MNIST_digit_recognition

A small project comparing various machine learning techniques for recognizing handwritten digits, using the MNIST dataset.

Installation

Install this project using poetry

petry install

Training data

This project is based on this exercise from Kaggle.com

To run this project, you need to download the data (train.csv) from this exercise and put it under ./data directory

Commands

This package offers several executable scripts:

• fit - train all models and save them in the ./models directory
• score - calculate accuracy of the models
• confusion_matrix - calculate confusion matrices for the models
• app - run the drawing app allowing to test the models

The main script executes these steps in sequence

To run the script using poetry run:

poetry run [script]

App

The drawing app allows the user to select a model, draw a digit and make a prediction with selected model.

Run the app with poetry run app

After running the app, select a model with the Model dropdown, draw a digit and hit Predict button to use the model to predict the digit. You can update the drawing after the prediction or select a different model to make another prediction.

Use the Clear button to clear the canvas and the prediction

Available models

Currently supported models are:

• Naive Bayes (from sklearn)
• Decision Tree (from sklearn)
• Random Forest (from sklearn)
• K Nearest Neighbours (k=5, from sklearn)
• Support Vector (from sklearn)