"initial"}},"7dc300f0a85241fe8385df414e8a5c97":{"model_module":"@jupyter-widgets/base","model_module_version":"1.2.0","model_name":"LayoutModel","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"819ad327d8de4df6ab1c59de568544d6":{"model_module":"@jupyter-widgets/base","model_module_version":"1.2.0","model_name":"LayoutModel","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"b1f711e1aaf24dc197f468a9a1d2dac2":{"model_module":"@jupyter-widgets/controls","model_module_version":"1.5.0","model_name":"DescriptionStyleModel","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"b2e45ab503b34cd68d5ffd0f81cb33f0":{"model_module":"@jupyter-widgets/controls","model_module_version":"1.5.0","model_name":"DescriptionStyleModel","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"b2e73348f1084afe92c3f03035d8606b":{"model_module":"@jupyter-widgets/controls","model_module_version":"1.5.0","model_name":"IntProgressModel","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"IntProgressModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"ProgressView","bar_style":"success","description":"100%","description_tooltip":null,"layout":"IPY_MODEL_3540aa1072634a95b56adbc63b871c44","max":50210,"min":0,"orientation":"horizontal","style":"IPY_MODEL_cfd07d8d18ed471fb9ea717da22823f0","value":50210}},"b57249febd6c457baa5d4ba55f379ec1":{"model_module":"@jupyter-widgets/controls","model_module_version":"1.5.0","model_name":"HTMLModel","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_dc99860fe34c4945b98db54863554527","placeholder":"​","style":"IPY_MODEL_570a0b20e27c418687a1a1b252721597","value":" 50210/50210 [00:34<00:00, 1472.92it/s]"}},"c40188b4079d4a32a3c40d43714ffdeb":{"model_module":"@jupyter-widgets/controls","model_module_version":"1.5.0","model_name":"IntProgressModel","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"IntProgressModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"ProgressView","bar_style":"success","description":"100%","description_tooltip":null,"layout":"IPY_MODEL_62710fc6ed3346d5894184aa95d3a0dd","max":50210,"min":0,"orientation":"horizontal","style":"IPY_MODEL_75b62e2e75604016b1b25b7dfafd055e","value":50210}},"c9970f6bf5c54ff8808f97fdeb286846":{"model_module":"@jupyter-widgets/base","model_module_version":"1.2.0","model_name":"LayoutModel","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"cd962aa4c7514f50a817344fcc4281c0":{"model_module":"@jupyter-widgets/controls","model_module_version":"1.5.0","model_name":"HTMLModel","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_7dc300f0a85241fe8385df414e8a5c97","placeholder":"​","style":"IPY_MODEL_b2e45ab503b34cd68d5ffd0f81cb33f0","value":" 50210/50210 [00:34<00:00, 1453.38it/s]"}},"cfd07d8d18ed471fb9ea717da22823f0":{"model_module":"@jupyter-widgets/controls","model_module_version":"1.5.0","model_name":"ProgressStyleModel","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"ProgressStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","bar_color":null,"description_width":"initial"}},"dc99860fe34c4945b98db54863554527":{"model_module":"@jupyter-widgets/base","model_module_version":"1.2.0","model_name":"LayoutModel","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"ddd90cb7822f4ceca0f5bbdb623d9153":{"model_module":"@jupyter-widgets/controls","model_module_version":"1.5.0","model_name":"HBoxModel","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HBoxModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HBoxView","box_style":"","children":["IPY_MODEL_108585220d384f1d87f3c8444587ae5b","IPY_MODEL_cd962aa4c7514f50a817344fcc4281c0"],"layout":"IPY_MODEL_e175ccd71bbc41ff8aa02b43f00de935"}},"e175ccd71bbc41ff8aa02b43f00de935":{"model_module":"@jupyter-widgets/base","model_module_version":"1.2.0","model_name":"LayoutModel","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"e4855ad98fad40deb899d57d401ae8c3":{"model_module":"@jupyter-widgets/base","model_module_version":"1.2.0","model_name":"LayoutModel","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"f0b119ea39cc420ea11296b456d078d8":{"model_module":"@jupyter-widgets/controls","model_module_version":"1.5.0","model_name":"ProgressStyleModel","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"ProgressStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","bar_color":null,"description_width":"initial"}},"f8a313a8731e4aee9ba940aa7bc795b0":{"model_module":"@jupyter-widgets/controls","model_module_version":"1.5.0","model_name":"HBoxModel","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HBoxModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HBoxView","box_style":"","children":["IPY_MODEL_c40188b4079d4a32a3c40d43714ffdeb","IPY_MODEL_ffa184123d1f45c5a7296f99465f2a08"],"layout":"IPY_MODEL_fcd628bac3174930bdb3727134928c69"}},"fcd628bac3174930bdb3727134928c69":{"model_module":"@jupyter-widgets/base","model_module_version":"1.2.0","model_name":"LayoutModel","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"ffa184123d1f45c5a7296f99465f2a08":{"model_module":"@jupyter-widgets/controls","model_module_version":"1.5.0","model_name":"HTMLModel","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_819ad327d8de4df6ab1c59de568544d6","placeholder":"​","style":"IPY_MODEL_b1f711e1aaf24dc197f468a9a1d2dac2","value":" 50210/50210 [00:34<00:00, 1436.32it/s]"}}},"version_major":2,"version_minor":0}},"kaggle":{"accelerator":"nvidiaTeslaT4","dataSources":[{"sourceId":14897,"databundleVersionId":1020216,"sourceType":"competition"},{"sourceId":861314,"sourceType":"datasetVersion","datasetId":456959}],"isInternetEnabled":false,"language":"python","sourceType":"notebook","isGpuEnabled":true}},"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 in \n\nimport numpy as np # linear algebra\nimport pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)\nfrom tqdm.auto import tqdm\nfrom glob import glob\nimport time, gc\nimport cv2\n\nfrom tensorflow import keras\nimport matplotlib.image as mpimg\nfrom tensorflow.keras.preprocessing.image import ImageDataGenerator\nfrom keras.models import Model\nfrom keras.models import clone_model\nfrom keras.layers import Dense,Conv2D,Flatten,MaxPool2D,Dropout,BatchNormalization, Input\nfrom keras.optimizers import Adam\nfrom keras.callbacks import ReduceLROnPlateau\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.metrics import confusion_matrix\nimport PIL.Image as Image, PIL.ImageDraw as ImageDraw, PIL.ImageFont as ImageFont\nfrom matplotlib import pyplot as plt\nimport seaborn as sns\n\n# Input data files are available in the \"../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# Any results you write to the current directory are saved as output.","metadata":{"_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","trusted":true,"execution":{"iopub.status.busy":"2025-01-07T17:14:58.208560Z","iopub.execute_input":"2025-01-07T17:14:58.208834Z","iopub.status.idle":"2025-01-07T17:14:58.814932Z","shell.execute_reply.started":"2025-01-07T17:14:58.208813Z","shell.execute_reply":"2025-01-07T17:14:58.814085Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train_df_ = pd.read_csv('/kaggle/input/bengaliai-cv19/train.csv')\ntest_df_ = pd.read_csv('/kaggle/input/bengaliai-cv19/test.csv')\nclass_map_df = pd.read_csv('/kaggle/input/bengaliai-cv19/class_map.csv')\nsample_sub_df = pd.read_csv('/kaggle/input/bengaliai-cv19/sample_submission.csv')","metadata":{"_cell_guid":"79c7e3d0-c299-4dcb-8224-4455121ee9b0","_uuid":"d629ff2d2480ee46fbb7e2d37f6b5fab8052498a","trusted":true,"execution":{"iopub.status.busy":"2025-01-07T17:15:07.104042Z","iopub.execute_input":"2025-01-07T17:15:07.104589Z","iopub.status.idle":"2025-01-07T17:15:07.388287Z","shell.execute_reply.started":"2025-01-07T17:15:07.104561Z","shell.execute_reply":"2025-01-07T17:15:07.387529Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train_df_.head()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T17:15:09.540536Z","iopub.execute_input":"2025-01-07T17:15:09.540815Z","iopub.status.idle":"2025-01-07T17:15:09.556371Z","shell.execute_reply.started":"2025-01-07T17:15:09.540793Z","shell.execute_reply":"2025-01-07T17:15:09.555682Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"test_df_.head()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T17:15:15.206295Z","iopub.execute_input":"2025-01-07T17:15:15.206563Z","iopub.status.idle":"2025-01-07T17:15:15.214414Z","shell.execute_reply.started":"2025-01-07T17:15:15.206543Z","shell.execute_reply":"2025-01-07T17:15:15.213470Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"sample_sub_df.head()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T17:15:17.390637Z","iopub.execute_input":"2025-01-07T17:15:17.390889Z","iopub.status.idle":"2025-01-07T17:15:17.398125Z","shell.execute_reply.started":"2025-01-07T17:15:17.390870Z","shell.execute_reply":"2025-01-07T17:15:17.397238Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"class_map_df.head()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T17:15:20.667966Z","iopub.execute_input":"2025-01-07T17:15:20.668260Z","iopub.status.idle":"2025-01-07T17:15:20.676174Z","shell.execute_reply.started":"2025-01-07T17:15:20.668240Z","shell.execute_reply":"2025-01-07T17:15:20.675309Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"print(f'Size of training data: {train_df_.shape}')\nprint(f'Size of test data: {test_df_.shape}')\nprint(f'Size of class map: {class_map_df.shape}')","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T17:15:22.467486Z","iopub.execute_input":"2025-01-07T17:15:22.467756Z","iopub.status.idle":"2025-01-07T17:15:22.472678Z","shell.execute_reply.started":"2025-01-07T17:15:22.467725Z","shell.execute_reply":"2025-01-07T17:15:22.471791Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"## Exploratory Data Analysis\nExploratory data analysis (EDA) is an approach to analyzing data sets to summarize their main characteristics, often with visual methods.","metadata":{}},{"cell_type":"code","source":"HEIGHT = 236\nWIDTH = 236\n\ndef get_n(df, field, n, top=True):\n    top_graphemes = df.groupby([field]).size().reset_index(name='counts')['counts'].sort_values(ascending=not top)[:n]\n    top_grapheme_roots = top_graphemes.index\n    top_grapheme_counts = top_graphemes.values\n    top_graphemes = class_map_df[class_map_df['component_type'] == field].reset_index().iloc[top_grapheme_roots]\n    top_graphemes.drop(['component_type', 'label'], axis=1, inplace=True)\n    top_graphemes.loc[:, 'count'] = top_grapheme_counts\n    return top_graphemes\n\ndef image_from_char(char):\n    image = Image.new('RGB', (WIDTH, HEIGHT))\n    draw = ImageDraw.Draw(image)\n    myfont = ImageFont.truetype('/kaggle/input/kalpurush-fonts/kalpurush-2.ttf', 120)\n    w, h = draw.textsize(char, font=myfont)\n    draw.text(((WIDTH - w) / 2,(HEIGHT - h) / 3), char, font=myfont)\n\n    return image","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T17:15:26.197552Z","iopub.execute_input":"2025-01-07T17:15:26.197859Z","iopub.status.idle":"2025-01-07T17:15:26.203934Z","shell.execute_reply.started":"2025-01-07T17:15:26.197831Z","shell.execute_reply":"2025-01-07T17:15:26.203078Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"### Number of unique values","metadata":{}},{"cell_type":"code","source":"print(f'Number of unique grapheme roots: {train_df_[\"grapheme_root\"].nunique()}')\nprint(f'Number of unique vowel diacritic: {train_df_[\"vowel_diacritic\"].nunique()}')\nprint(f'Number of unique consonant diacritic: {train_df_[\"consonant_diacritic\"].nunique()}')","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T17:15:30.780560Z","iopub.execute_input":"2025-01-07T17:15:30.780835Z","iopub.status.idle":"2025-01-07T17:15:30.792471Z","shell.execute_reply.started":"2025-01-07T17:15:30.780815Z","shell.execute_reply":"2025-01-07T17:15:30.791751Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"### Most used top 10 Grapheme Roots in training set","metadata":{}},{"cell_type":"code","source":"top_10_roots = get_n(train_df_, 'grapheme_root', 10)\ntop_10_roots","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T17:15:32.706470Z","iopub.execute_input":"2025-01-07T17:15:32.706748Z","iopub.status.idle":"2025-01-07T17:15:32.727431Z","shell.execute_reply.started":"2025-01-07T17:15:32.706728Z","shell.execute_reply":"2025-01-07T17:15:32.726695Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#f, ax = plt.subplots(2, 5, figsize=(16, 8))\n#ax = ax.flatten()\n\n#for i in range(10):\n    #ax[i].imshow(image_from_char(top_10_roots['component'].iloc[i]), cmap='Greys')","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T17:16:15.181214Z","iopub.execute_input":"2025-01-07T17:16:15.181602Z","iopub.status.idle":"2025-01-07T17:16:15.185028Z","shell.execute_reply.started":"2025-01-07T17:16:15.181567Z","shell.execute_reply":"2025-01-07T17:16:15.184249Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"### Least used 10 Grapheme Roots in training set","metadata":{}},{"cell_type":"code","source":"bottom_10_roots = get_n(train_df_, 'grapheme_root', 10, False)\nbottom_10_roots","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T17:16:18.289198Z","iopub.execute_input":"2025-01-07T17:16:18.289471Z","iopub.status.idle":"2025-01-07T17:16:18.302879Z","shell.execute_reply.started":"2025-01-07T17:16:18.289450Z","shell.execute_reply":"2025-01-07T17:16:18.302182Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#f, ax = plt.subplots(2, 5, figsize=(16, 8))\n#ax = ax.flatten()\n\n#for i in range(10):\n    #ax[i].imshow(image_from_char(bottom_10_roots['component'].iloc[i]), cmap='Greys')","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T17:16:57.099156Z","iopub.execute_input":"2025-01-07T17:16:57.099496Z","iopub.status.idle":"2025-01-07T17:16:57.102883Z","shell.execute_reply.started":"2025-01-07T17:16:57.099460Z","shell.execute_reply":"2025-01-07T17:16:57.102038Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"### Top 5 Vowel Diacritic in taining data","metadata":{}},{"cell_type":"code","source":"top_5_vowels = get_n(train_df_, 'vowel_diacritic', 5)\ntop_5_vowels","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T17:16:58.548394Z","iopub.execute_input":"2025-01-07T17:16:58.548674Z","iopub.status.idle":"2025-01-07T17:16:58.562252Z","shell.execute_reply.started":"2025-01-07T17:16:58.548653Z","shell.execute_reply":"2025-01-07T17:16:58.561536Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#f, ax = plt.subplots(1, 5, figsize=(16, 8))\n#ax = ax.flatten()\n\n#for i in range(5):\n    #ax[i].imshow(image_from_char(top_5_vowels['component'].iloc[i]), cmap='Greys')","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T17:17:16.738962Z","iopub.execute_input":"2025-01-07T17:17:16.739299Z","iopub.status.idle":"2025-01-07T17:17:16.742579Z","shell.execute_reply.started":"2025-01-07T17:17:16.739271Z","shell.execute_reply":"2025-01-07T17:17:16.741753Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"### Top 5 Consonant Diacritic in training data","metadata":{}},{"cell_type":"code","source":"top_5_consonants = get_n(train_df_, 'consonant_diacritic', 5)\ntop_5_consonants","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T17:17:19.543113Z","iopub.execute_input":"2025-01-07T17:17:19.543387Z","iopub.status.idle":"2025-01-07T17:17:19.556447Z","shell.execute_reply.started":"2025-01-07T17:17:19.543366Z","shell.execute_reply":"2025-01-07T17:17:19.555798Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"f, ax = plt.subplots(1, 5, figsize=(16, 8))\nax = ax.flatten()\n\nfor i in range(5):\n    ax[i].imshow(image_from_char(top_5_consonants['component'].iloc[i]), cmap='Greys')","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train_df_ = train_df_.drop(['grapheme'], axis=1, inplace=False)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T17:17:26.047404Z","iopub.execute_input":"2025-01-07T17:17:26.047696Z","iopub.status.idle":"2025-01-07T17:17:26.057852Z","shell.execute_reply.started":"2025-01-07T17:17:26.047672Z","shell.execute_reply":"2025-01-07T17:17:26.056640Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train_df_[['grapheme_root', 'vowel_diacritic', 'consonant_diacritic']] = train_df_[['grapheme_root', 'vowel_diacritic', 'consonant_diacritic']].astype('uint8')","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T17:17:27.715158Z","iopub.execute_input":"2025-01-07T17:17:27.715460Z","iopub.status.idle":"2025-01-07T17:17:27.725047Z","shell.execute_reply.started":"2025-01-07T17:17:27.715433Z","shell.execute_reply":"2025-01-07T17:17:27.724158Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"IMG_SIZE=64\nN_CHANNELS=1","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T17:17:30.326805Z","iopub.execute_input":"2025-01-07T17:17:30.327126Z","iopub.status.idle":"2025-01-07T17:17:30.330751Z","shell.execute_reply.started":"2025-01-07T17:17:30.327099Z","shell.execute_reply":"2025-01-07T17:17:30.329849Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"Let's apply some image processing (credits: [this kernel](https://www.kaggle.com/shawon10/bangla-graphemes-image-processing-deep-cnn)) while resizing the images, which will center crop the region of interest from the original images.","metadata":{}},{"cell_type":"code","source":"def resize(df, size=64, need_progress_bar=True):\n    resized = {}\n    resize_size=64\n    if need_progress_bar:\n        for i in tqdm(range(df.shape[0])):\n            image=df.loc[df.index[i]].values.reshape(137,236)\n            _, thresh = cv2.threshold(image, 30, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)\n            contours, _ = cv2.findContours(thresh,cv2.RETR_LIST,cv2.CHAIN_APPROX_SIMPLE)[-2:]\n\n            idx = 0 \n            ls_xmin = []\n            ls_ymin = []\n            ls_xmax = []\n            ls_ymax = []\n            for cnt in contours:\n                idx += 1\n                x,y,w,h = cv2.boundingRect(cnt)\n                ls_xmin.append(x)\n                ls_ymin.append(y)\n                ls_xmax.append(x + w)\n                ls_ymax.append(y + h)\n            xmin = min(ls_xmin)\n            ymin = min(ls_ymin)\n            xmax = max(ls_xmax)\n            ymax = max(ls_ymax)\n\n            roi = image[ymin:ymax,xmin:xmax]\n            resized_roi = cv2.resize(roi, (resize_size, resize_size),interpolation=cv2.INTER_AREA)\n            resized[df.index[i]] = resized_roi.reshape(-1)\n    else:\n        for i in range(df.shape[0]):\n            #image = cv2.resize(df.loc[df.index[i]].values.reshape(137,236),(size,size),None,fx=0.5,fy=0.5,interpolation=cv2.INTER_AREA)\n            image=df.loc[df.index[i]].values.reshape(137,236)\n            _, thresh = cv2.threshold(image, 30, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)\n            contours, _ = cv2.findContours(thresh,cv2.RETR_LIST,cv2.CHAIN_APPROX_SIMPLE)[-2:]\n\n            idx = 0 \n            ls_xmin = []\n            ls_ymin = []\n            ls_xmax = []\n            ls_ymax = []\n            for cnt in contours:\n                idx += 1\n                x,y,w,h = cv2.boundingRect(cnt)\n                ls_xmin.append(x)\n                ls_ymin.append(y)\n                ls_xmax.append(x + w)\n                ls_ymax.append(y + h)\n            xmin = min(ls_xmin)\n            ymin = min(ls_ymin)\n            xmax = max(ls_xmax)\n            ymax = max(ls_ymax)\n\n            roi = image[ymin:ymax,xmin:xmax]\n            resized_roi = cv2.resize(roi, (resize_size, resize_size),interpolation=cv2.INTER_AREA)\n            resized[df.index[i]] = resized_roi.reshape(-1)\n    resized = pd.DataFrame(resized).T\n    return resized","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T17:17:32.647442Z","iopub.execute_input":"2025-01-07T17:17:32.647765Z","iopub.status.idle":"2025-01-07T17:17:32.658083Z","shell.execute_reply.started":"2025-01-07T17:17:32.647737Z","shell.execute_reply":"2025-01-07T17:17:32.657123Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"def get_dummies(df):\n    cols = []\n    for col in df:\n        cols.append(pd.get_dummies(df[col].astype(str)))\n    return pd.concat(cols, axis=1)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T17:17:35.700523Z","iopub.execute_input":"2025-01-07T17:17:35.700812Z","iopub.status.idle":"2025-01-07T17:17:35.704904Z","shell.execute_reply.started":"2025-01-07T17:17:35.700786Z","shell.execute_reply":"2025-01-07T17:17:35.704091Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"## Basic Model","metadata":{}},{"cell_type":"code","source":"inputs = Input(shape = (IMG_SIZE, IMG_SIZE, 1))\ndef build_model():\n    model = Conv2D(filters=32, kernel_size=(3, 3), padding='SAME', activation='relu', input_shape=(IMG_SIZE, IMG_SIZE, 1))(inputs)\n    model = Conv2D(filters=32, kernel_size=(3, 3), padding='SAME', activation='relu')(model)\n    model = Conv2D(filters=32, kernel_size=(3, 3), padding='SAME', activation='relu')(model)\n    model = Conv2D(filters=32, kernel_size=(3, 3), padding='SAME', activation='relu')(model)\n    model = BatchNormalization(momentum=0.15)(model)\n    model = MaxPool2D(pool_size=(2, 2))(model)\n    model = Conv2D(filters=32, kernel_size=(5, 5), padding='SAME', activation='relu')(model)\n    model = Dropout(rate=0.3)(model)\n    \n    model = Conv2D(filters=64, kernel_size=(3, 3), padding='SAME', activation='relu')(model)\n    model = Conv2D(filters=64, kernel_size=(3, 3), padding='SAME', activation='relu')(model)\n    model = Conv2D(filters=64, kernel_size=(3, 3), padding='SAME', activation='relu')(model)\n    model = Conv2D(filters=64, kernel_size=(3, 3), padding='SAME', activation='relu')(model)\n    model = BatchNormalization(momentum=0.15)(model)\n    model = MaxPool2D(pool_size=(2, 2))(model)\n    model = Conv2D(filters=64, kernel_size=(5, 5), padding='SAME', activation='relu')(model)\n    model = BatchNormalization(momentum=0.15)(model)\n    model = Dropout(rate=0.3)(model)\n    \n    model = Conv2D(filters=128, kernel_size=(3, 3), padding='SAME', activation='relu')(model)\n    model = Conv2D(filters=128, kernel_size=(3, 3), padding='SAME', activation='relu')(model)\n    model = Conv2D(filters=128, kernel_size=(3, 3), padding='SAME', activation='relu')(model)\n    model = Conv2D(filters=128, kernel_size=(3, 3), padding='SAME', activation='relu')(model)\n    model = BatchNormalization(momentum=0.15)(model)\n    model = MaxPool2D(pool_size=(2, 2))(model)\n    model = Conv2D(filters=128, kernel_size=(5, 5), padding='SAME', activation='relu')(model)\n    model = BatchNormalization(momentum=0.15)(model)\n    model = Dropout(rate=0.3)(model)\n    \n    model = Conv2D(filters=256, kernel_size=(3, 3), padding='SAME', activation='relu')(model)\n    model = Conv2D(filters=256, kernel_size=(3, 3), padding='SAME', activation='relu')(model)\n    model = Conv2D(filters=256, kernel_size=(3, 3), padding='SAME', activation='relu')(model)\n    model = Conv2D(filters=256, kernel_size=(3, 3), padding='SAME', activation='relu')(model)\n    model = BatchNormalization(momentum=0.15)(model)\n    model = MaxPool2D(pool_size=(2, 2))(model)\n    model = Conv2D(filters=256, kernel_size=(5, 5), padding='SAME', activation='relu')(model)\n    model = BatchNormalization(momentum=0.15)(model)\n    model = Dropout(rate=0.3)(model)\n    \n    model = Flatten()(model)\n    model = Dense(1024, activation = \"relu\")(model)\n    model = Dropout(rate=0.3)(model)\n    dense = Dense(512, activation = \"relu\")(model)\n    \n    head_root = Dense(168, activation = 'softmax', name='output_root')(dense)\n    head_vowel = Dense(11, activation = 'softmax', name='output_vowel')(dense)\n    head_consonant = Dense(7, activation = 'softmax', name='output_consonant')(dense)\n    \n    model = Model(inputs=inputs, outputs=[head_root, head_vowel, head_consonant])\n    return model","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T17:40:18.397563Z","iopub.execute_input":"2025-01-07T17:40:18.398067Z","iopub.status.idle":"2025-01-07T17:40:18.418789Z","shell.execute_reply.started":"2025-01-07T17:40:18.398017Z","shell.execute_reply":"2025-01-07T17:40:18.417923Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#model.summary()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T17:17:51.090057Z","iopub.execute_input":"2025-01-07T17:17:51.090351Z","iopub.status.idle":"2025-01-07T17:17:51.093737Z","shell.execute_reply.started":"2025-01-07T17:17:51.090326Z","shell.execute_reply":"2025-01-07T17:17:51.092809Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"Let's visualize the 3-tailed (3 output) CNN by plotting it.","metadata":{}},{"cell_type":"code","source":"#from keras.utils import plot_model\n#plot_model(model, to_file='model.png')","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T17:18:00.681650Z","iopub.execute_input":"2025-01-07T17:18:00.681940Z","iopub.status.idle":"2025-01-07T17:18:00.685454Z","shell.execute_reply.started":"2025-01-07T17:18:00.681917Z","shell.execute_reply":"2025-01-07T17:18:00.684555Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T17:18:17.806134Z","iopub.execute_input":"2025-01-07T17:18:17.806409Z","iopub.status.idle":"2025-01-07T17:18:17.809913Z","shell.execute_reply.started":"2025-01-07T17:18:17.806389Z","shell.execute_reply":"2025-01-07T17:18:17.809050Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"# Set a learning rate annealer. Learning rate will be half after 3 epochs if accuracy is not increased\nlearning_rate_reduction_root = ReduceLROnPlateau(monitor='dense_3_accuracy', \n                                            patience=3, \n                                            verbose=1,\n                                            factor=0.5, \n                                            min_lr=0.00001)\nlearning_rate_reduction_vowel = ReduceLROnPlateau(monitor='dense_4_accuracy', \n                                            patience=3, \n                                            verbose=1,\n                                            factor=0.5, \n                                            min_lr=0.00001)\nlearning_rate_reduction_consonant = ReduceLROnPlateau(monitor='dense_5_accuracy', \n                                            patience=3, \n                                            verbose=1,\n                                            factor=0.5, \n                                            min_lr=0.00001)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T17:18:18.407225Z","iopub.execute_input":"2025-01-07T17:18:18.407530Z","iopub.status.idle":"2025-01-07T17:18:18.412576Z","shell.execute_reply.started":"2025-01-07T17:18:18.407506Z","shell.execute_reply":"2025-01-07T17:18:18.411591Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"batch_size = 256\nepochs = 3","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T18:01:22.780173Z","iopub.execute_input":"2025-01-07T18:01:22.780616Z","iopub.status.idle":"2025-01-07T18:01:22.784910Z","shell.execute_reply.started":"2025-01-07T18:01:22.780576Z","shell.execute_reply":"2025-01-07T18:01:22.783905Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"class MultiOutputDataGenerator(keras.preprocessing.image.ImageDataGenerator):\n\n    def flow(self,\n             x,\n             y=None,\n             batch_size=32,\n             shuffle=True,\n             sample_weight=None,\n             seed=None,\n             save_to_dir=None,\n             save_prefix='',\n             save_format='png',\n             subset=None):\n\n        targets = None\n        target_lengths = {}\n        ordered_outputs = []\n        for output, target in y.items():\n            if targets is None:\n                targets = target\n            else:\n                targets = np.concatenate((targets, target), axis=1)\n            target_lengths[output] = target.shape[1]\n            ordered_outputs.append(output)\n\n\n        for flowx, flowy in super().flow(x, targets, batch_size=batch_size,\n                                         shuffle=shuffle):\n            target_dict = {}\n            i = 0\n            for output in ordered_outputs:\n                target_length = target_lengths[output]\n                target_dict[output] = flowy[:, i: i + target_length]\n                i += target_length\n\n            yield flowx, target_dict","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T18:02:59.269744Z","iopub.execute_input":"2025-01-07T18:02:59.270194Z","iopub.status.idle":"2025-01-07T18:02:59.278304Z","shell.execute_reply.started":"2025-01-07T18:02:59.270154Z","shell.execute_reply":"2025-01-07T18:02:59.277054Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"HEIGHT = 137\nWIDTH = 236","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T18:03:01.655657Z","iopub.execute_input":"2025-01-07T18:03:01.656115Z","iopub.status.idle":"2025-01-07T18:03:01.660203Z","shell.execute_reply.started":"2025-01-07T18:03:01.656074Z","shell.execute_reply":"2025-01-07T18:03:01.659239Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train_df = pd.merge(pd.read_parquet(f'/kaggle/input/bengaliai-cv19/train_image_data_{i}.parquet'), train_df_, on='image_id').drop(['image_id'], axis=1)\n\n# Visualize few samples of current training dataset\nfig, ax = plt.subplots(nrows=3, ncols=4, figsize=(16, 8))\ncount=0\nfor row in ax:\n    for col in row:\n        col.imshow(resize(train_df.drop(['grapheme_root', 'vowel_diacritic', 'consonant_diacritic'], axis=1).iloc[[count]], need_progress_bar=False).values.reshape(-1).reshape(IMG_SIZE, IMG_SIZE).astype(np.float64))\n        count += 1\nplt.show()\n\nX_train = train_df.drop(['grapheme_root', 'vowel_diacritic', 'consonant_diacritic'], axis=1)\nX_train = resize(X_train)/255\n\n# CNN takes images in shape `(batch_size, h, w, channels)`, so reshape the images\nX_train = X_train.values.reshape(-1, IMG_SIZE, IMG_SIZE, N_CHANNELS)\n\nY_train_root = pd.get_dummies(train_df['grapheme_root']).values\nY_train_vowel = pd.get_dummies(train_df['vowel_diacritic']).values\nY_train_consonant = pd.get_dummies(train_df['consonant_diacritic']).values\n\nprint(f'Training images: {X_train.shape}')\nprint(f'Training labels root: {Y_train_root.shape}')\nprint(f'Training labels vowel: {Y_train_vowel.shape}')\nprint(f'Training labels consonants: {Y_train_consonant.shape}')\n\n# Divide the data into training and validation set\nx_train, x_test, y_train_root, y_test_root, y_train_vowel, y_test_vowel, y_train_consonant, y_test_consonant = train_test_split(X_train, Y_train_root, Y_train_vowel, Y_train_consonant, test_size=0.08, random_state=666)\ndel train_df\ndel X_train\ndel Y_train_root, Y_train_vowel, Y_train_consonant\n\n# Data augmentation for creating more training data\ndatagen = MultiOutputDataGenerator(\n    featurewise_center=False,  # set input mean to 0 over the dataset\n    samplewise_center=False,  # set each sample mean to 0\n    featurewise_std_normalization=False,  # divide inputs by std of the dataset\n    samplewise_std_normalization=False,  # divide each input by its std\n    zca_whitening=False,  # apply ZCA whitening\n    rotation_range=8,  # randomly rotate images in the range (degrees, 0 to 180)\n    zoom_range = 0.15, # Randomly zoom image \n    width_shift_range=0.15,  # randomly shift images horizontally (fraction of total width)\n    height_shift_range=0.15,  # randomly shift images vertically (fraction of total height)\n    horizontal_flip=False,  # randomly flip images\n    vertical_flip=False)  # randomly flip images\n\n\n# This will just calculate parameters required to augment the given data. This won't perform any augmentations\ndatagen.fit(x_train)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T18:03:02.666301Z","iopub.execute_input":"2025-01-07T18:03:02.666735Z","iopub.status.idle":"2025-01-07T18:04:47.211165Z","shell.execute_reply.started":"2025-01-07T18:03:02.666698Z","shell.execute_reply":"2025-01-07T18:04:47.210091Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"model = build_model()\n\nmodel.compile(\n    optimizer='adam',\n    loss={\n        'output_root': 'categorical_crossentropy',\n        'output_vowel': 'categorical_crossentropy',\n        'output_consonant': 'categorical_crossentropy'\n    },\n    metrics={\n        'output_root': ['accuracy'],\n        'output_vowel': ['accuracy'],\n        'output_consonant': ['accuracy']\n    }\n)\n\n \nhistory = model.fit(\n    datagen.flow(\n        x_train,\n        {\n            'output_root': y_train_root,\n            'output_vowel': y_train_vowel,\n            'output_consonant': y_train_consonant\n        },\n        batch_size=batch_size\n    ),\n    epochs=epochs,\n    validation_data=(x_test, [y_test_root, y_test_vowel, y_test_consonant]),\n    steps_per_epoch=x_train.shape[0] // batch_size\n)\n","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T17:42:53.696253Z","iopub.execute_input":"2025-01-07T17:42:53.696711Z","iopub.status.idle":"2025-01-07T17:44:36.127801Z","shell.execute_reply.started":"2025-01-07T17:42:53.696665Z","shell.execute_reply":"2025-01-07T17:44:36.126180Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"### Training loop","metadata":{}},{"cell_type":"code","source":"import psutil\nimport cProfile\nimport pstats\nimport time\nimport tensorflow as tf\nwith tf.device('/CPU:0'):\n    model = build_model()\n\n    model.compile(\n        optimizer='adam',\n        loss={\n            'output_root': 'categorical_crossentropy',\n            'output_vowel': 'categorical_crossentropy',\n            'output_consonant': 'categorical_crossentropy'\n        },\n        metrics={\n            'output_root': ['accuracy'],\n            'output_vowel': ['accuracy'],\n            'output_consonant': ['accuracy']\n        }\n    )\n    \n    # Record initial memory usage\n    process = psutil.Process()\n    initial_memory = process.memory_info().rss / 1024 ** 2  # Memory in MB\n    profiler = cProfile.Profile()\n    profiler.enable()\n    \n    start_time = time.time()\n    history = model.fit(\n        datagen.flow(\n            x_train,\n            {\n                'output_root': y_train_root,\n                'output_vowel': y_train_vowel,\n                'output_consonant': y_train_consonant\n            },\n            batch_size=batch_size\n        ),\n        epochs=epochs,\n        validation_data=(x_test, [y_test_root, y_test_vowel, y_test_consonant]),\n        steps_per_epoch=10\n    )\n\n    gpu_time = time.time() - start_time\n    \n    # Final memory usage\n    final_memory = process.memory_info().rss / 1024 ** 2  # Memory in MB\n    memory_usage = final_memory - initial_memory\n    \n    # Calculate throughput (samples per second)\n    throughput = len(x_train) / gpu_time\n\n    profiler.disable()\n\n    # Print the profiling results\n    stats = pstats.Stats(profiler)\n    stats.sort_stats('time').print_stats(10)  # Top 10 time-consuming functions\n\n    print()\n    \n    print(f'CPU Training time: {gpu_time:.4f} seconds')\n    print(f'CPU Memory usage: {memory_usage:.4f} MB')\n    print(f'CPU Throughput: {throughput:.4f} samples/second')\n\n    print('='*50)\n#-------------------------------------------------------------------------------------------\n    print('Scheduler :')\n    from tensorflow.keras.callbacks import ReduceLROnPlateau\n\n    # Reduce learning rate when validation loss stops improving\n    # Set a learning rate annealer. Learning rate will be half after 3 epochs if accuracy is not increased\n    learning_rate_reduction_root = ReduceLROnPlateau(monitor='dense_3_accuracy', \n                                                patience=3, \n                                                verbose=1,\n                                                factor=0.5, \n                                                min_lr=0.00001)\n    learning_rate_reduction_vowel = ReduceLROnPlateau(monitor='dense_4_accuracy', \n                                                patience=3, \n                                                verbose=1,\n                                                factor=0.5, \n                                                min_lr=0.00001)\n    learning_rate_reduction_consonant = ReduceLROnPlateau(monitor='dense_5_accuracy', \n                                                patience=3, \n                                                verbose=1,\n                                                factor=0.5, \n                                                min_lr=0.00001)\n\n    # Record initial memory usage\n    process = psutil.Process()\n    initial_memory = process.memory_info().rss / 1024 ** 2  # Memory in MB\n    profiler = cProfile.Profile()\n    profiler.enable()\n    \n    start_time = time.time()\n    history = model.fit(\n        datagen.flow(\n            x_train,\n            {\n                'output_root': y_train_root,\n                'output_vowel': y_train_vowel,\n                'output_consonant': y_train_consonant\n            },\n            batch_size=batch_size\n        ),\n        epochs=epochs,\n        validation_data=(x_test, [y_test_root, y_test_vowel, y_test_consonant]),\n        steps_per_epoch=10,\n        callbacks=[learning_rate_reduction_root, learning_rate_reduction_vowel, learning_rate_reduction_consonant]\n    )\n    gpu_time = time.time() - start_time\n    \n    # Final memory usage\n    final_memory = process.memory_info().rss / 1024 ** 2  # Memory in MB\n    memory_usage = final_memory - initial_memory\n    \n    # Calculate throughput (samples per second)\n    throughput = len(x_train) / gpu_time\n\n    profiler.disable()\n\n    # Print the profiling results\n    stats = pstats.Stats(profiler)\n    stats.sort_stats('time').print_stats(10)  # Top 10 time-consuming functions\n\n    print()\n    \n    print(f'CPU Training time: {gpu_time:.4f} seconds')\n    print(f'CPU Memory usage: {memory_usage:.4f} MB')\n    print(f'CPU Throughput: {throughput:.4f} samples/second')\n\n    print('='*50)\n    #-------------------------------------------------------------------------------------------\n    print('Early stopping :')\n    from tensorflow.keras.callbacks import EarlyStopping\n\n    # Stop training if validation loss doesn't improve for 5 epochs\n    early_stopping = EarlyStopping(monitor='val_loss', patience=5, restore_best_weights=True, verbose=1)\n\n\n    # Record initial memory usage\n    process = psutil.Process()\n    initial_memory = process.memory_info().rss / 1024 ** 2  # Memory in MB\n    profiler = cProfile.Profile()\n    profiler.enable()\n    \n    start_time = time.time()\n    history = model.fit(\n        datagen.flow(\n            x_train,\n            {\n                'output_root': y_train_root,\n                'output_vowel': y_train_vowel,\n                'output_consonant': y_train_consonant\n            },\n            batch_size=batch_size\n        ),\n        epochs=epochs,\n        validation_data=(x_test, [y_test_root, y_test_vowel, y_test_consonant]),\n        steps_per_epoch=10,\n        callbacks=[early_stopping]\n    )\n                              \n    gpu_time = time.time() - start_time\n    \n    # Final memory usage\n    final_memory = process.memory_info().rss / 1024 ** 2  # Memory in MB\n    memory_usage = final_memory - initial_memory\n    \n    # Calculate throughput (samples per second)\n    throughput = len(x_train) / gpu_time\n\n    profiler.disable()\n\n    # Print the profiling results\n    stats = pstats.Stats(profiler)\n    stats.sort_stats('time').print_stats(10)  # Top 10 time-consuming functions\n\n    print()\n    \n    print(f'CPU Training time: {gpu_time:.4f} seconds')\n    print(f'CPU Memory usage: {memory_usage:.4f} MB')\n    print(f'CPU Throughput: {throughput:.4f} samples/second')\n\n    print('='*50)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T18:16:36.814520Z","iopub.execute_input":"2025-01-07T18:16:36.814968Z","iopub.status.idle":"2025-01-07T19:02:35.647975Z","shell.execute_reply.started":"2025-01-07T18:16:36.814927Z","shell.execute_reply":"2025-01-07T19:02:35.647196Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"import psutil\nimport cProfile\nimport pstats\nimport time\nwith tf.device('/GPU:0'):\n    model = build_model()\n\n    model.compile(\n        optimizer='adam',\n        loss={\n            'output_root': 'categorical_crossentropy',\n            'output_vowel': 'categorical_crossentropy',\n            'output_consonant': 'categorical_crossentropy'\n        },\n        metrics={\n            'output_root': ['accuracy'],\n            'output_vowel': ['accuracy'],\n            'output_consonant': ['accuracy']\n        }\n    )\n    # Record initial memory usage\n    process = psutil.Process()\n    initial_memory = process.memory_info().rss / 1024 ** 2  # Memory in MB\n    profiler = cProfile.Profile()\n    profiler.enable()\n    \n    start_time = time.time()\n    history = model.fit(\n        datagen.flow(\n            x_train,\n            {\n                'output_root': y_train_root,\n                'output_vowel': y_train_vowel,\n                'output_consonant': y_train_consonant\n            },\n            batch_size=batch_size\n        ),\n        epochs=epochs,\n        validation_data=(x_test, [y_test_root, y_test_vowel, y_test_consonant]),\n        steps_per_epoch=10\n    )\n    gpu_time = time.time() - start_time\n    \n    # Final memory usage\n    final_memory = process.memory_info().rss / 1024 ** 2  # Memory in MB\n    memory_usage = final_memory - initial_memory\n    \n    # Calculate throughput (samples per second)\n    throughput = len(x_train) / gpu_time\n\n    profiler.disable()\n\n    # Print the profiling results\n    stats = pstats.Stats(profiler)\n    stats.sort_stats('time').print_stats(10)  # Top 10 time-consuming functions\n\n    print()\n    \n    print(f'GPU Training time: {gpu_time:.4f} seconds')\n    print(f'GPU Memory usage: {memory_usage:.4f} MB')\n    print(f'GPU Throughput: {throughput:.4f} samples/second')\n\n    print('='*50)\n#-------------------------------------------------------------------------------------------\n    print('Scheduler :')\n    from tensorflow.keras.callbacks import ReduceLROnPlateau\n\n    # Reduce learning rate when validation loss stops improving\n    # Set a learning rate annealer. Learning rate will be half after 3 epochs if accuracy is not increased\n    learning_rate_reduction_root = ReduceLROnPlateau(monitor='dense_3_accuracy', \n                                                patience=3, \n                                                verbose=1,\n                                                factor=0.5, \n                                                min_lr=0.00001)\n    learning_rate_reduction_vowel = ReduceLROnPlateau(monitor='dense_4_accuracy', \n                                                patience=3, \n                                                verbose=1,\n                                                factor=0.5, \n                                                min_lr=0.00001)\n    learning_rate_reduction_consonant = ReduceLROnPlateau(monitor='dense_5_accuracy', \n                                                patience=3, \n                                                verbose=1,\n                                                factor=0.5, \n                                                min_lr=0.00001)\n\n    # Record initial memory usage\n    process = psutil.Process()\n    initial_memory = process.memory_info().rss / 1024 ** 2  # Memory in MB\n    profiler = cProfile.Profile()\n    profiler.enable()\n    \n    start_time = time.time()\n    history = model.fit(\n        datagen.flow(\n            x_train,\n            {\n                'output_root': y_train_root,\n                'output_vowel': y_train_vowel,\n                'output_consonant': y_train_consonant\n            },\n            batch_size=batch_size\n        ),\n        epochs=epochs,\n        validation_data=(x_test, [y_test_root, y_test_vowel, y_test_consonant]),\n        steps_per_epoch=10,\n        callbacks=[learning_rate_reduction_root, learning_rate_reduction_vowel, learning_rate_reduction_consonant] )\n    gpu_time = time.time() - start_time\n    \n    # Final memory usage\n    final_memory = process.memory_info().rss / 1024 ** 2  # Memory in MB\n    memory_usage = final_memory - initial_memory\n    \n    # Calculate throughput (samples per second)\n    throughput = len(x_train) / gpu_time\n\n    profiler.disable()\n\n    # Print the profiling results\n    stats = pstats.Stats(profiler)\n    stats.sort_stats('time').print_stats(10)  # Top 10 time-consuming functions\n\n    print()\n    \n    print(f'GPU Training time: {gpu_time:.4f} seconds')\n    print(f'GPU Memory usage: {memory_usage:.4f} MB')\n    print(f'GPU Throughput: {throughput:.4f} samples/second')\n\n    print('='*50)\n    #-------------------------------------------------------------------------------------------\n    print('Early stopping :')\n    from tensorflow.keras.callbacks import EarlyStopping\n\n    # Stop training if validation loss doesn't improve for 5 epochs\n    early_stopping = EarlyStopping(monitor='val_loss', patience=5, restore_best_weights=True, verbose=1)\n\n\n    # Record initial memory usage\n    process = psutil.Process()\n    initial_memory = process.memory_info().rss / 1024 ** 2  # Memory in MB\n    profiler = cProfile.Profile()\n    profiler.enable()\n    \n    start_time = time.time()\n    history = model.fit(\n        datagen.flow(\n            x_train,\n            {\n                'output_root': y_train_root,\n                'output_vowel': y_train_vowel,\n                'output_consonant': y_train_consonant\n            },\n            batch_size=batch_size\n        ),\n        epochs=epochs,\n        validation_data=(x_test, [y_test_root, y_test_vowel, y_test_consonant]),\n        steps_per_epoch=10,\n        callbacks=[early_stopping] )\n    gpu_time = time.time() - start_time\n    \n    # Final memory usage\n    final_memory = process.memory_info().rss / 1024 ** 2  # Memory in MB\n    memory_usage = final_memory - initial_memory\n    \n    # Calculate throughput (samples per second)\n    throughput = len(x_train) / gpu_time\n\n    profiler.disable()\n\n    # Print the profiling results\n    stats = pstats.Stats(profiler)\n    stats.sort_stats('time').print_stats(10)  # Top 10 time-consuming functions\n\n    print()\n    \n    print(f'GPU Training time: {gpu_time:.4f} seconds')\n    print(f'GPU Memory usage: {memory_usage:.4f} MB')\n    print(f'GPU Throughput: {throughput:.4f} samples/second')\n\n    print('='*50)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T19:02:35.649166Z","iopub.execute_input":"2025-01-07T19:02:35.649488Z","iopub.status.idle":"2025-01-07T19:03:33.178832Z","shell.execute_reply.started":"2025-01-07T19:02:35.649442Z","shell.execute_reply":"2025-01-07T19:03:33.177918Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#-------------------------------------------------------------------------------------------\nprint('Mirror Strategy :')\n\n    # Define the strategy\nstrategy = tf.distribute.MirroredStrategy()\n\n# Print the number of devices\nprint(f\"Number of devices: {strategy.num_replicas_in_sync}\")\n\nwith strategy.scope():\n    model = build_model()\n\n    model.compile(\n        optimizer='adam',\n        loss={\n            'output_root': 'categorical_crossentropy',\n            'output_vowel': 'categorical_crossentropy',\n            'output_consonant': 'categorical_crossentropy'\n        },\n        metrics={\n            'output_root': ['accuracy'],\n            'output_vowel': ['accuracy'],\n            'output_consonant': ['accuracy']\n        }\n    )\n\n\n\n# Record initial memory usage\nprocess = psutil.Process()\ninitial_memory = process.memory_info().rss / 1024 ** 2  # Memory in MB\nprofiler = cProfile.Profile()\nprofiler.enable()\n\nstart_time = time.time()\nhistory = model.fit(\n        datagen.flow(\n            x_train,\n            {\n                'output_root': y_train_root,\n                'output_vowel': y_train_vowel,\n                'output_consonant': y_train_consonant\n            },\n            batch_size=batch_size\n        ),\n        epochs=epochs,\n        validation_data=(x_test, [y_test_root, y_test_vowel, y_test_consonant]),\n        steps_per_epoch=10\n    )\ngpu_time = time.time() - start_time\n\n# Final memory usage\nfinal_memory = process.memory_info().rss / 1024 ** 2  # Memory in MB\nmemory_usage = final_memory - initial_memory\n\n# Calculate throughput (samples per second)\nthroughput = len(x_train) / gpu_time\n\nprofiler.disable()\n\n# Print the profiling results\nstats = pstats.Stats(profiler)\nstats.sort_stats('time').print_stats(10)  # Top 10 time-consuming functions\n\nprint()\n\nprint(f'GPU Training time: {gpu_time:.4f} seconds')\nprint(f'GPU Memory usage: {memory_usage:.4f} MB')\nprint(f'GPU Throughput: {throughput:.4f} samples/second')\n\nprint('='*50)\n#-------------------------------------------------------------------------------------------\nprint('Multi worker Mirror strategy :')\n\n# Define the strategy\nstrategy = tf.distribute.MultiWorkerMirroredStrategy()\n\nwith strategy.scope():\n    model = build_model()\n\n    model.compile(\n        optimizer='adam',\n        loss={\n            'output_root': 'categorical_crossentropy',\n            'output_vowel': 'categorical_crossentropy',\n            'output_consonant': 'categorical_crossentropy'\n        },\n        metrics={\n            'output_root': ['accuracy'],\n            'output_vowel': ['accuracy'],\n            'output_consonant': ['accuracy']\n        }\n    )\n\n\n# Record initial memory usage\nprocess = psutil.Process()\ninitial_memory = process.memory_info().rss / 1024 ** 2  # Memory in MB\nprofiler = cProfile.Profile()\nprofiler.enable()\n\nstart_time = time.time()\nhistory = model.fit(\n        datagen.flow(\n            x_train,\n            {\n                'output_root': y_train_root,\n                'output_vowel': y_train_vowel,\n                'output_consonant': y_train_consonant\n            },\n            batch_size=batch_size\n        ),\n        epochs=epochs,\n        validation_data=(x_test, [y_test_root, y_test_vowel, y_test_consonant]),\n        steps_per_epoch=10\n    )\ngpu_time = time.time() - start_time\n\n# Final memory usage\nfinal_memory = process.memory_info().rss / 1024 ** 2  # Memory in MB\nmemory_usage = final_memory - initial_memory\n\n# Calculate throughput (samples per second)\nthroughput = len(x_train) / gpu_time\n\nprofiler.disable()\n\n# Print the profiling results\nstats = pstats.Stats(profiler)\nstats.sort_stats('time').print_stats(10)  # Top 10 time-consuming functions\n\nprint()\n\nprint(f'GPU Training time: {gpu_time:.4f} seconds')\nprint(f'GPU Memory usage: {memory_usage:.4f} MB')\nprint(f'GPU Throughput: {throughput:.4f} samples/second')\n\nprint('='*50)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-01-07T19:03:33.180256Z","iopub.execute_input":"2025-01-07T19:03:33.180500Z","iopub.status.idle":"2025-01-07T19:05:07.405813Z","shell.execute_reply.started":"2025-01-07T19:03:33.180478Z","shell.execute_reply":"2025-01-07T19:05:07.405102Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"histories = []\nfor i in range(1):\n    train_df = pd.merge(pd.read_parquet(f'/kaggle/input/bengaliai-cv19/train_image_data_{i}.parquet'), train_df_, on='image_id').drop(['image_id'], axis=1)\n    \n    # Visualize few samples of current training dataset\n    fig, ax = plt.subplots(nrows=3, ncols=4, figsize=(16, 8))\n    count=0\n    for row in ax:\n        for col in row:\n            col.imshow(resize(train_df.drop(['grapheme_root', 'vowel_diacritic', 'consonant_diacritic'], axis=1).iloc[[count]], need_progress_bar=False).values.reshape(-1).reshape(IMG_SIZE, IMG_SIZE).astype(np.float64))\n            count += 1\n    plt.show()\n    \n    X_train = train_df.drop(['grapheme_root', 'vowel_diacritic', 'consonant_diacritic'], axis=1)\n    X_train = resize(X_train)/255\n    \n    # CNN takes images in shape `(batch_size, h, w, channels)`, so reshape the images\n    X_train = X_train.values.reshape(-1, IMG_SIZE, IMG_SIZE, N_CHANNELS)\n    \n    Y_train_root = pd.get_dummies(train_df['grapheme_root']).values\n    Y_train_vowel = pd.get_dummies(train_df['vowel_diacritic']).values\n    Y_train_consonant = pd.get_dummies(train_df['consonant_diacritic']).values\n\n    print(f'Training images: {X_train.shape}')\n    print(f'Training labels root: {Y_train_root.shape}')\n    print(f'Training labels vowel: {Y_train_vowel.shape}')\n    print(f'Training labels consonants: {Y_train_consonant.shape}')\n\n    # Divide the data into training and validation set\n    x_train, x_test, y_train_root, y_test_root, y_train_vowel, y_test_vowel, y_train_consonant, y_test_consonant = train_test_split(X_train, Y_train_root, Y_train_vowel, Y_train_consonant, test_size=0.08, random_state=666)\n    del train_df\n    del X_train\n    del Y_train_root, Y_train_vowel, Y_train_consonant\n\n    # Data augmentation for creating more training data\n    datagen = MultiOutputDataGenerator(\n        featurewise_center=False,  # set input mean to 0 over the dataset\n        samplewise_center=False,  # set each sample mean to 0\n        featurewise_std_normalization=False,  # divide inputs by std of the dataset\n        samplewise_std_normalization=False,  # divide each input by its std\n        zca_whitening=False,  # apply ZCA whitening\n        rotation_range=8,  # randomly rotate images in the range (degrees, 0 to 180)\n        zoom_range = 0.15, # Randomly zoom image \n        width_shift_range=0.15,  # randomly shift images horizontally (fraction of total width)\n        height_shift_range=0.15,  # randomly shift images vertically (fraction of total height)\n        horizontal_flip=False,  # randomly flip images\n        vertical_flip=False)  # randomly flip images\n\n\n    # This will just calculate parameters required to augment the given data. This won't perform any augmentations\n    datagen.fit(x_train)\n\n    # Fit the model\n    history = model.fit_generator(datagen.flow(x_train, {'dense_3': y_train_root, 'dense_4': y_train_vowel, 'dense_5': y_train_consonant}, batch_size=batch_size),\n                              epochs = epochs, validation_data = (x_test, [y_test_root, y_test_vowel, y_test_consonant]), \n                              steps_per_epoch=x_train.shape[0] // batch_size, \n                              callbacks=[learning_rate_reduction_root, learning_rate_reduction_vowel, learning_rate_reduction_consonant])\n\n    histories.append(history)\n    \n    # Delete to reduce memory usage\n    del x_train\n    del x_test\n    del y_train_root\n    del y_test_root\n    del y_train_vowel\n    del y_test_vowel\n    del y_train_consonant\n    del y_test_consonant\n    gc.collect()","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"%matplotlib inline\ndef plot_loss(his, epoch, title):\n    plt.style.use('ggplot')\n    plt.figure()\n    plt.plot(np.arange(0, epoch), his.history['loss'], label='train_loss')\n    plt.plot(np.arange(0, epoch), his.history['dense_3_loss'], label='train_root_loss')\n    plt.plot(np.arange(0, epoch), his.history['dense_4_loss'], label='train_vowel_loss')\n    plt.plot(np.arange(0, epoch), his.history['dense_5_loss'], label='train_consonant_loss')\n    \n    plt.plot(np.arange(0, epoch), his.history['val_dense_3_loss'], label='val_train_root_loss')\n    plt.plot(np.arange(0, epoch), his.history['val_dense_4_loss'], label='val_train_vowel_loss')\n    plt.plot(np.arange(0, epoch), his.history['val_dense_5_loss'], label='val_train_consonant_loss')\n    \n    plt.title(title)\n    plt.xlabel('Epoch #')\n    plt.ylabel('Loss')\n    plt.legend(loc='upper right')\n    plt.show()\n\ndef plot_acc(his, epoch, title):\n    plt.style.use('ggplot')\n    plt.figure()\n    plt.plot(np.arange(0, epoch), his.history['dense_3_accuracy'], label='train_root_acc')\n    plt.plot(np.arange(0, epoch), his.history['dense_4_accuracy'], label='train_vowel_accuracy')\n    plt.plot(np.arange(0, epoch), his.history['dense_5_accuracy'], label='train_consonant_accuracy')\n    \n    plt.plot(np.arange(0, epoch), his.history['val_dense_3_accuracy'], label='val_root_acc')\n    plt.plot(np.arange(0, epoch), his.history['val_dense_4_accuracy'], label='val_vowel_accuracy')\n    plt.plot(np.arange(0, epoch), his.history['val_dense_5_accuracy'], label='val_consonant_accuracy')\n    plt.title(title)\n    plt.xlabel('Epoch #')\n    plt.ylabel('Accuracy')\n    plt.legend(loc='upper right')\n    plt.show()","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"for dataset in range(4):\n    plot_loss(histories[dataset], epochs, f'Training Dataset: {dataset}')\n    plot_acc(histories[dataset], epochs, f'Training Dataset: {dataset}')","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"del histories\ngc.collect()","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"preds_dict = {\n    'grapheme_root': [],\n    'vowel_diacritic': [],\n    'consonant_diacritic': []\n}","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"components = ['consonant_diacritic', 'grapheme_root', 'vowel_diacritic']\ntarget=[] # model predictions placeholder\nrow_id=[] # row_id place holder\nfor i in range(4):\n    df_test_img = pd.read_parquet('/kaggle/input/bengaliai-cv19/test_image_data_{}.parquet'.format(i)) \n    df_test_img.set_index('image_id', inplace=True)\n\n    X_test = resize(df_test_img, need_progress_bar=False)/255\n    X_test = X_test.values.reshape(-1, IMG_SIZE, IMG_SIZE, N_CHANNELS)\n    \n    preds = model.predict(X_test)\n\n    for i, p in enumerate(preds_dict):\n        preds_dict[p] = np.argmax(preds[i], axis=1)\n\n    for k,id in enumerate(df_test_img.index.values):  \n        for i,comp in enumerate(components):\n            id_sample=id+'_'+comp\n            row_id.append(id_sample)\n            target.append(preds_dict[comp][k])\n    del df_test_img\n    del X_test\n    gc.collect()\n\ndf_sample = pd.DataFrame(\n    {\n        'row_id': row_id,\n        'target':target\n    },\n    columns = ['row_id','target'] \n)\ndf_sample.to_csv('submission.csv',index=False)\ndf_sample.head()","metadata":{"trusted":true},"outputs":[],"execution_count":null}]}