lipeeeee / neuralnetwork_scratch

Neural Network without Tensorflow

Neural Network with math and numpy

No tensorflow and/or pytorch, pure numpy and math equations using the MNIST dataset.

Model

1. Activation Function: ReLU and ReLU_deriv
2. Last Activation Function: softmax
3. Propagation: Foward and Backwards
4. Encoding: one hot encoding
5. Optimization: Gradient descent

Iteration 500

Model had a 87% accuracy on Iteration 500

Iteration:  490
[2 7 7 ... 8 6 1] [2 7 9 ... 8 6 1]
0.8702439024390244

neuralnetwork_scratch's Issues

Gradient descent alpha misnumbered