Install Requirements– pip install -r requirements.txt

Running– Preferred: Run main.ipynb notebook. Approx. Total runtime (three simple models) ~ 6min python main.py or python -m main produces the same results (but matplotlib will pause running code for its interactive graphs)

What to Do

Go to models directory to find/add your model, and configure how they work there Go to main.ipynb and look for "Model Initialization Logic Here" to change how you want to initialize your the model(s) Use train_model() helper function to train your model Within the notebooks, you can comment out or delete cells with other other models' training-validation(not testing) loops for your own. You can also always use backup_main.ipynb as well for convenience. Use evaluate_model() helper function to evaluate your model You can also refer to the later cells as a guide to training your models If you're not using this helper function then manually plug your model into an Evaluator and run eval() Test your final model at the end against the other models, under the "Final Testing and Evaluations" header

Major Components

Dataset– Looks like the mnist_reader provided reads data from the .gz files directly, so no need to extract them dataevaluator.py contains the DataEvaluator class, defining how we preprocess the data. Currently it only runs a simple split and plots a graph

Models– models directory contains basic models for kNN, Logistic, and MLP Models MUST be plugged into Evaluator class to be evaluated properly models/logsdirectory contains logs of model runs and results when plugged in to Evaluator

Evaluator– modelevaluator.py contains the Evaluator class. Requires only the model and a name to initialize. eval() is the best function to run all evaluations.

Helper Functions– Stored in helper.py to simplify code. Refer to notebooks/code for implementation examples train_model will train your model using model.fit(). Logs into models/logs and returns training time evaluate_model will evaluate your model. logs into models/logs. Returns an Evaluator, useful for evaluating the same model later. Also returns evaluation outputs and time

Other

Seed- seed.py contains the fixed seed

Backup Main- backup_main.ipynb is a copy main.ipynb for convenience