A 3-layer neural network for MNIST using keras
The jupyter notebook contains my submission for the Digit Recognizer Challenge on Kaggle the same neural network. The neural network got a score of 0.97071 on 1 and the submission was ranked 1722 out of 2459 on a rolling scoreboard.
- Input layer
- Layer with
392
neurons [relu activation] - Layer with
196
neurons [relu activation] - Output layer with
10
neurons [softmax activation]
This model achieves 98.5% test accuracy in 40 epochs using rmsprop
optimizer
Download MNIST dataset in a folder and name it 'mnist'