Project_Yusuf_GUR_2532810_30_07_benim.ipynb

{"nbformat":4,"nbformat_minor":0,"metadata":{"anaconda-cloud":{},"kernelspec":{"name":"python3","display_name":"Python 3"},"language_info":{"codemirror_mode":{"name":"ipython","version":3},"file_extension":".py","mimetype":"text/x-python","name":"python","nbconvert_exporter":"python","pygments_lexer":"ipython3","version":"3.6.3"},"colab":{"name":"Project_Yusuf_GUR_2532810_30_07_benim.ipynb","provenance":[],"collapsed_sections":["ewM7CGodswEy","AH7hEqiJzjrO"]},"accelerator":"GPU","widgets":{"application/vnd.jupyter.widget-state+json":{"d784668f947048c9b273b974d7de63e3":{"model_module":"@jupyter-widgets/controls","model_name":"HBoxModel","state":{"_view_name":"HBoxView","_dom_classes":[],"_model_name":"HBoxModel","_view_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_view_count":null,"_view_module_version":"1.5.0","box_style":"","layout":"IPY_MODEL_1687ef5de81f492284e17b0937a509b0","_model_module":"@jupyter-widgets/controls","children":["IPY_MODEL_5d8b4627dea24864b2619be7022a9505","IPY_MODEL_c33761d8df394b32aa90e27c444042fd"]}},"1687ef5de81f492284e17b0937a509b0":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","state":{"_view_name":"LayoutView","grid_template_rows":null,"right":null,"justify_content":null,"_view_module":"@jupyter-widgets/base","overflow":null,"_model_module_version":"1.2.0","_view_count":null,"flex_flow":null,"width":null,"min_width":null,"border":null,"align_items":null,"bottom":null,"_model_module":"@jupyter-widgets/base","top":null,"grid_column":null,"overflow_y":null,"overflow_x":null,"grid_auto_flow":null,"grid_area":null,"grid_template_columns":null,"flex":null,"_model_name":"LayoutModel","justify_items":null,"grid_row":null,"max_height":null,"align_content":null,"visibility":null,"align_self":null,"height":null,"min_height":null,"padding":null,"grid_auto_rows":null,"grid_gap":null,"max_width":null,"order":null,"_view_module_version":"1.2.0","grid_template_areas":null,"object_position":null,"object_fit":null,"grid_auto_columns":null,"margin":null,"display":null,"left":null}},"5d8b4627dea24864b2619be7022a9505":{"model_module":"@jupyter-widgets/controls","model_name":"FloatProgressModel","state":{"_view_name":"ProgressView","style":"IPY_MODEL_bf57c810dbc7424e93b7290ac71e6f14","_dom_classes":[],"description":"100%","_model_name":"FloatProgressModel","bar_style":"success","max":46830571,"_view_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","value":46830571,"_view_count":null,"_view_module_version":"1.5.0","orientation":"horizontal","min":0,"description_tooltip":null,"_model_module":"@jupyter-widgets/controls","layout":"IPY_MODEL_ce131cdc0a774dac92f2481f2323aea7"}},"c33761d8df394b32aa90e27c444042fd":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","state":{"_view_name":"HTMLView","style":"IPY_MODEL_d972c5066342443d85c00d4832938ad3","_dom_classes":[],"description":"","_model_name":"HTMLModel","placeholder":"​","_view_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","value":" 44.7M/44.7M [01:58&lt;00:00, 394kB/s]","_view_count":null,"_view_module_version":"1.5.0","description_tooltip":null,"_model_module":"@jupyter-widgets/controls","layout":"IPY_MODEL_5e3b828e7ba7406b9b32ecc302c8b3c8"}},"bf57c810dbc7424e93b7290ac71e6f14":{"model_module":"@jupyter-widgets/controls","model_name":"ProgressStyleModel","state":{"_view_name":"StyleView","_model_name":"ProgressStyleModel","description_width":"initial","_view_module":"@jupyter-widgets/base","_model_module_version":"1.5.0","_view_count":null,"_view_module_version":"1.2.0","bar_color":null,"_model_module":"@jupyter-widgets/controls"}},"ce131cdc0a774dac92f2481f2323aea7":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","state":{"_view_name":"LayoutView","grid_template_rows":null,"right":null,"justify_content":null,"_view_module":"@jupyter-widgets/base","overflow":null,"_model_module_version":"1.2.0","_view_count":null,"flex_flow":null,"width":null,"min_width":null,"border":null,"align_items":null,"bottom":null,"_model_module":"@jupyter-widgets/base","top":null,"grid_column":null,"overflow_y":null,"overflow_x":null,"grid_auto_flow":null,"grid_area":null,"grid_template_columns":null,"flex":null,"_model_name":"LayoutModel","justify_items":null,"grid_row":null,"max_height":null,"align_content":null,"visibility":null,"align_self":null,"height":null,"min_height":null,"padding":null,"grid_auto_rows":null,"grid_gap":null,"max_width":null,"order":null,"_view_module_version":"1.2.0","grid_template_areas":null,"object_position":null,"object_fit":null,"grid_auto_columns":null,"margin":null,"display":null,"left":null}},"d972c5066342443d85c00d4832938ad3":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","state":{"_view_name":"StyleView","_model_name":"DescriptionStyleModel","description_width":"","_view_module":"@jupyter-widgets/base","_model_module_version":"1.5.0","_view_count":null,"_view_module_version":"1.2.0","_model_module":"@jupyter-widgets/controls"}},"5e3b828e7ba7406b9b32ecc302c8b3c8":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","state":{"_view_name":"LayoutView","grid_template_rows":null,"right":null,"justify_content":null,"_view_module":"@jupyter-widgets/base","overflow":null,"_model_module_version":"1.2.0","_view_count":null,"flex_flow":null,"width":null,"min_width":null,"border":null,"align_items":null,"bottom":null,"_model_module":"@jupyter-widgets/base","top":null,"grid_column":null,"overflow_y":null,"overflow_x":null,"grid_auto_flow":null,"grid_area":null,"grid_template_columns":null,"flex":null,"_model_name":"LayoutModel","justify_items":null,"grid_row":null,"max_height":null,"align_content":null,"visibility":null,"align_self":null,"height":null,"min_height":null,"padding":null,"grid_auto_rows":null,"grid_gap":null,"max_width":null,"order":null,"_view_module_version":"1.2.0","grid_template_areas":null,"object_position":null,"object_fit":null,"grid_auto_columns":null,"margin":null,"display":null,"left":null}},"2ea8243a5213455abffc7cbc3b6ad676":{"model_module":"@jupyter-widgets/controls","model_name":"HBoxModel","state":{"_view_name":"HBoxView","_dom_classes":[],"_model_name":"HBoxModel","_view_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_view_count":null,"_view_module_version":"1.5.0","box_style":"","layout":"IPY_MODEL_c7209ac6ac4e4185bb2d435ae217e4d2","_model_module":"@jupyter-widgets/controls","children":["IPY_MODEL_7d085f86859441b2a3d49b00ebdc6056","IPY_MODEL_b607ffe4713045fbae64a0725129ecb5"]}},"c7209ac6ac4e4185bb2d435ae217e4d2":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","state":{"_view_name":"LayoutView","grid_template_rows":null,"right":null,"justify_content":null,"_view_module":"@jupyter-widgets/base","overflow":null,"_model_module_version":"1.2.0","_view_count":null,"flex_flow":null,"width":null,"min_width":null,"border":null,"align_items":null,"bottom":null,"_model_module":"@jupyter-widgets/base","top":null,"grid_column":null,"overflow_y":null,"overflow_x":null,"grid_auto_flow":null,"grid_area":null,"grid_template_columns":null,"flex":null,"_model_name":"LayoutModel","justify_items":null,"grid_row":null,"max_height":null,"align_content":null,"visibility":null,"align_self":null,"height":null,"min_height":null,"padding":null,"grid_auto_rows":null,"grid_gap":null,"max_width":null,"order":null,"_view_module_version":"1.2.0","grid_template_areas":null,"object_position":null,"object_fit":null,"grid_auto_columns":null,"margin":null,"display":null,"left":null}},"7d085f86859441b2a3d49b00ebdc6056":{"model_module":"@jupyter-widgets/controls","model_name":"FloatProgressModel","state":{"_view_name":"ProgressView","style":"IPY_MODEL_fc357c806c1e4942a6f0c476e8f6e121","_dom_classes":[],"description":"100%","_model_name":"FloatProgressModel","bar_style":"success","max":244408911,"_view_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","value":244408911,"_view_count":null,"_view_module_version":"1.5.0","orientation":"horizontal","min":0,"description_tooltip":null,"_model_module":"@jupyter-widgets/controls","layout":"IPY_MODEL_60142a84850e4188ac2c00555fa2f244"}},"b607ffe4713045fbae64a0725129ecb5":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","state":{"_view_name":"HTMLView","style":"IPY_MODEL_c91fad0d6d2f4791a976b60e55e7d555","_dom_classes":[],"description":"","_model_name":"HTMLModel","placeholder":"​","_view_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","value":" 233M/233M [00:03&lt;00:00, 63.6MB/s]","_view_count":null,"_view_module_version":"1.5.0","description_tooltip":null,"_model_module":"@jupyter-widgets/controls","layout":"IPY_MODEL_1843b25577244547bec2d20c89e2f4ea"}},"fc357c806c1e4942a6f0c476e8f6e121":{"model_module":"@jupyter-widgets/controls","model_name":"ProgressStyleModel","state":{"_view_name":"StyleView","_model_name":"ProgressStyleModel","description_width":"initial","_view_module":"@jupyter-widgets/base","_model_module_version":"1.5.0","_view_count":null,"_view_module_version":"1.2.0","bar_color":null,"_model_module":"@jupyter-widgets/controls"}},"60142a84850e4188ac2c00555fa2f244":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","state":{"_view_name":"LayoutView","grid_template_rows":null,"right":null,"justify_content":null,"_view_module":"@jupyter-widgets/base","overflow":null,"_model_module_version":"1.2.0","_view_count":null,"flex_flow":null,"width":null,"min_width":null,"border":null,"align_items":null,"bottom":null,"_model_module":"@jupyter-widgets/base","top":null,"grid_column":null,"overflow_y":null,"overflow_x":null,"grid_auto_flow":null,"grid_area":null,"grid_template_columns":null,"flex":null,"_model_name":"LayoutModel","justify_items":null,"grid_row":null,"max_height":null,"align_content":null,"visibility":null,"align_self":null,"height":null,"min_height":null,"padding":null,"grid_auto_rows":null,"grid_gap":null,"max_width":null,"order":null,"_view_module_version":"1.2.0","grid_template_areas":null,"object_position":null,"object_fit":null,"grid_auto_columns":null,"margin":null,"display":null,"left":null}},"c91fad0d6d2f4791a976b60e55e7d555":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","state":{"_view_name":"StyleView","_model_name":"DescriptionStyleModel","description_width":"","_view_module":"@jupyter-widgets/base","_model_module_version":"1.5.0","_view_count":null,"_view_module_version":"1.2.0","_model_module":"@jupyter-widgets/controls"}},"1843b25577244547bec2d20c89e2f4ea":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","state":{"_view_name":"LayoutView","grid_template_rows":null,"right":null,"justify_content":null,"_view_module":"@jupyter-widgets/base","overflow":null,"_model_module_version":"1.2.0","_view_count":null,"flex_flow":null,"width":null,"min_width":null,"border":null,"align_items":null,"bottom":null,"_model_module":"@jupyter-widgets/base","top":null,"grid_column":null,"overflow_y":null,"overflow_x":null,"grid_auto_flow":null,"grid_area":null,"grid_template_columns":null,"flex":null,"_model_name":"LayoutModel","justify_items":null,"grid_row":null,"max_height":null,"align_content":null,"visibility":null,"align_self":null,"height":null,"min_height":null,"padding":null,"grid_auto_rows":null,"grid_gap":null,"max_width":null,"order":null,"_view_module_version":"1.2.0","grid_template_areas":null,"object_position":null,"object_fit":null,"grid_auto_columns":null,"margin":null,"display":null,"left":null}},"43a87190275247519df23a8ec12af148":{"model_module":"@jupyter-widgets/controls","model_name":"HBoxModel","state":{"_view_name":"HBoxView","_dom_classes":[],"_model_name":"HBoxModel","_view_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_view_count":null,"_view_module_version":"1.5.0","box_style":"","layout":"IPY_MODEL_002134be836445759ffa27f869c5338a","_model_module":"@jupyter-widgets/controls","children":["IPY_MODEL_c147e7d150ba4b919f6686270df32570","IPY_MODEL_e37954f32d98450da20006ea3d1561f5"]}},"002134be836445759ffa27f869c5338a":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","state":{"_view_name":"LayoutView","grid_template_rows":null,"right":null,"justify_content":null,"_view_module":"@jupyter-widgets/base","overflow":null,"_model_module_version":"1.2.0","_view_count":null,"flex_flow":null,"width":null,"min_width":null,"border":null,"align_items":null,"bottom":null,"_model_module":"@jupyter-widgets/base","top":null,"grid_column":null,"overflow_y":null,"overflow_x":null,"grid_auto_flow":null,"grid_area":null,"grid_template_columns":null,"flex":null,"_model_name":"LayoutModel","justify_items":null,"grid_row":null,"max_height":null,"align_content":null,"visibility":null,"align_self":null,"height":null,"min_height":null,"padding":null,"grid_auto_rows":null,"grid_gap":null,"max_width":null,"order":null,"_view_module_version":"1.2.0","grid_template_areas":null,"object_position":null,"object_fit":null,"grid_auto_columns":null,"margin":null,"display":null,"left":null}},"c147e7d150ba4b919f6686270df32570":{"model_module":"@jupyter-widgets/controls","model_name":"FloatProgressModel","state":{"_view_name":"ProgressView","style":"IPY_MODEL_bf495ab5d5b54edf8f838b012b88369c","_dom_classes":[],"description":"100%","_model_name":"FloatProgressModel","bar_style":"success","max":553433881,"_view_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","value":553433881,"_view_count":null,"_view_module_version":"1.5.0","orientation":"horizontal","min":0,"description_tooltip":null,"_model_module":"@jupyter-widgets/controls","layout":"IPY_MODEL_673234a3aed14d47aa31ce429c70245b"}},"e37954f32d98450da20006ea3d1561f5":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","state":{"_view_name":"HTMLView","style":"IPY_MODEL_fbb3f93f49ae470082b5a904c7abb969","_dom_classes":[],"description":"","_model_name":"HTMLModel","placeholder":"​","_view_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","value":" 528M/528M [01:52&lt;00:00, 4.91MB/s]","_view_count":null,"_view_module_version":"1.5.0","description_tooltip":null,"_model_module":"@jupyter-widgets/controls","layout":"IPY_MODEL_a0040a4028d348d8afd59a29bbd4af30"}},"bf495ab5d5b54edf8f838b012b88369c":{"model_module":"@jupyter-widgets/controls","model_name":"ProgressStyleModel","state":{"_view_name":"StyleView","_model_name":"ProgressStyleModel","description_width":"initial","_view_module":"@jupyter-widgets/base","_model_module_version":"1.5.0","_view_count":null,"_view_module_version":"1.2.0","bar_color":null,"_model_module":"@jupyter-widgets/controls"}},"673234a3aed14d47aa31ce429c70245b":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","state":{"_view_name":"LayoutView","grid_template_rows":null,"right":null,"justify_content":null,"_view_module":"@jupyter-widgets/base","overflow":null,"_model_module_version":"1.2.0","_view_count":null,"flex_flow":null,"width":null,"min_width":null,"border":null,"align_items":null,"bottom":null,"_model_module":"@jupyter-widgets/base","top":null,"grid_column":null,"overflow_y":null,"overflow_x":null,"grid_auto_flow":null,"grid_area":null,"grid_template_columns":null,"flex":null,"_model_name":"LayoutModel","justify_items":null,"grid_row":null,"max_height":null,"align_content":null,"visibility":null,"align_self":null,"height":null,"min_height":null,"padding":null,"grid_auto_rows":null,"grid_gap":null,"max_width":null,"order":null,"_view_module_version":"1.2.0","grid_template_areas":null,"object_position":null,"object_fit":null,"grid_auto_columns":null,"margin":null,"display":null,"left":null}},"fbb3f93f49ae470082b5a904c7abb969":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","state":{"_view_name":"StyleView","_model_name":"DescriptionStyleModel","description_width":"","_view_module":"@jupyter-widgets/base","_model_module_version":"1.5.0","_view_count":null,"_view_module_version":"1.2.0","_model_module":"@jupyter-widgets/controls"}},"a0040a4028d348d8afd59a29bbd4af30":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","state":{"_view_name":"LayoutView","grid_template_rows":null,"right":null,"justify_content":null,"_view_module":"@jupyter-widgets/base","overflow":null,"_model_module_version":"1.2.0","_view_count":null,"flex_flow":null,"width":null,"min_width":null,"border":null,"align_items":null,"bottom":null,"_model_module":"@jupyter-widgets/base","top":null,"grid_column":null,"overflow_y":null,"overflow_x":null,"grid_auto_flow":null,"grid_area":null,"grid_template_columns":null,"flex":null,"_model_name":"LayoutModel","justify_items":null,"grid_row":null,"max_height":null,"align_content":null,"visibility":null,"align_self":null,"height":null,"min_height":null,"padding":null,"grid_auto_rows":null,"grid_gap":null,"max_width":null,"order":null,"_view_module_version":"1.2.0","grid_template_areas":null,"object_position":null,"object_fit":null,"grid_auto_columns":null,"margin":null,"display":null,"left":null}}}}},"cells":[{"cell_type":"markdown","metadata":{"id":"PkRJW7QiswEw"},"source":["# Labeling Images That Have Weapon/Knife Content\n"," *by Yusuf GÜR (2532810)*\n"]},{"cell_type":"markdown","metadata":{"id":"ewM7CGodswEy"},"source":["## Introduction\n","\n","This notebook implements a fight labelling in crowd areas, especially in stadiums.\n"]},{"cell_type":"markdown","metadata":{"id":"AH7hEqiJzjrO"},"source":["## Notes\n","\n","The dataset is located in the link below.\n","https://we.tl/t-yHftCY3Cg3\n","\n","It can be downloaded, unzipped and used for training\n","\n","The trained weights are also located in this link:\n","https://we.tl/t-94m9JBxB5q"]},{"cell_type":"markdown","metadata":{"id":"ciGvlFGzswEz"},"source":["## Imports"]},{"cell_type":"code","metadata":{"id":"OzHYeJ9wswE0","executionInfo":{"status":"ok","timestamp":1625000380772,"user_tz":-180,"elapsed":5415,"user":{"displayName":"Yusuf GÜR","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GifMwXdY1_KGHe-vPWD-1Noese7ceIkU8eKmlY-g7Q=s64","userId":"13124731809299192028"}}},"source":["# required libraries for the model\n","from numpy.random import seed, shuffle\n","import numpy as np\n","import matplotlib.pyplot as plt\n","import time as tm \n","import sys\n","import logging\n","import cv2\n","import time as tm\n","%matplotlib inline\n","\n","# Imports\n","import torch\n","import torch.nn as nn  # All neural network modules, nn.Linear, nn.Conv2d, BatchNorm, Loss functions\n","import torch.optim as optim  # For all Optimization algorithms, SGD, Adam, etc.\n","from torch.optim import lr_scheduler  #\n","import torch.nn.functional as F  # All functions that don't have any parameters\n","import torchvision.transforms as transforms  # Transformations we can perform on our dataset\n","from torchvision import datasets, models, transforms\n","from torch.utils.data import (DataLoader,DataLoader)  # Gives easier dataset managment and creates mini batches\n","import torchvision.transforms as transforms  # Transformations we can perform on our dataset\n","import torchvision\n","import os\n","import pandas as pd\n","from skimage import io\n","\n","from google.colab import drive\n","\n","import shutil\n","import copy"],"execution_count":1,"outputs":[]},{"cell_type":"markdown","metadata":{"id":"9Ib5vqiCswE3"},"source":["**Setting**  Device and **Initiating** Google_Drive"]},{"cell_type":"code","metadata":{"id":"PK7t9VhVswE3","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1625000485317,"user_tz":-180,"elapsed":98646,"user":{"displayName":"Yusuf GÜR","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GifMwXdY1_KGHe-vPWD-1Noese7ceIkU8eKmlY-g7Q=s64","userId":"13124731809299192028"}},"outputId":"188cd1ee-e63a-433f-c586-2c8d841b0bf2"},"source":["# Set device\n","device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n","# Connect Drive \n","drive.mount(\"/content/gdrive\")\n","# I am using my own Google Drive. All files are in labIssues folder\n","dataset_path = '/content/gdrive/MyDrive/METU/01.courses/METU_MMI_727/project/source'"],"execution_count":2,"outputs":[{"output_type":"stream","text":["Mounted at /content/gdrive\n"],"name":"stdout"}]},{"cell_type":"markdown","metadata":{"id":"vs5AA0AgvqI-"},"source":["## Prepare Datasets"]},{"cell_type":"code","metadata":{"id":"7_9YOD3xZuc0","executionInfo":{"status":"ok","timestamp":1625000487595,"user_tz":-180,"elapsed":274,"user":{"displayName":"Yusuf GÜR","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GifMwXdY1_KGHe-vPWD-1Noese7ceIkU8eKmlY-g7Q=s64","userId":"13124731809299192028"}}},"source":["#!rm -r /content/gdrive/MyDrive/METU/01.courses/METU_MMI_727/project/source_new\n","#!rm -r /content/source_new\n","#%cd /content/gdrive/MyDrive/METU/01.courses/METU_MMI_727/project/source\n","#%cp -av directorySource directoryTarget  # using linux terminal\n","#!rm -r /content/source\n","#!rm -r /content/test\n","#!rm -r /content/train\n","#!rm -r /content/val"],"execution_count":3,"outputs":[]},{"cell_type":"markdown","metadata":{"id":"0n_5dUsDGL17"},"source":["# Functions to be used for preparing Datasets"]},{"cell_type":"code","metadata":{"id":"Mc2cW30bhbfd","executionInfo":{"status":"ok","timestamp":1625000491756,"user_tz":-180,"elapsed":311,"user":{"displayName":"Yusuf GÜR","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GifMwXdY1_KGHe-vPWD-1Noese7ceIkU8eKmlY-g7Q=s64","userId":"13124731809299192028"}}},"source":["# Checking source and destination file sizes if some files missed\n","def countFiles(directorySource,directoryDestination):\n","  fileSrc = os.listdir(directorySource) # fileSrc is source directory path\n","  fileDst = os.listdir(directoryDestination) # fileDst is destination directory path\n","  print(\"Source Folder Length:\",len(fileSrc),\"Destination Folder Length:\",len(fileDst))"],"execution_count":4,"outputs":[]},{"cell_type":"code","metadata":{"id":"-ukfL3-BGy69","executionInfo":{"status":"ok","timestamp":1625000492973,"user_tz":-180,"elapsed":2,"user":{"displayName":"Yusuf GÜR","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GifMwXdY1_KGHe-vPWD-1Noese7ceIkU8eKmlY-g7Q=s64","userId":"13124731809299192028"}}},"source":["# Using linux terminal to transfer files from one path to another\n","def linuxMove(src,dst,Src,Dst):\n","  directorySrc=src+Src\n","  directoryDst=dst+Dst\n","  %mv -b $directorySrc $directoryDst  # using linux terminal\n","  countFiles(directorySrc,directoryDst)"],"execution_count":5,"outputs":[]},{"cell_type":"code","metadata":{"id":"clUkry2DzW7N","executionInfo":{"status":"ok","timestamp":1625000495139,"user_tz":-180,"elapsed":3,"user":{"displayName":"Yusuf GÜR","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GifMwXdY1_KGHe-vPWD-1Noese7ceIkU8eKmlY-g7Q=s64","userId":"13124731809299192028"}}},"source":["# Using linux terminal to transfer files from one path to another\n","def linuxCopy(src,dst,Src,Dst):\n","  directorySrc=src+Src\n","  directoryDst=dst+Dst\n","  if not os.path.exists(directoryDst):\n","    os.makedirs(directoryDst)\n","  %cp -ar $directorySrc/* $directoryDst  # using linux terminal\n","  countFiles(directorySrc,directoryDst)"],"execution_count":6,"outputs":[]},{"cell_type":"code","metadata":{"id":"0Ij1Ix8Ma13R","executionInfo":{"status":"ok","timestamp":1625000496943,"user_tz":-180,"elapsed":356,"user":{"displayName":"Yusuf GÜR","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GifMwXdY1_KGHe-vPWD-1Noese7ceIkU8eKmlY-g7Q=s64","userId":"13124731809299192028"}}},"source":["# preparing data paths to copy from gdrive\n","src=\"/content/gdrive/MyDrive/METU/01.courses/METU_MMI_727/project/source\"\n","dst=\"/content/source\"\n","#dst=\"/content/gdrive/MyDrive/METU/01.courses/METU_MMI_727/project\"\n","\n","knifeImgSrc='/Knife_Detection/Images'\n","knifeImgDst='/images/Knifes'\n","knifeAntSrc='/Knife_Detection/Annotations'\n","knifeAntDst='/annotations/Knifes_annotations'\n","weaponImgSrc='/Weapon_Detection/Images'\n","weaponImgDst='/images/Weapons'\n","weaponAntSrc='/Weapon_Detection/Annotations'\n","weaponAntDst='/annotations/Weapons_annotations'\n","\n","if not os.path.exists(dst):\n","  os.makedirs(dst)\n","\n","Src=[knifeImgSrc,knifeAntSrc,weaponImgSrc,weaponAntSrc]\n","Dst=[knifeImgDst,knifeAntDst,weaponImgDst,weaponAntDst]"],"execution_count":7,"outputs":[]},{"cell_type":"code","metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"MKpVeUfzzrtS","executionInfo":{"status":"ok","timestamp":1624912497740,"user_tz":-180,"elapsed":3673603,"user":{"displayName":"Yusuf GÜR","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GifMwXdY1_KGHe-vPWD-1Noese7ceIkU8eKmlY-g7Q=s64","userId":"13124731809299192028"}},"outputId":"8102aa2d-4123-4bde-ef9c-b6684a505076"},"source":["'''\n","# copy folders into colab\n","start_time = tm.time() \n","for i in range(len(Src)):\n","  linuxCopy(src,dst,Src[i],Dst[i])\n","end_time = tm.time()                                                                               \n","duration_time = np.subtract(end_time, start_time)\n","print('Duration of Copy {:.0f}m {:.0f}s'.format(\n","        duration_time // 60, duration_time % 60))"],"execution_count":null,"outputs":[{"output_type":"stream","text":["Source Folder Length: 2078 Destination Folder Length: 2078\n","Source Folder Length: 2078 Destination Folder Length: 2078\n","Source Folder Length: 3000 Destination Folder Length: 3000\n","Source Folder Length: 3000 Destination Folder Length: 3000\n","Duration of Copy 61m 13s\n"],"name":"stdout"}]},{"cell_type":"markdown","metadata":{"id":"xVeQnH-I6cPa"},"source":["# Splitting dataset into train, test and validation"]},{"cell_type":"code","metadata":{"id":"CuqYOaKG6c9d","executionInfo":{"status":"ok","timestamp":1625000513177,"user_tz":-180,"elapsed":288,"user":{"displayName":"Yusuf GÜR","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GifMwXdY1_KGHe-vPWD-1Noese7ceIkU8eKmlY-g7Q=s64","userId":"13124731809299192028"}}},"source":["# Defining the function to be used\n","def split_dataset(path_to_dataset, train_ratio, valid_ratio):\n","    #split the dataset in the given path into three subsets(test,val,train)\n","    _, sub_dirs, _ = next(iter(os.walk(path_to_dataset)))  # retrieve name of subdirectories\n","    sub_dir_item_cnt = [0 for i in range(len(sub_dirs))]  # list for counting items in each sub directory(class)\n","\n","    # directories where the splitted dataset will copy\n","    dir_train = os.path.join(os.path.dirname(path_to_dataset), 'train')\n","    dir_test = os.path.join(os.path.dirname(path_to_dataset), 'test')\n","    dir_valid = os.path.join(os.path.dirname(path_to_dataset), 'val')\n","    \n","\n","    for i, sub_dir in enumerate(sub_dirs):  # walking on all sub directories\n","\n","        dir_train_dst = os.path.join(dir_train, sub_dir)  # directory for destination of train dataset\n","        dir_test_dst = os.path.join(dir_test, sub_dir)  # directory for destination of test dataset\n","        dir_valid_dst = os.path.join(dir_valid, sub_dir)  # directory for destination of validation dataset\n","        \n","        # variables to save the sub directory name(class name) and to count the images of each sub directory(class)\n","        class_name = sub_dir\n","        sub_dir = os.path.join(path_to_dataset, sub_dir)\n","        sub_dir_item_cnt[i] = len(os.listdir(sub_dir))\n","\n","        items = os.listdir(sub_dir)\n","\n","        # transfer data to trainset\n","        for item_idx in range(round(sub_dir_item_cnt[i] * train_ratio)):\n","            if not os.path.exists(dir_train_dst):\n","                os.makedirs(dir_train_dst)\n","\n","            source_file = os.path.join(sub_dir, items[item_idx])\n","            dst_file = os.path.join(dir_train_dst, items[item_idx])\n","            shutil.copyfile(source_file, dst_file)\n","\n","        # transfer data to testset\n","        for item_idx in range(round(sub_dir_item_cnt[i] * (train_ratio + valid_ratio)) + 1, sub_dir_item_cnt[i]):\n","            if not os.path.exists(dir_test_dst):\n","                os.makedirs(dir_test_dst)\n","\n","            source_file = os.path.join(sub_dir, items[item_idx])\n","            dst_file = os.path.join(dir_test_dst, items[item_idx])\n","            shutil.copyfile(source_file, dst_file)\n","\n","        # transfer data to validation\n","        for item_idx in range(round(sub_dir_item_cnt[i] * train_ratio) + 1,\n","                              round(sub_dir_item_cnt[i] * (train_ratio + valid_ratio))):\n","            if not os.path.exists(dir_valid_dst):\n","                os.makedirs(dir_valid_dst)\n","\n","            source_file = os.path.join(sub_dir, items[item_idx])\n","            dst_file = os.path.join(dir_valid_dst, items[item_idx])\n","            shutil.copyfile(source_file, dst_file)\n","    return\n","\n","\n","\n"],"execution_count":8,"outputs":[]},{"cell_type":"code","metadata":{"id":"nk8tSr0iNdQ-"},"source":["'''\n","#Splitting data into 3 sets\n","path_to_dataset=\"/content/source/images\"\n","train_ratio=0.7\n","valid_ratio=0.15\n","split_dataset(path_to_dataset, train_ratio, valid_ratio)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"kVa3Fb3AwMRF","executionInfo":{"status":"ok","timestamp":1625000618842,"user_tz":-180,"elapsed":35831,"user":{"displayName":"Yusuf GÜR","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GifMwXdY1_KGHe-vPWD-1Noese7ceIkU8eKmlY-g7Q=s64","userId":"13124731809299192028"}},"outputId":"ef4f70d9-9a0a-45cc-ae57-8550748ece8a"},"source":["#Checking data sizes in each 3 sets\n","dstTrain=\"/content/gdrive/MyDrive/METU/01.courses/METU_MMI_727/project/train\"\n","dstTest=\"/content/gdrive/MyDrive/METU/01.courses/METU_MMI_727/project/test\"\n","dstValidation=\"/content/gdrive/MyDrive/METU/01.courses/METU_MMI_727/project/val\"\n","\n","knifeDst='/Knifes'\n","weaponDst='/Weapons'\n","\n","Folders=[dstTrain,dstTest,dstValidation]\n","Dst=[knifeDst,weaponDst]\n","\n","for i in range(len(Folders)):\n","  for j in range(len(Dst)):\n","    directoryDst=Folders[i]+Dst[j]\n","    fileDst = os.listdir(directoryDst) # fileDst is directory path\n","    print(directoryDst,\"Folder Length:\",len(fileDst))\n"],"execution_count":9,"outputs":[{"output_type":"stream","text":["/content/gdrive/MyDrive/METU/01.courses/METU_MMI_727/project/train/Knifes Folder Length: 1455\n","/content/gdrive/MyDrive/METU/01.courses/METU_MMI_727/project/train/Weapons Folder Length: 2100\n","/content/gdrive/MyDrive/METU/01.courses/METU_MMI_727/project/test/Knifes Folder Length: 311\n","/content/gdrive/MyDrive/METU/01.courses/METU_MMI_727/project/test/Weapons Folder Length: 449\n","/content/gdrive/MyDrive/METU/01.courses/METU_MMI_727/project/val/Knifes Folder Length: 310\n","/content/gdrive/MyDrive/METU/01.courses/METU_MMI_727/project/val/Weapons Folder Length: 449\n"],"name":"stdout"}]},{"cell_type":"markdown","metadata":{"id":"3sM_z44Xxem-"},"source":["# Define Transfoms Functions"]},{"cell_type":"code","metadata":{"id":"56eNndudxdHx","executionInfo":{"status":"ok","timestamp":1625000623056,"user_tz":-180,"elapsed":290,"user":{"displayName":"Yusuf GÜR","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GifMwXdY1_KGHe-vPWD-1Noese7ceIkU8eKmlY-g7Q=s64","userId":"13124731809299192028"}}},"source":["#------------- (Dataset Transforms)---------------------------------#\n","# 1) Resize image to H x W\n","# 2) Convert image to torch tensor.\n","# 3) Normalize the tensor.\n","#-------------------------------------------------------------------#\n","\n","mean = np.array([0.485, 0.456, 0.406])\n","std = np.array([0.229, 0.224, 0.225])\n","\n","data_transforms = {\n","    'train': transforms.Compose([\n","        transforms.RandomResizedCrop(224),\n","        transforms.RandomHorizontalFlip(),\n","        transforms.ToTensor(),\n","        transforms.Normalize(mean, std)\n","    ]),\n","    'test': transforms.Compose([\n","        transforms.Resize(256),\n","        transforms.CenterCrop(224),\n","        transforms.ToTensor(),\n","        transforms.Normalize(mean, std)\n","    ]),\n","    'val': transforms.Compose([\n","        transforms.Resize(256),\n","        transforms.CenterCrop(224),\n","        transforms.ToTensor(),\n","        transforms.Normalize(mean, std)\n","    ]),\n","}\n","\n","#Imshow for Tensor.\n","def imshow(img, title):\n","    \"\"\"Imshow for Tensor.\"\"\"\n","    img = img.numpy().transpose((1, 2, 0))\n","    img = std * img + mean\n","    img = np.clip(img, 0, 1)\n","    plt.imshow(img)\n","    plt.title(title)\n","    plt.show()\n","\n"],"execution_count":10,"outputs":[]},{"cell_type":"markdown","metadata":{"id":"3E2cTCVmSMKp"},"source":["## Define Hyperparamaters to be used"]},{"cell_type":"code","metadata":{"colab":{"base_uri":"https://localhost:8080/","height":1000,"referenced_widgets":["d784668f947048c9b273b974d7de63e3","1687ef5de81f492284e17b0937a509b0","5d8b4627dea24864b2619be7022a9505","c33761d8df394b32aa90e27c444042fd","bf57c810dbc7424e93b7290ac71e6f14","ce131cdc0a774dac92f2481f2323aea7","d972c5066342443d85c00d4832938ad3","5e3b828e7ba7406b9b32ecc302c8b3c8","2ea8243a5213455abffc7cbc3b6ad676","c7209ac6ac4e4185bb2d435ae217e4d2","7d085f86859441b2a3d49b00ebdc6056","b607ffe4713045fbae64a0725129ecb5","fc357c806c1e4942a6f0c476e8f6e121","60142a84850e4188ac2c00555fa2f244","c91fad0d6d2f4791a976b60e55e7d555","1843b25577244547bec2d20c89e2f4ea","43a87190275247519df23a8ec12af148","002134be836445759ffa27f869c5338a","c147e7d150ba4b919f6686270df32570","e37954f32d98450da20006ea3d1561f5","bf495ab5d5b54edf8f838b012b88369c","673234a3aed14d47aa31ce429c70245b","fbb3f93f49ae470082b5a904c7abb969","a0040a4028d348d8afd59a29bbd4af30"]},"id":"EAC-p3dPncii","executionInfo":{"status":"ok","timestamp":1625000639874,"user_tz":-180,"elapsed":10699,"user":{"displayName":"Yusuf GÜR","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GifMwXdY1_KGHe-vPWD-1Noese7ceIkU8eKmlY-g7Q=s64","userId":"13124731809299192028"}},"outputId":"7ab302f6-786d-48f3-83b2-a74c38cee195"},"source":["## Models for Transfer Learning\n","#resnet18 = models.resnet18(pretrained=True)\n","#print(resnet18)\n","alexnet = models.alexnet(pretrained=True)\n","print(alexnet)\n","vgg16 = models.vgg16(pretrained=True)\n","print(vgg16)"],"execution_count":11,"outputs":[{"output_type":"stream","text":["Downloading: \"https://download.pytorch.org/models/resnet18-f37072fd.pth\" to /root/.cache/torch/hub/checkpoints/resnet18-f37072fd.pth\n"],"name":"stderr"},{"output_type":"display_data","data":{"application/vnd.jupyter.widget-view+json":{"model_id":"d784668f947048c9b273b974d7de63e3","version_minor":0,"version_major":2},"text/plain":["HBox(children=(FloatProgress(value=0.0, max=46830571.0), HTML(value='')))"]},"metadata":{"tags":[]}},{"output_type":"stream","text":["\n","ResNet(\n","  (conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)\n","  (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","  (relu): ReLU(inplace=True)\n","  (maxpool): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)\n","  (layer1): Sequential(\n","    (0): BasicBlock(\n","      (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (relu): ReLU(inplace=True)\n","      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","    )\n","    (1): BasicBlock(\n","      (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (relu): ReLU(inplace=True)\n","      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","    )\n","  )\n","  (layer2): Sequential(\n","    (0): BasicBlock(\n","      (conv1): Conv2d(64, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n","      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (relu): ReLU(inplace=True)\n","      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (downsample): Sequential(\n","        (0): Conv2d(64, 128, kernel_size=(1, 1), stride=(2, 2), bias=False)\n","        (1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      )\n","    )\n","    (1): BasicBlock(\n","      (conv1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (relu): ReLU(inplace=True)\n","      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","    )\n","  )\n","  (layer3): Sequential(\n","    (0): BasicBlock(\n","      (conv1): Conv2d(128, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n","      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (relu): ReLU(inplace=True)\n","      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (downsample): Sequential(\n","        (0): Conv2d(128, 256, kernel_size=(1, 1), stride=(2, 2), bias=False)\n","        (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      )\n","    )\n","    (1): BasicBlock(\n","      (conv1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (relu): ReLU(inplace=True)\n","      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","    )\n","  )\n","  (layer4): Sequential(\n","    (0): BasicBlock(\n","      (conv1): Conv2d(256, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n","      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (relu): ReLU(inplace=True)\n","      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (downsample): Sequential(\n","        (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n","        (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      )\n","    )\n","    (1): BasicBlock(\n","      (conv1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (relu): ReLU(inplace=True)\n","      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","    )\n","  )\n","  (avgpool): AdaptiveAvgPool2d(output_size=(1, 1))\n","  (fc): Linear(in_features=512, out_features=1000, bias=True)\n",")\n"],"name":"stdout"},{"output_type":"stream","text":["Downloading: \"https://download.pytorch.org/models/alexnet-owt-7be5be79.pth\" to /root/.cache/torch/hub/checkpoints/alexnet-owt-7be5be79.pth\n"],"name":"stderr"},{"output_type":"display_data","data":{"application/vnd.jupyter.widget-view+json":{"model_id":"2ea8243a5213455abffc7cbc3b6ad676","version_minor":0,"version_major":2},"text/plain":["HBox(children=(FloatProgress(value=0.0, max=244408911.0), HTML(value='')))"]},"metadata":{"tags":[]}},{"output_type":"stream","text":["\n","AlexNet(\n","  (features): Sequential(\n","    (0): Conv2d(3, 64, kernel_size=(11, 11), stride=(4, 4), padding=(2, 2))\n","    (1): ReLU(inplace=True)\n","    (2): MaxPool2d(kernel_size=3, stride=2, padding=0, dilation=1, ceil_mode=False)\n","    (3): Conv2d(64, 192, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))\n","    (4): ReLU(inplace=True)\n","    (5): MaxPool2d(kernel_size=3, stride=2, padding=0, dilation=1, ceil_mode=False)\n","    (6): Conv2d(192, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n","    (7): ReLU(inplace=True)\n","    (8): Conv2d(384, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n","    (9): ReLU(inplace=True)\n","    (10): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n","    (11): ReLU(inplace=True)\n","    (12): MaxPool2d(kernel_size=3, stride=2, padding=0, dilation=1, ceil_mode=False)\n","  )\n","  (avgpool): AdaptiveAvgPool2d(output_size=(6, 6))\n","  (classifier): Sequential(\n","    (0): Dropout(p=0.5, inplace=False)\n","    (1): Linear(in_features=9216, out_features=4096, bias=True)\n","    (2): ReLU(inplace=True)\n","    (3): Dropout(p=0.5, inplace=False)\n","    (4): Linear(in_features=4096, out_features=4096, bias=True)\n","    (5): ReLU(inplace=True)\n","    (6): Linear(in_features=4096, out_features=1000, bias=True)\n","  )\n",")\n"],"name":"stdout"},{"output_type":"stream","text":["Downloading: \"https://download.pytorch.org/models/vgg16-397923af.pth\" to /root/.cache/torch/hub/checkpoints/vgg16-397923af.pth\n"],"name":"stderr"},{"output_type":"display_data","data":{"application/vnd.jupyter.widget-view+json":{"model_id":"43a87190275247519df23a8ec12af148","version_minor":0,"version_major":2},"text/plain":["HBox(children=(FloatProgress(value=0.0, max=553433881.0), HTML(value='')))"]},"metadata":{"tags":[]}},{"output_type":"stream","text":["\n","VGG(\n","  (features): Sequential(\n","    (0): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n","    (1): ReLU(inplace=True)\n","    (2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n","    (3): ReLU(inplace=True)\n","    (4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n","    (5): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n","    (6): ReLU(inplace=True)\n","    (7): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n","    (8): ReLU(inplace=True)\n","    (9): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n","    (10): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n","    (11): ReLU(inplace=True)\n","    (12): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n","    (13): ReLU(inplace=True)\n","    (14): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n","    (15): ReLU(inplace=True)\n","    (16): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n","    (17): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n","    (18): ReLU(inplace=True)\n","    (19): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n","    (20): ReLU(inplace=True)\n","    (21): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n","    (22): ReLU(inplace=True)\n","    (23): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n","    (24): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n","    (25): ReLU(inplace=True)\n","    (26): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n","    (27): ReLU(inplace=True)\n","    (28): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n","    (29): ReLU(inplace=True)\n","    (30): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n","  )\n","  (avgpool): AdaptiveAvgPool2d(output_size=(7, 7))\n","  (classifier): Sequential(\n","    (0): Linear(in_features=25088, out_features=4096, bias=True)\n","    (1): ReLU(inplace=True)\n","    (2): Dropout(p=0.5, inplace=False)\n","    (3): Linear(in_features=4096, out_features=4096, bias=True)\n","    (4): ReLU(inplace=True)\n","    (5): Dropout(p=0.5, inplace=False)\n","    (6): Linear(in_features=4096, out_features=1000, bias=True)\n","  )\n",")\n"],"name":"stdout"}]},{"cell_type":"code","metadata":{"id":"Az67VAI3SW2o","executionInfo":{"status":"ok","timestamp":1625000666025,"user_tz":-180,"elapsed":261,"user":{"displayName":"Yusuf GÜR","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GifMwXdY1_KGHe-vPWD-1Noese7ceIkU8eKmlY-g7Q=s64","userId":"13124731809299192028"}}},"source":["## Transfer Learning aproaches\n","transferLearnings=['fineTune','fromScratch']\n","## Transfer Learning aproaches\n","#architectures=[resnet18, alexnet, vgg16]\n","architectures=[alexnet, vgg16]\n","## Transfer Learning aproaches\n","#dropouts=[0.0,0.1,0.2]\n","dropouts=[0.0,0.2]\n","## Transfer Learning aproaches\n","#learningRates=[1e-4 , 1e-3, 1e-2]\n","learningRates=[1e-4 , 1e-3,]\n","## Learning Schedulers\n","#learningSchedulerSteps=[2,5,10]\n","learningSchedulerSteps=[2,5]\n","## Epochs used with Learning Schedulers according to order\n","#epochs=[10,20,30]\n","epochs=[10,20]\n","## Batch Sizes\n","#batchSizes=[4,16,64]\n","batchSizes=[16,64]\n","## Loss Functions\n","criterion=nn.CrossEntropyLoss()\n","## Optimizer\n","optimizer=optim.SGD\n"],"execution_count":12,"outputs":[]},{"cell_type":"code","metadata":{"id":"c_V3oPysh7VV","colab":{"base_uri":"https://localhost:8080/","height":120},"executionInfo":{"status":"ok","timestamp":1624913044309,"user_tz":-180,"elapsed":418,"user":{"displayName":"Yusuf GÜR","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GifMwXdY1_KGHe-vPWD-1Noese7ceIkU8eKmlY-g7Q=s64","userId":"13124731809299192028"}},"outputId":"7b4954dc-1c5a-498d-a5af-24b5f7fe0cf4"},"source":["'''\n","# classifiers for each model to be used in transfer learning\n","\n","resnetClassifier=nn.Sequential(\n","                        nn.Linear(512,10),\n","                        nn.ReLU(),\n","                        nn.BatchNorm1d(10),\n","                        nn.Dropout(dropOut),\n","                        nn.Linear(10,2),\n",")\n","\n","alexnetClassifier= nn.Sequential(\n","\t\t\t\t                nn.Dropout(dropOut),\n","                        nn.Linear(9216, 4096),\n","                        nn.ReLU(),\n","                        nn.Dropout(dropOut),\n","                        nn.Linear(4096, 1000),\n","                        nn.ReLU(),\n","                        nn.Dropout(dropOut),\n","                        nn.Linear(1000,10),\n","                        nn.ReLU(),\n","                        nn.Dropout(dropOut),\n","                        nn.Linear(10,2),\n",")\n","\n","vgg16netClassifier= nn.Sequential(\n","\t\t\t\t                nn.Dropout(dropOut),\n","                        nn.Linear(25088, 4096),\n","                        nn.ReLU(),\n","                        nn.Dropout(dropOut),\n","                        nn.Linear(4096, 1000),\n","                        nn.ReLU(),\n","                        nn.Dropout(dropOut),\n","                        nn.Linear(1000,10),\n","                        nn.ReLU(),\n","                        nn.Dropout(dropOut),\n","                        nn.Linear(10,2),\n",")\n","'''"],"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"application/vnd.google.colaboratory.intrinsic+json":{"type":"string"},"text/plain":["'\\n# classifiers for each model to be used in transfer learning\\n\\nresnetClassifier=nn.Sequential(\\n                        nn.Linear(512,10),\\n                        nn.ReLU(),\\n                        nn.BatchNorm1d(10),\\n                        nn.Dropout(dropOut),\\n                        nn.Linear(10,2),\\n)\\n\\nalexnetClassifier= nn.Sequential(\\n\\t\\t\\t\\t                nn.Dropout(dropOut),\\n                        nn.Linear(9216, 4096),\\n                        nn.ReLU(),\\n                        nn.Dropout(dropOut),\\n                        nn.Linear(4096, 1000),\\n                        nn.ReLU(),\\n                        nn.Dropout(dropOut),\\n                        nn.Linear(1000,10),\\n                        nn.ReLU(),\\n                        nn.Dropout(dropOut),\\n                        nn.Linear(10,2),\\n)\\n\\nvgg16netClassifier= nn.Sequential(\\n\\t\\t\\t\\t                nn.Dropout(dropOut),\\n                        nn.Linear(25088, 4096),\\n                        nn.ReLU(),\\n                        nn.Dropout(dropOut),\\n                        nn.Linear(4096, 1000),\\n                        nn.ReLU(),\\n                        nn.Dropout(dropOut),\\n                        nn.Linear(1000,10),\\n                        nn.ReLU(),\\n                        nn.Dropout(dropOut),\\n                        nn.Linear(10,2),\\n)\\n'"]},"metadata":{"tags":[]},"execution_count":22}]},{"cell_type":"markdown","metadata":{"id":"ynzcHbvdswE5"},"source":["## Load Data\n","\n","The data is imported to Colab from Google Drive "]},{"cell_type":"code","metadata":{"colab":{"base_uri":"https://localhost:8080/","height":182},"id":"eGieKOgvUEZK","executionInfo":{"status":"ok","timestamp":1625000750537,"user_tz":-180,"elapsed":2210,"user":{"displayName":"Yusuf GÜR","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GifMwXdY1_KGHe-vPWD-1Noese7ceIkU8eKmlY-g7Q=s64","userId":"13124731809299192028"}},"outputId":"6b1d5b06-efb3-4c5d-ab35-fa144ab1a26c"},"source":["data_dir = '/content/gdrive/MyDrive/METU/01.courses/METU_MMI_727/project/'\n","sets = ['train', 'test', 'val']\n","\n","# defining datasets for each set\n","image_datasets = {x: datasets.ImageFolder(os.path.join(data_dir, x),\n","                                          data_transforms[x])\n","                  for x in sets}\n","# defining dataloader for each set\n","dataloaders = {x: torch.utils.data.DataLoader(image_datasets[x], batch_size=4,\n","                                             shuffle=True, num_workers=0)\n","              for x in sets}\n","\n","# defining dataset sizes for each set\n","dataset_sizes = {x: len(image_datasets[x]) for x in sets}\n","\n","# Target class names according to folder names\n","class_names = image_datasets['train'].classes\n","\n","print(\"Class Names:\",class_names)\n","print(\"Dataset sizes:\",dataset_sizes)\n","\n","# Get a batch of training data\n","inputs, classes = next(iter(dataloaders['train']))\n","\n","# Make a grid from batch\n","out = torchvision.utils.make_grid(inputs)\n","\n","imshow(out, title=[class_names[x] for x in classes])"],"execution_count":13,"outputs":[{"output_type":"stream","text":["Class Names: ['Knifes', 'Weapons']\n","Dataset sizes: {'train': 3555, 'test': 760, 'val': 759}\n"],"name":"stdout"},{"output_type":"display_data","data":{"image/png":"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\n","text/plain":["<Figure size 432x288 with 1 Axes>"]},"metadata":{"tags":[],"needs_background":"light"}}]},{"cell_type":"markdown","metadata":{"id":"wn-DBSgucDdd"},"source":["# training  Model"]},{"cell_type":"code","metadata":{"id":"4_kivCVacETf","executionInfo":{"status":"ok","timestamp":1625000770197,"user_tz":-180,"elapsed":607,"user":{"displayName":"Yusuf GÜR","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GifMwXdY1_KGHe-vPWD-1Noese7ceIkU8eKmlY-g7Q=s64","userId":"13124731809299192028"}}},"source":["def train_model(model, criterion, optimizer, scheduler, num_epochs=5):\n","    #Start Timer\n","    start_time = tm.time()\n","\n","    trainLoss=[]\n","    trainAccuracy=[]\n","    testLoss=[]\n","    testAccuracy=[]\n","\n","    best_model_wts = copy.deepcopy(model.state_dict())\n","    best_acc = 0.0\n","\n","    for epoch in range(num_epochs):\n","        print('Epoch {}/{}'.format(epoch+1, num_epochs))\n","        print('-' * 10)\n","\n","        # Each epoch has a training and test phase\n","        for phase in ['train', 'test']:\n","            if phase == 'train':\n","                model.train()  # Set model to training mode\n","            else:\n","                model.eval()   # Set model to evaluate mode\n","\n","            running_loss = 0.0\n","            running_corrects = 0\n","\n","            # Iterate over data.\n","            for inputs, labels in dataloaders[phase]:\n","                inputs = inputs.to(device)\n","                labels = labels.to(device)\n","\n","                # forward\n","                # track history if only in train\n","                with torch.set_grad_enabled(phase == 'train'):\n","                    outputs = model(inputs)\n","                    _, preds = torch.max(outputs, 1)\n","                    loss = criterion(outputs, labels)\n","\n","                    # backward + optimize only if in training phase\n","                    if phase == 'train':\n","                        optimizer.zero_grad()\n","                        loss.backward()\n","                        optimizer.step()\n","\n","                # statistics\n","                running_loss += loss.item() * inputs.size(0)\n","                running_corrects += torch.sum(preds == labels.data)\n","\n","            if phase == 'train':\n","                scheduler.step()\n","\n","            epoch_loss = running_loss / dataset_sizes[phase]\n","            epoch_acc = running_corrects.double() / dataset_sizes[phase]\n","            if phase == 'train':\n","                trainLoss.append(epoch_loss) \n","                trainAccuracy.append(epoch_acc)\n","            else:\n","                testLoss.append(epoch_loss) \n","                testAccuracy.append(epoch_acc)\n","\n","            #print to screen    \n","            print('{} Loss: {:.4f} Acc: {:.4f}'.format(\n","                phase, epoch_loss, epoch_acc))\n","            # deep copy the model\n","            if phase == 'test' and epoch_acc > best_acc:\n","                best_acc = epoch_acc\n","                best_model_wts = copy.deepcopy(model.state_dict())\n","\n","        print()\n","\n","    end_time = tm.time()\n","    duration_time = np.subtract(end_time, start_time) \n","\n","    print('Training complete in {:.0f}m {:.0f}s'.format(\n","        duration_time // 60, duration_time % 60))\n","    print('Best Test Acc: {:4f}'.format(best_acc))\n","\n","    # load best model weights\n","    model.load_state_dict(best_model_wts)\n","    return model,trainLoss, trainAccuracy, testLoss, testAccuracy, duration_time"],"execution_count":14,"outputs":[]},{"cell_type":"code","metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"Wa2XP_AEIZeA","outputId":"0d9d2ff0-e42e-4488-ba55-de4c5b44d005"},"source":["  for transferLearning in transferLearnings:\n","    for architecture in architectures:\n","      for dropOut in dropouts:\n","        for learningRate in learningRates:\n","          for batchSize in batchSizes:\n","            for epoch in epochs:\n","              # Load a pretrained model and reset final fully connected layer.\n","              model = architecture\n","              # Here, we need to freeze all the network except the final layer.\n","              # We need to set requires_grad == False to freeze the parameters so that the gradients are not computed in backward()\n","              if (transferLearning=='fineTune'):\n","                for param in model.parameters():\n","                    param.requires_grad = False\n","              # Define new classifier according to binary output. # Parameters of newly constructed modules have requires_grad=True by default\n","              if (architecture==alexnet):\n","                modelName='alexnet'\n","                model.classifier = nn.Sequential(\n","                                nn.Dropout(dropOut),\n","                                nn.Linear(9216, 4096),\n","                                nn.ReLU(),\n","                                nn.Dropout(dropOut),\n","                                nn.Linear(4096, 1000),\n","                                nn.ReLU(),\n","                                nn.Dropout(dropOut),\n","                                nn.Linear(1000,10),\n","                                nn.ReLU(),\n","                                nn.Dropout(dropOut),\n","                                nn.Linear(10,2),\n","              )\n","              elif (architecture==vgg16):\n","                modelName='vgg16'\n","                model.classifier =  nn.Sequential(\n","                                nn.Dropout(dropOut),\n","                                nn.Linear(25088, 4096),\n","                                nn.ReLU(),\n","                                nn.Dropout(dropOut),\n","                                nn.Linear(4096, 1000),\n","                                nn.ReLU(),\n","                                nn.Dropout(dropOut),\n","                                nn.Linear(1000,10),\n","                                nn.ReLU(),\n","                                nn.Dropout(dropOut),\n","                                nn.Linear(10,2),\n","              )\n","              print('modelName:',modelName)\n","              print('transferLearning:',transferLearning)\n","              print('batchSize:',batchSize)\n","              print('dropOut:',dropOut)\n","              print('learningRate:',learningRate)\n","              print('epoch:',epoch)\n","              \n","              model = model.to(device)\n","              # Observe that all parameters are being optimized\n","              optimizer = optim.SGD(model.parameters(), lr=learningRate, momentum=0.9)\n","              # Define Loss Function\n","              criterion = nn.CrossEntropyLoss()\n","              # Define learning scheduler according to the epoch_nums\n","              if (epoch==10):\n","                step_lr_scheduler = lr_scheduler.StepLR(optimizer, step_size=learningSchedulerSteps[0], gamma=0.1)\n","              elif (epoch==20):\n","                step_lr_scheduler = lr_scheduler.StepLR(optimizer, step_size=learningSchedulerSteps[1], gamma=0.1)\n","              else:\n","                step_lr_scheduler = lr_scheduler.StepLR(optimizer, step_size=learningSchedulerSteps[2], gamma=0.1)\n","              modelNew, trainLoss, trainAccuracy, testLoss, testAccuracy, duration_time = train_model(model, criterion, optimizer, step_lr_scheduler, epoch)\n","\n","              #visualize and save loss&accuracy\n","              plt.style.use(\"ggplot\")\n","              plt.figure()\n","              plt.plot(np.arange(1, epoch+1), trainAccuracy, label=\"train_acc\")\n","              plt.plot(np.arange(1, epoch+1), testAccuracy, label=\"test_acc\")\n","              plt.title(\"Accuracy on Dataset \\n (CNN:{}, Transfer Learning Type:{}, Batch Size: {} \\n Drop: {}, Learn: {}, Epochs: {})\".format(modelName, transferLearning, batchSize, dropOut, learningRate, epoch) )\n","              plt.xlabel(\"Epoch #\")\n","              plt.ylabel(\"Accuracy\")\n","              plt.ylim(0.3,1.0)\n","              plt.legend(loc=\"lower right\")\n","              #plt.savefig(\"Plot_Accuracy_%s_%s_%s_%s_%s_%s.png\" % (modelName, transferLearning, batchSize, dropOut, learningRate, epoch))\n","              plt.savefig('/content/gdrive/MyDrive/METU/01.courses/METU_MMI_727/project/results/'+\"Plot_Accuracy_%s_%s_%s_%s_%s_%s.png\" % (modelName, transferLearning, batchSize, dropOut, learningRate, epoch))\n","\n","              plt.style.use(\"ggplot\")\n","              plt.figure()\n","              plt.plot(np.arange(1, epoch+1), trainLoss, label=\"train_loss\")\n","              plt.plot(np.arange(1, epoch+1), testLoss, label=\"test_loss\")\n","              plt.title(\"Loss on Dataset \\n (CNN:{}, Transfer Learning Type:{}, Batch Size: {} \\n Drop: {}, Learn: {}, Epochs: {})\".format(modelName, transferLearning, batchSize, dropOut, learningRate, epoch) )\n","              plt.xlabel(\"Epoch #\")\n","              plt.ylabel(\"Loss\")\n","              plt.ylim(0.3,1.0)\n","              plt.legend(loc=\"lower right\")\n","              #plt.savefig(\"Plot_Loss_%s_%s_%s_%s_%s_%s.png\" % (modelName, transferLearning, batchSize, dropOut, learningRate, epoch))\n","              plt.savefig(\"/content/gdrive/MyDrive/METU/01.courses/METU_MMI_727/project/results/\"+\"Plot_Loss_%s_%s_%s_%s_%s_%s.png\" % (modelName, transferLearning, batchSize, dropOut, learningRate, epoch))\n","\n","              model_save_name = \"Model_%s_%s_%s_%s_%s_%s.pt\" % (modelName, transferLearning, batchSize, dropOut, learningRate, epoch)\n","              #path = '/content/source/'+model_save_name\n","              path = '/content/gdrive/MyDrive/METU/01.courses/METU_MMI_727/project/results/'+model_save_name\n","              torch.save(modelNew, path)\n","            \n","              # dictionary of lists \n","              dict = {'trainLoss': trainLoss, 'trainAccuracy': trainAccuracy, 'testLoss': testLoss, 'testAccuracy':testAccuracy, 'duration_time':duration_time} \n","              #pd.DataFrame(dict).to_csv(\"Results_%s_%s_%s_%s_%s_%s.csv\" % (modelName, transferLearning, batchSize, dropOut, learningRate, epoch))\n","              pd.DataFrame(dict).to_csv('/content/gdrive/MyDrive/METU/01.courses/METU_MMI_727/project/results/'+\"Results_%s_%s_%s_%s_%s_%s.csv\" % (modelName, transferLearning, batchSize, dropOut, learningRate, epoch))\n","\n","    "],"execution_count":null,"outputs":[{"output_type":"stream","text":["modelName: alexnet\n","transferLearning: fineTune\n","batchSize: 16\n","dropOut: 0.0\n","learningRate: 0.0001\n","epoch: 10\n","Epoch 1/10\n","----------\n","train Loss: 0.5126 Acc: 0.7480\n","test Loss: 0.3464 Acc: 0.8355\n","\n","Epoch 2/10\n","----------\n","train Loss: 0.3240 Acc: 0.8658\n","test Loss: 0.2543 Acc: 0.8908\n","\n","Epoch 3/10\n","----------\n","train Loss: 0.2687 Acc: 0.8892\n","test Loss: 0.2494 Acc: 0.8895\n","\n","Epoch 4/10\n","----------\n","train Loss: 0.2718 Acc: 0.8880\n","test Loss: 0.2438 Acc: 0.8934\n","\n","Epoch 5/10\n","----------\n","train Loss: 0.2634 Acc: 0.8897\n","test Loss: 0.2437 Acc: 0.8921\n","\n","Epoch 6/10\n","----------\n","train Loss: 0.2696 Acc: 0.8937\n","test Loss: 0.2430 Acc: 0.8934\n","\n","Epoch 7/10\n","----------\n","train Loss: 0.2676 Acc: 0.8889\n","test Loss: 0.2429 Acc: 0.8934\n","\n","Epoch 8/10\n","----------\n","train Loss: 0.2702 Acc: 0.8895\n","test Loss: 0.2428 Acc: 0.8934\n","\n","Epoch 9/10\n","----------\n","train Loss: 0.2764 Acc: 0.8819\n","test Loss: 0.2428 Acc: 0.8934\n","\n","Epoch 10/10\n","----------\n","train Loss: 0.2634 Acc: 0.8895\n","test Loss: 0.2428 Acc: 0.8934\n","\n","Training complete in 34m 27s\n","Best Test Acc: 0.893421\n","modelName: alexnet\n","transferLearning: fineTune\n","batchSize: 16\n","dropOut: 0.0\n","learningRate: 0.0001\n","epoch: 20\n","Epoch 1/20\n","----------\n","train Loss: 0.5170 Acc: 0.7547\n","test Loss: 0.3743 Acc: 0.8250\n","\n","Epoch 2/20\n","----------\n","train Loss: 0.3338 Acc: 0.8506\n","test Loss: 0.2591 Acc: 0.8895\n","\n","Epoch 3/20\n","----------\n","train Loss: 0.2923 Acc: 0.8706\n","test Loss: 0.2217 Acc: 0.9053\n","\n","Epoch 4/20\n","----------\n","train Loss: 0.2765 Acc: 0.8852\n","test Loss: 0.2090 Acc: 0.9171\n","\n","Epoch 5/20\n","----------\n","train Loss: 0.2518 Acc: 0.8976\n","test Loss: 0.1937 Acc: 0.9184\n","\n","Epoch 6/20\n","----------\n","train Loss: 0.2238 Acc: 0.9052\n","test Loss: 0.1885 Acc: 0.9171\n","\n","Epoch 7/20\n","----------\n","train Loss: 0.2224 Acc: 0.9072\n","test Loss: 0.1872 Acc: 0.9158\n","\n","Epoch 8/20\n","----------\n","train Loss: 0.2208 Acc: 0.9120\n","test Loss: 0.1860 Acc: 0.9171\n","\n","Epoch 9/20\n","----------\n","train Loss: 0.2281 Acc: 0.9069\n","test Loss: 0.1834 Acc: 0.9237\n","\n","Epoch 10/20\n","----------\n","train Loss: 0.2136 Acc: 0.9083\n","test Loss: 0.1797 Acc: 0.9263\n","\n","Epoch 11/20\n","----------\n","train Loss: 0.2189 Acc: 0.9117\n","test Loss: 0.1793 Acc: 0.9224\n","\n","Epoch 12/20\n","----------\n","train Loss: 0.2200 Acc: 0.9117\n","test Loss: 0.1793 Acc: 0.9224\n","\n","Epoch 13/20\n","----------\n","train Loss: 0.2163 Acc: 0.9108\n","test Loss: 0.1792 Acc: 0.9237\n","\n","Epoch 14/20\n","----------\n","train Loss: 0.2126 Acc: 0.9114\n","test Loss: 0.1791 Acc: 0.9237\n","\n","Epoch 15/20\n","----------\n","train Loss: 0.2031 Acc: 0.9243\n","test Loss: 0.1793 Acc: 0.9237\n","\n","Epoch 16/20\n","----------\n","train Loss: 0.2214 Acc: 0.9058\n","test Loss: 0.1793 Acc: 0.9237\n","\n","Epoch 17/20\n","----------\n","train Loss: 0.2109 Acc: 0.9089\n","test Loss: 0.1794 Acc: 0.9237\n","\n","Epoch 18/20\n","----------\n","train Loss: 0.2066 Acc: 0.9153\n","test Loss: 0.1793 Acc: 0.9237\n","\n","Epoch 19/20\n","----------\n","train Loss: 0.2119 Acc: 0.9111\n","test Loss: 0.1794 Acc: 0.9237\n","\n","Epoch 20/20\n","----------\n","train Loss: 0.2146 Acc: 0.9139\n","test Loss: 0.1793 Acc: 0.9237\n","\n","Training complete in 24m 57s\n","Best Test Acc: 0.926316\n","modelName: alexnet\n","transferLearning: fineTune\n","batchSize: 64\n","dropOut: 0.0\n","learningRate: 0.0001\n","epoch: 10\n","Epoch 1/10\n","----------\n","train Loss: 0.4928 Acc: 0.7688\n","test Loss: 0.3445 Acc: 0.8342\n","\n","Epoch 2/10\n","----------\n","train Loss: 0.3336 Acc: 0.8504\n","test Loss: 0.2525 Acc: 0.8855\n","\n","Epoch 3/10\n","----------\n","train Loss: 0.2776 Acc: 0.8774\n","test Loss: 0.2453 Acc: 0.8947\n","\n","Epoch 4/10\n","----------\n","train Loss: 0.2766 Acc: 0.8805\n","test Loss: 0.2439 Acc: 0.8947\n","\n","Epoch 5/10\n","----------\n","train Loss: 0.2697 Acc: 0.8796\n","test Loss: 0.2418 Acc: 0.8961\n","\n","Epoch 6/10\n","----------\n","train Loss: 0.2739 Acc: 0.8802\n","test Loss: 0.2412 Acc: 0.8987\n","\n","Epoch 7/10\n","----------\n","train Loss: 0.2743 Acc: 0.8897\n","test Loss: 0.2411 Acc: 0.8987\n","\n","Epoch 8/10\n","----------\n","train Loss: 0.2643 Acc: 0.8864\n","test Loss: 0.2412 Acc: 0.8987\n","\n","Epoch 9/10\n","----------\n","train Loss: 0.2759 Acc: 0.8788\n","test Loss: 0.2412 Acc: 0.8987\n","\n","Epoch 10/10\n","----------\n","train Loss: 0.2692 Acc: 0.8858\n","test Loss: 0.2412 Acc: 0.8987\n","\n","Training complete in 12m 9s\n","Best Test Acc: 0.898684\n","modelName: alexnet\n","transferLearning: fineTune\n","batchSize: 64\n","dropOut: 0.0\n","learningRate: 0.0001\n","epoch: 20\n","Epoch 1/20\n","----------\n","train Loss: 0.5794 Acc: 0.7623\n","test Loss: 0.4743 Acc: 0.8013\n","\n","Epoch 2/20\n","----------\n","train Loss: 0.3891 Acc: 0.8512\n","test Loss: 0.2874 Acc: 0.8697\n","\n","Epoch 3/20\n","----------\n","train Loss: 0.3051 Acc: 0.8644\n","test Loss: 0.2391 Acc: 0.8895\n","\n","Epoch 4/20\n","----------\n","train Loss: 0.2705 Acc: 0.8830\n","test Loss: 0.2820 Acc: 0.8842\n","\n","Epoch 5/20\n","----------\n","train Loss: 0.2487 Acc: 0.8987\n","test Loss: 0.2037 Acc: 0.9013\n","\n","Epoch 6/20\n","----------\n","train Loss: 0.2316 Acc: 0.9044\n","test Loss: 0.2012 Acc: 0.9066\n","\n","Epoch 7/20\n","----------\n","train Loss: 0.2241 Acc: 0.9069\n","test Loss: 0.2003 Acc: 0.9132\n","\n","Epoch 8/20\n","----------\n","train Loss: 0.2235 Acc: 0.9069\n","test Loss: 0.1978 Acc: 0.9145\n","\n","Epoch 9/20\n","----------\n","train Loss: 0.2190 Acc: 0.9125\n","test Loss: 0.1895 Acc: 0.9171\n","\n","Epoch 10/20\n","----------\n","train Loss: 0.2194 Acc: 0.9097\n","test Loss: 0.1867 Acc: 0.9211\n","\n","Epoch 11/20\n","----------\n","train Loss: 0.2193 Acc: 0.9103\n","test Loss: 0.1864 Acc: 0.9211\n","\n","Epoch 12/20\n","----------\n","train Loss: 0.2078 Acc: 0.9148\n","test Loss: 0.1859 Acc: 0.9197\n","\n","Epoch 13/20\n","----------\n","train Loss: 0.2230 Acc: 0.9089\n","test Loss: 0.1858 Acc: 0.9211\n","\n","Epoch 14/20\n","----------\n","train Loss: 0.2184 Acc: 0.9136\n","test Loss: 0.1853 Acc: 0.9211\n","\n","Epoch 15/20\n","----------\n","train Loss: 0.2238 Acc: 0.9100\n","test Loss: 0.1850 Acc: 0.9211\n","\n","Epoch 16/20\n","----------\n","train Loss: 0.2149 Acc: 0.9049\n","test Loss: 0.1849 Acc: 0.9197\n","\n","Epoch 17/20\n","----------\n","train Loss: 0.2139 Acc: 0.9122\n","test Loss: 0.1850 Acc: 0.9211\n","\n","Epoch 18/20\n","----------\n","train Loss: 0.2168 Acc: 0.9041\n","test Loss: 0.1848 Acc: 0.9197\n","\n","Epoch 19/20\n","----------\n","train Loss: 0.2209 Acc: 0.9086\n","test Loss: 0.1847 Acc: 0.9197\n","\n","Epoch 20/20\n","----------\n","train Loss: 0.2144 Acc: 0.9083\n","test Loss: 0.1846 Acc: 0.9171\n","\n","Training complete in 24m 40s\n","Best Test Acc: 0.921053\n","modelName: alexnet\n","transferLearning: fineTune\n","batchSize: 16\n","dropOut: 0.0\n","learningRate: 0.001\n","epoch: 10\n","Epoch 1/10\n","----------\n","train Loss: 0.3777 Acc: 0.8281\n","test Loss: 0.2256 Acc: 0.9039\n","\n","Epoch 2/10\n","----------\n","train Loss: 0.2691 Acc: 0.8875\n","test Loss: 0.1953 Acc: 0.9197\n","\n","Epoch 3/10\n","----------\n","train Loss: 0.2327 Acc: 0.9044\n","test Loss: 0.1838 Acc: 0.9211\n","\n","Epoch 4/10\n","----------\n","train Loss: 0.2206 Acc: 0.9058\n","test Loss: 0.1817 Acc: 0.9184\n","\n","Epoch 5/10\n","----------\n","train Loss: 0.2055 Acc: 0.9131\n","test Loss: 0.1771 Acc: 0.9250\n","\n","Epoch 6/10\n","----------\n","train Loss: 0.2118 Acc: 0.9103\n","test Loss: 0.1762 Acc: 0.9263\n","\n","Epoch 7/10\n","----------\n","train Loss: 0.2079 Acc: 0.9131\n","test Loss: 0.1761 Acc: 0.9250\n","\n","Epoch 8/10\n","----------\n","train Loss: 0.2085 Acc: 0.9136\n","test Loss: 0.1759 Acc: 0.9250\n","\n","Epoch 9/10\n","----------\n","train Loss: 0.2070 Acc: 0.9167\n","test Loss: 0.1759 Acc: 0.9250\n","\n","Epoch 10/10\n","----------\n","train Loss: 0.2046 Acc: 0.9111\n","test Loss: 0.1759 Acc: 0.9250\n","\n","Training complete in 12m 24s\n","Best Test Acc: 0.926316\n","modelName: alexnet\n","transferLearning: fineTune\n","batchSize: 16\n","dropOut: 0.0\n","learningRate: 0.001\n","epoch: 20\n","Epoch 1/20\n","----------\n","train Loss: 0.3846 Acc: 0.8169\n","test Loss: 0.2593 Acc: 0.8803\n","\n","Epoch 2/20\n","----------\n","train Loss: 0.2706 Acc: 0.8827\n","test Loss: 0.2006 Acc: 0.9184\n","\n","Epoch 3/20\n","----------\n","train Loss: 0.2415 Acc: 0.8962\n","test Loss: 0.1939 Acc: 0.9197\n","\n","Epoch 4/20\n","----------\n","train Loss: 0.2173 Acc: 0.9058\n","test Loss: 0.2083 Acc: 0.9224\n","\n","Epoch 5/20\n","----------\n","train Loss: 0.2005 Acc: 0.9204\n","test Loss: 0.1895 Acc: 0.9276\n","\n","Epoch 6/20\n","----------\n","train Loss: 0.1710 Acc: 0.9302\n","test Loss: 0.1536 Acc: 0.9434\n","\n","Epoch 7/20\n","----------\n","train Loss: 0.1722 Acc: 0.9288\n","test Loss: 0.1463 Acc: 0.9434\n","\n","Epoch 8/20\n","----------\n","train Loss: 0.1607 Acc: 0.9339\n","test Loss: 0.1455 Acc: 0.9461\n","\n","Epoch 9/20\n","----------\n","train Loss: 0.1442 Acc: 0.9401\n","test Loss: 0.1527 Acc: 0.9434\n","\n","Epoch 10/20\n","----------\n","train Loss: 0.1563 Acc: 0.9342\n","test Loss: 0.1439 Acc: 0.9461\n","\n","Epoch 11/20\n","----------\n","train Loss: 0.1390 Acc: 0.9429\n","test Loss: 0.1442 Acc: 0.9474\n","\n","Epoch 12/20\n","----------\n","train Loss: 0.1385 Acc: 0.9446\n","test Loss: 0.1448 Acc: 0.9474\n","\n","Epoch 13/20\n","----------\n","train Loss: 0.1505 Acc: 0.9378\n","test Loss: 0.1441 Acc: 0.9474\n","\n","Epoch 14/20\n","----------\n","train Loss: 0.1421 Acc: 0.9435\n","test Loss: 0.1424 Acc: 0.9474\n","\n","Epoch 15/20\n","----------\n","train Loss: 0.1469 Acc: 0.9370\n","test Loss: 0.1415 Acc: 0.9474\n","\n","Epoch 16/20\n","----------\n","train Loss: 0.1401 Acc: 0.9432\n","test Loss: 0.1416 Acc: 0.9474\n","\n","Epoch 17/20\n","----------\n","train Loss: 0.1452 Acc: 0.9446\n","test Loss: 0.1417 Acc: 0.9474\n","\n","Epoch 18/20\n","----------\n","train Loss: 0.1552 Acc: 0.9347\n","test Loss: 0.1416 Acc: 0.9474\n","\n","Epoch 19/20\n","----------\n","train Loss: 0.1527 Acc: 0.9387\n","test Loss: 0.1415 Acc: 0.9474\n","\n","Epoch 20/20\n","----------\n","train Loss: 0.1463 Acc: 0.9390\n","test Loss: 0.1415 Acc: 0.9474\n","\n","Training complete in 24m 44s\n","Best Test Acc: 0.947368\n","modelName: alexnet\n","transferLearning: fineTune\n","batchSize: 64\n","dropOut: 0.0\n","learningRate: 0.001\n","epoch: 10\n","Epoch 1/10\n","----------\n","train Loss: 0.3754 Acc: 0.8267\n","test Loss: 0.3257 Acc: 0.8724\n","\n","Epoch 2/10\n","----------\n","train Loss: 0.2701 Acc: 0.8796\n","test Loss: 0.2071 Acc: 0.9145\n","\n","Epoch 3/10\n","----------\n","train Loss: 0.2281 Acc: 0.9010\n","test Loss: 0.1996 Acc: 0.9145\n","\n","Epoch 4/10\n","----------\n","train Loss: 0.2119 Acc: 0.9139\n","test Loss: 0.2024 Acc: 0.9184\n","\n","Epoch 5/10\n","----------\n","train Loss: 0.2075 Acc: 0.9122\n","test Loss: 0.1980 Acc: 0.9184\n","\n","Epoch 6/10\n","----------\n","train Loss: 0.1983 Acc: 0.9150\n","test Loss: 0.1986 Acc: 0.9184\n","\n","Epoch 7/10\n","----------\n","train Loss: 0.1932 Acc: 0.9179\n","test Loss: 0.1980 Acc: 0.9184\n","\n","Epoch 8/10\n","----------\n","train Loss: 0.2085 Acc: 0.9089\n","test Loss: 0.1978 Acc: 0.9184\n","\n","Epoch 9/10\n","----------\n","train Loss: 0.1989 Acc: 0.9179\n","test Loss: 0.1978 Acc: 0.9184\n","\n","Epoch 10/10\n","----------\n","train Loss: 0.2092 Acc: 0.9091\n","test Loss: 0.1977 Acc: 0.9184\n","\n","Training complete in 12m 24s\n","Best Test Acc: 0.918421\n","modelName: alexnet\n","transferLearning: fineTune\n","batchSize: 64\n","dropOut: 0.0\n","learningRate: 0.001\n","epoch: 20\n","Epoch 1/20\n","----------\n","train Loss: 0.3828 Acc: 0.8191\n","test Loss: 0.2639 Acc: 0.8921\n","\n","Epoch 2/20\n","----------\n","train Loss: 0.2842 Acc: 0.8754\n","test Loss: 0.2101 Acc: 0.9171\n","\n","Epoch 3/20\n","----------\n","train Loss: 0.2337 Acc: 0.8923\n","test Loss: 0.1852 Acc: 0.9197\n","\n","Epoch 4/20\n","----------\n","train Loss: 0.2329 Acc: 0.9038\n","test Loss: 0.1634 Acc: 0.9263\n","\n","Epoch 5/20\n","----------\n","train Loss: 0.2013 Acc: 0.9193\n","test Loss: 0.1904 Acc: 0.9250\n","\n","Epoch 6/20\n","----------\n","train Loss: 0.1820 Acc: 0.9207\n","test Loss: 0.1462 Acc: 0.9408\n","\n","Epoch 7/20\n","----------\n","train Loss: 0.1609 Acc: 0.9319\n","test Loss: 0.1470 Acc: 0.9395\n","\n","Epoch 8/20\n","----------\n","train Loss: 0.1587 Acc: 0.9322\n","test Loss: 0.1422 Acc: 0.9434\n","\n","Epoch 9/20\n","----------\n","train Loss: 0.1572 Acc: 0.9350\n","test Loss: 0.1352 Acc: 0.9447\n","\n","Epoch 10/20\n","----------\n","train Loss: 0.1542 Acc: 0.9395\n","test Loss: 0.1386 Acc: 0.9395\n","\n","Epoch 11/20\n","----------\n","train Loss: 0.1492 Acc: 0.9378\n","test Loss: 0.1371 Acc: 0.9408\n","\n","Epoch 12/20\n","----------\n","train Loss: 0.1404 Acc: 0.9437\n","test Loss: 0.1366 Acc: 0.9408\n","\n","Epoch 13/20\n","----------\n","train Loss: 0.1527 Acc: 0.9364\n","test Loss: 0.1373 Acc: 0.9382\n","\n","Epoch 14/20\n","----------\n","train Loss: 0.1475 Acc: 0.9384\n","test Loss: 0.1373 Acc: 0.9395\n","\n","Epoch 15/20\n","----------\n","train Loss: 0.1465 Acc: 0.9415\n","test Loss: 0.1404 Acc: 0.9395\n","\n","Epoch 16/20\n","----------\n","train Loss: 0.1500 Acc: 0.9401\n","test Loss: 0.1396 Acc: 0.9395\n","\n","Epoch 17/20\n","----------\n","train Loss: 0.1491 Acc: 0.9418\n","test Loss: 0.1392 Acc: 0.9408\n","\n","Epoch 18/20\n","----------\n","train Loss: 0.1438 Acc: 0.9429\n","test Loss: 0.1387 Acc: 0.9395\n","\n","Epoch 19/20\n","----------\n","train Loss: 0.1445 Acc: 0.9426\n","test Loss: 0.1384 Acc: 0.9395\n","\n","Epoch 20/20\n","----------\n","train Loss: 0.1458 Acc: 0.9421\n","test Loss: 0.1382 Acc: 0.9395\n","\n","Training complete in 25m 21s\n","Best Test Acc: 0.944737\n","modelName: alexnet\n","transferLearning: fineTune\n","batchSize: 16\n","dropOut: 0.2\n","learningRate: 0.0001\n","epoch: 10\n","Epoch 1/10\n","----------\n","train Loss: 0.5198 Acc: 0.7412\n","test Loss: 0.3615 Acc: 0.8342\n","\n","Epoch 2/10\n","----------\n","train Loss: 0.3797 Acc: 0.8273\n","test Loss: 0.2770 Acc: 0.8776\n","\n","Epoch 3/10\n","----------\n","train Loss: 0.3326 Acc: 0.8560\n","test Loss: 0.2714 Acc: 0.8789\n","\n","Epoch 4/10\n","----------\n","train Loss: 0.3271 Acc: 0.8563\n","test Loss: 0.2647 Acc: 0.8829\n","\n","Epoch 5/10\n","----------\n","train Loss: 0.3204 Acc: 0.8655\n","test Loss: 0.2645 Acc: 0.8829\n","\n","Epoch 6/10\n","----------\n","train Loss: 0.3246 Acc: 0.8613\n","test Loss: 0.2636 Acc: 0.8829\n","\n","Epoch 7/10\n","----------\n","train Loss: 0.3193 Acc: 0.8678\n","test Loss: 0.2637 Acc: 0.8829\n","\n","Epoch 8/10\n","----------\n","train Loss: 0.3226 Acc: 0.8641\n","test Loss: 0.2636 Acc: 0.8829\n","\n","Epoch 9/10\n","----------\n","train Loss: 0.3234 Acc: 0.8579\n","test Loss: 0.2636 Acc: 0.8829\n","\n","Epoch 10/10\n","----------\n","train Loss: 0.3173 Acc: 0.8684\n","test Loss: 0.2636 Acc: 0.8829\n","\n","Training complete in 12m 56s\n","Best Test Acc: 0.882895\n","modelName: alexnet\n","transferLearning: fineTune\n","batchSize: 16\n","dropOut: 0.2\n","learningRate: 0.0001\n","epoch: 20\n","Epoch 1/20\n","----------\n","train Loss: 0.5653 Acc: 0.6996\n","test Loss: 0.4018 Acc: 0.8211\n","\n","Epoch 2/20\n","----------\n","train Loss: 0.3978 Acc: 0.8284\n","test Loss: 0.3034 Acc: 0.8671\n","\n","Epoch 3/20\n","----------\n","train Loss: 0.3451 Acc: 0.8419\n","test Loss: 0.2636 Acc: 0.8855\n","\n","Epoch 4/20\n","----------\n","train Loss: 0.3244 Acc: 0.8563\n","test Loss: 0.2808 Acc: 0.8684\n","\n","Epoch 5/20\n","----------\n","train Loss: 0.2980 Acc: 0.8765\n","test Loss: 0.2459 Acc: 0.8934\n","\n","Epoch 6/20\n","----------\n","train Loss: 0.2861 Acc: 0.8824\n","test Loss: 0.2285 Acc: 0.8934\n","\n","Epoch 7/20\n","----------\n","train Loss: 0.2676 Acc: 0.8875\n","test Loss: 0.2262 Acc: 0.9053\n","\n","Epoch 8/20\n","----------\n","train Loss: 0.2795 Acc: 0.8793\n","test Loss: 0.2239 Acc: 0.9039\n","\n","Epoch 9/20\n","----------\n","train Loss: 0.2778 Acc: 0.8875\n","test Loss: 0.2245 Acc: 0.9066\n","\n","Epoch 10/20\n","----------\n","train Loss: 0.2746 Acc: 0.8897\n","test Loss: 0.2254 Acc: 0.9053\n","\n","Epoch 11/20\n","----------\n","train Loss: 0.2683 Acc: 0.8925\n","test Loss: 0.2210 Acc: 0.9053\n","\n","Epoch 12/20\n","----------\n","train Loss: 0.2764 Acc: 0.8895\n","test Loss: 0.2206 Acc: 0.9053\n","\n","Epoch 13/20\n","----------\n","train Loss: 0.2787 Acc: 0.8833\n","test Loss: 0.2199 Acc: 0.9039\n","\n","Epoch 14/20\n","----------\n","train Loss: 0.2708 Acc: 0.8931\n","test Loss: 0.2193 Acc: 0.9026\n","\n","Epoch 15/20\n","----------\n","train Loss: 0.2665 Acc: 0.8937\n","test Loss: 0.2204 Acc: 0.9066\n","\n","Epoch 16/20\n","----------\n","train Loss: 0.2742 Acc: 0.8850\n","test Loss: 0.2202 Acc: 0.9066\n","\n","Epoch 17/20\n","----------\n"],"name":"stdout"}]},{"cell_type":"markdown","metadata":{"id":"rql-L8X1dWMO"},"source":["# Transfer Learning Approach"]},{"cell_type":"code","metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"vc68VUrOdZBE","outputId":"0a85880b-ec25-44b7-906e-2042f5ccf493"},"source":["'''\n","\n","# Load a pretrained model and reset final fully connected layer.\n","model = models.resnet18(pretrained=True)\n","# Here the size of each output sample is set to 2.\n","# Alternatively, it can be generalized to nn.Linear(num_ftrs, len(class_names)).\n","model.fc = classifier\n","#model.classifier = classifier\n","\n","model = model.to(device)\n","\n","criterion = nn.CrossEntropyLoss()\n","\n","# Observe that all parameters are being optimized\n","optimizer = optim.SGD(model.parameters(), lr=0.001)\n","\n","# StepLR Decays the learning rate of each parameter group by gamma every step_size epochs\n","# Decay LR by a factor of 0.1 every 7 epochs\n","# Learning rate scheduling should be applied after optimizer’s update\n","# e.g., you should write your code this way:\n","# for epoch in range(100):\n","#     train(...)\n","#     validate(...)\n","#     scheduler.step()\n","\n","step_lr_scheduler = lr_scheduler.StepLR(optimizer, step_size=7, gamma=0.1)\n","\n","model = train_model(model, criterion, optimizer, step_lr_scheduler, num_epochs=5)\n","\n","\n","#### ConvNet as fixed feature extractor ####\n","# Here, we need to freeze all the network except the final layer.\n","# We need to set requires_grad == False to freeze the parameters so that the gradients are not computed in backward()\n","model_conv = torchvision.models.resnet18(pretrained=True)\n","for param in model_conv.parameters():\n","    param.requires_grad = False\n","\n","# Parameters of newly constructed modules have requires_grad=True by default\n","\n","model_conv.fc = classifier\n","#model_conv.classifier = classifier\n","\n","\n","model_conv = model_conv.to(device)\n","\n","criterion = nn.CrossEntropyLoss()\n","\n","# Observe that only parameters of final layer are being optimized as\n","# opposed to before.\n","optimizer_conv = optim.SGD(model_conv.fc.parameters(), lr=0.001, momentum=0.9)\n","\n","# Decay LR by a factor of 0.1 every 7 epochs\n","exp_lr_scheduler = lr_scheduler.StepLR(optimizer_conv, step_size=7, gamma=0.1)\n","\n","model_conv = train_model(model_conv, criterion, optimizer_conv,\n","                         exp_lr_scheduler, num_epochs=5)"],"execution_count":null,"outputs":[{"output_type":"stream","text":["Epoch 1/5\n","----------\n","train Loss: 0.5647 Acc: 0.7170\n","test Loss: 0.3186 Acc: 0.8921\n","\n","Epoch 2/5\n","----------\n"],"name":"stdout"}]},{"cell_type":"code","metadata":{"id":"m8C_RvwOeH95"},"source":["'''\n","#after this training session, the weights are saved. \n","model.save_weights('trained_weights.h5')"],"execution_count":null,"outputs":[]}]}