Kaggle - Digit Recognizer

Learn computer vision fundamentals with the famous MNIST data.

Introduction

Overview

The overview of the Kaggle competition can be found here: LINK

Dataset

A detailed dataset description can be found here: LINK

Model Architecture

The model architecture is based on convolutional networks, implemented in PyTorch and PyTorch-Lightning. See: ./model/classifier.py

Instructions

This is a step-by-step guide on how to replicate the Kaggle submission.

Python Environment

• Create virtual environment:

$ pipenv --python path/to/python3.7

• Activate the virtual environemt:

$ pipenv shell

• Install the required Python packages:

$ pipenv sync

Submission Script

Training the model on the provided training set and running predictions on the provided test set can be done with the use of the script ./scripts/mnist.py.

The required command-line argumets are described by running the following command:

$ python scripts/mnist.py --help

The default model configuration can be found in ./scripts/config.json.

If the train.csv and test.csv are not found inside the provided folder under the -dir_in argument, then it is necessary to provide a Kaggle API token: LINK