A small project comparing various machine learning techniques for recognizing handwritten digits, using the MNIST dataset.
Install this project using poetry
petry install
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
This package offers several executable scripts:
fit- train all models and save them in the./modelsdirectoryscore- calculate accuracy of the modelsconfusion_matrix- calculate confusion matrices for the modelsapp- 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]
TODO
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)