mnist-classifier 1.0.1

Basic mnist classifier example of a Reproducible Research Project in Python

Classifying digits using 28x28px images in one of 10 classes

Small classifier for 28x28px handwritten digits based on M-NIST dataset

License

MIT

Requirements

For this project to run properly you will need:

• conda (installation instructions)

Installation

To use and reproduce this project, first clone this repository in the directory of your choice

cd /path/to/your/directory
git clone https://github.com/sandrich/classifying_digits_mnist.git

Then create a conda environment with the correct dependencies:

conda env create --file environment.yml

Once the conda has finished installing all the dependencies, activate it:

conda activate mnist_classifier

Usage

The program can run without parameters which will take our researched value. Feel free to use different parameters to play with the data and algorithm

$ python mnist_predict.py -h
usage: mnist_predict.py [-h] [--trees TREES] [--depth DEPTH] [--impurity_method {entropy,gini}]

Run MNIST classifier

optional arguments:
  -h, --help            show this help message and exit
  --trees TREES         Number of trees
  --depth DEPTH         Maximum tree depth
  --impurity_method {entropy,gini}
                        Impurity method

Example

# python mnist_predict.py 
No local fit dataset found.
Downloading fit data
['================================================='>'']]
Downloading fit labels
['================================================='>'']
No local test dataset found.
Downloading test data
['================================================='>'']]
Downloading test labels
['================================================='>'']
Starting training...
Done training.
Predicting...
Predicting...
Classification stats:
-----------------
Max tree depth: 9
Number of trees: 20
Impurity method: entropy
-----------------
Train Accuracy: 0.946
Train Accuracy: 0.935

Authors

@sandrich - Christian Sandrini @bigskapinsky - Calixte Mayoraz