Deploy our PyTorch model with Flask and Heroku. Create a simple Flask app with a REST API that returns the result as json data, and then deploy it to Heroku. Here we will do image classification, and we can send images to our heroku app and then predict it with our live running app.

https://pytorch-deploy-test.herokuapp.com/predict

1. Training code, model output and metrics.csv is located in train folder. Initialized repo with dvc and tracked changes using dvc.
2. Achieved more than 70% accuracy on all classes:
   Accuracy on the test images: 85 %
   Accuracy for class airplane is: 87.7 %
   Accuracy for class automobile is: 93.4 %
   Accuracy for class bird is: 78.5 %
   Accuracy for class cat is: 75.2 %
   Accuracy for class deer is: 83.3 %
   Accuracy for class dog is: 74.1 %
   Accuracy for class frog is: 89.2 %
   Accuracy for class horse is: 92.5 %
   Accuracy for class ship is: 90.1 %
   Accuracy for class truck is: 92.3 %
3. Wrote following 6 test cases in unittest.py. Used pytest for writing unit tests.
   test_check_if_model_file_present_in_root_folder
   test_check_if_data_folder_present_in_root_folder
   test_check_if_metrics_csv_present_in_root_folder
   test_validate_train_accuracy_greater_than_70_pct
   test_validate_test_accuracy_greater_than_70_pct
   test_validate_individual_class_accuracy_greater_than_70_pct
4. Model training output: Model Training Output