MNIST_MultiLayer-Perceptron

A simple Python script to train a neural network with the MNIST dataset using the PyTorch framework.

Prerequisites

Bellow There are steps for installing all the prerequites to run this Python program. I recommend creating a virtual environment to avoid version conflict. The steps are done in Linux using Pip. If you are not in Linux or using Pip, check requirements.txt for a list of needed package/modules.

Install the Pytorch with:

pip install torch==1.8.1+cpu torchvision==0.9.1+cpu torchaudio==0.8.1 -f https://download.pytorch.org/whl/torch_stable.html

This will download and install PyTorch and its prerequisites.

Install the tqdm loading bar:

pip install tqdm

After this you can just run the script normally:

python script.py

What does it do?

The script will download the MNIST dataset and use it to train the parameters of a neural network, and then finally will print its precision.

Neural Network Details

This neural network is a relatively simple multilayer perceptron model.

• The input has 784 values, that are equivalent to each grayscale pixel from each of the images.
• This input gets through a linear transformation that results in the 500 values of the first hidden layer.
• These 500 values are then sent through a ReLU function (max(x,0)), resulting in 500 values of the second hidden layer.
• Finally these 500 values are sent through another linear transformation, resulting in 10 different values. Each of these values represent a different number, and the higher the value, the higher the chance of the image being that specific number.