This project has been implemented from scratch & without using related frameworks like TensorFlow and PyTorch. Also, it is based on MNIST handwritten digit dataset and uses Momentum technique to increase speed and accuracy of model.
This project only comprise numpy and matplotlib as third-party libraries.

We can set hyperparameters in this part of the code:

layersLength = [784, 16, 16, 10]
batchSize = 50
numOfEpochs = 5
learningRate = 1

Also, momentum parameters can be configured in this section:

isMomentumEnabled = False
momentumTerm = []
momentumRatio = 0.7

In the end, we can start the training process by just running the NN.py code.

• t10k-images.idx3-ubyte
• t10k-labels.idx1-ubyte