In this notebook I've implemeneted a fully paramtrized neural network from scratch using only numpy library.
To follow this notebook you have to install numpy, pandas and matplotlib:
pip install numpy pandas matplotlib
or if you're using Anaconda Distribution
conda install numpy pandas matplotlib
The data can be downloaded from Kaggle.
In this notebook I've implemented forward and backpropagation algorithms using the partial derivatives. For updating weight i've used gradient descent algortihms.
Note that you can try the network with different number of hidden layers, nodes and activation function. However the last layer will always be softamx because I was trying to solve a classification problem and I've used softmax with cross-entropy loss.
I implemented SGD optimizer as well and you can change the batch size each time.
Some results:
