mnist_using_pytorch

MNIST digit recognizer Neural Net using Pytorch.

Details

Although Pytorch contains MNIST dataset, but I wanted to write my own custom Dataset and hence loaded the MNIST dataset from Kaggle into Pytorch using Dataset class.

Version 1 (v1) MNIST digit recognizer using 3 Fully connected layers in Pytorch and LogSoftmax as the final layer

Here I tried to build MNIST digit recognizer with a simple 3-fully-connected layer neural network written using Pytorch. Reached 91% accuracy.

Version 2 (v2) MNIST digit recognizer uses 2 Convolution layers, a dropout layer, and 2 Fully connected layers in Pytorch and LogSoftmax as the final layer

Here I tried to participate in Kaggle's MNIST challenge, my network is not quite as accurate compared to others on leaderboard, but I managed to get about 94% accuracy and a decent place on leaderboard.

Kaggle Notebook link: https://www.kaggle.com/code/tas820/mnist-using-deep-neural-net-pytorch

Dataset

Used from Kaggle