Using classical and neural image embeddings and finetuned end-to-end networks to achieve top-tier performance on a vehicle type classification task. Containerized and deployed model as a web app
Input: all features train and test files. to construct X in the training run you should drop all string variables, and have 'Class', as the y variable. to construct X in the prediction run you should drop all string variables and there should not be a y
Output:
[x] joblib file of model.
[x] Excel file containing time to fit/train overall, train and validation accuracies.
[x] Pickle file of dictionaries of all hyperparameters searched and optimal hyperparameters
[x] Excel file of predictions - this should be all string columns in all features test + one column called classifier_name + '_Prediction'. Excel file of inference time.
Input: all features train and test files. to construct X in the training run you should drop all string variables, and have 'Class', as the y variable. to construct X in the prediction run you should drop all string variables and there should not be a y
Output: