chisomdaniel / mnist-digit-classifier

MNIST handwritten digit classifier

MNIST-Digit-Classifier

This is an image classification project that classify handwritten digits using the MNIST dataset.

It is a simple project which makes use of a simple architecture in order to correctly classify handwritten digits with a high test accuracy of 97%.

Table of contents

• Project description
• Technologies used
• Setup
• Image example

Technologies

• python 3.6
• opencv-python-headless 4.5.3.56
• matplotlib 3.4.3
• numpy 1.21.2
• pillow 7.0.0
• bokeh 2.1.1
• torch 1.11.0
• torchvision 0.12.0
• tqdm 4.63.0
• ipywidgets 7.7.0
• livelossplot 0.5.4
• pytest 7.1.1
• pandas 1.3.5
• seaborn 0.11.2
• jupyter 1.0.0
• ipykernel 4.10.0
  NB: Please note that the version numbers noted here are not strict.

Setup

The project is easy to setup and run.

• First you want to ensure you have python installed on your machine.

• Secondly having anaconda setup on your system is highly recommended.

• Install the project dependencies using the requirement.txt file: (or do it through the notebook)

  pip install -r requirements.txt

• Run the notebook cells one after the other.

Here is an example of the model being used to predict the digit written in an image: