NN example implementation
The layout is a 2 hidden layer fully connected NN + an output layer of size 10
This repo contains the pre-trained parameters stored in params/
to run the model: Extract the training set (train.csv.zip) into train.csv
then execute:
python main.py
It has currently around 85% success rate ( the original had around 84%, big gains)
Inspired by (almost all code): Simple MNIST NN from scratch (numpy, no TF/Keras) | Kaggle