So...
This my neural network to predict numbers by MINIST dataset.
To know how to launch and see how it work, check next.
- Input layer - 784(There are pixels of photo)
- First hidden layer - 128
- Second hidden layer - 64
- Output layer - 10
Sigmoid: 1/(1 + np.exp(-x))
And of course softmax: (this function doesn`t have graph) np.exp(x - x.max()) / np.sum(exps, axis=0)
Sigmoid is used in two layers (input and first hidden)
You must write to console: python main.py
And the algorithms (backpropagation) will start working.
Also I have first(bad) version who doesn`t work.
If you want you can fix it)