neural network from scratch without using any library, build using only numpy arrays and maths. It's a good demonstration how neural networks works from inside.

this neural network is build for digit dataset https://www.kaggle.com/competitions/digit-recognizer

cross validation accuracy is over 90%

have only one hidden layer with 10 nodes. input layer contains 28x28 nodes, because images are 28x28 pixels. output layer have only 10 nodes because their are only 10 node to predict.

have only 2 dependencies:

• pandas: for loading data
• numpy: for array calculation