This is a neural network coded from scratch that classifies the MNIST handwritten digits.
In Q1 I test the effect of different output formats, learning rates, batch sizes and ...
Q2 runs the code using the best features we tested in Q1 and it reaches 90% accuracy !!
The code in Q3 runs a model using only two layers (input and output, there is no hidden layer) on the same dataset and reaches 80% accuracy.
in order to run the code cd to the directory and run python3 main.py