This repository was an exercice to use Pytorch to create 3 Neural Network models and train/test them on the Fashion MNIST dataset.
3 NN models were made:
- An easy model with a single linear layer
- A medium model with Fully-Connected Network with 2 hidden layers.
- An advanced model with a multitude of different layers to try and get the best results.
Please take in account that this code was written in a few days without any professional review/standard as an exercice.
All 3 NN models are located in "models.py". Look there if you want more information on the architecture of each model.
The file "fashion_mnist.py" contains all the code to load the data set, run the models and visualize the results with a confusion matrix.
You can run all the models with the following command:
python3 fashion_mnist.py
We run each model with for 2 epochs, a batch size of 100 and a learning rate of 0.001. The following are the accuracy results and confusion matrix for each model.
~80 percent accuracy
~83 percent accuracy
~91 percent accuracy
This project is licensed under the MIT License - see the LICENSE.md file for details