This repo is a model trained to predict images of hand-written digits contained in the MNIST database.It is a basic code designed to be a self-project based implemented using PyTorch and numpy. Using logistic regression the accuracy achieved is 87-89% with 10-20 epochs and upto 92% with larger epochs(this process can be speeded up if you simply use a neural network with a hidden layer but without any activation function).
However using a Neural network with hidden layer of size 32 and activation function ReLU an accuracy of upto 97% is achieved.The accuracy can be increased even further using complex learning techniques like CNN.
References :-