NeuralNetworkFromScratch

A simple python class for neural network from scratch

Hi everyone

It's a simple python code which contains a MLP neural network and wrote from scratch. It's training algorithm is Marquart-Levenbeurg method.

How to use

1. Download and copy beside your main python code

2. import it like below:

from NN import NeuralNetwork

3. Make a NN in your code

myNN = NeuralNetwork(
              number_of_inputs= place number of inputs here,
              number_of_hidden_neurons= [h1,h2,....], # place each layer number of neurons as hi e.g [5,5]
              number_of_outputs= place number of inputs here, 
              #activation_function=lambda x: 2/(1+numpy.exp(-2*x))-1, # you can set any activation function you like
              #activation_function_derivation=lambda x: 1+(2/(1+numpy.exp(-2*x))-1)**2, # wrote down activation function derivation
              max_epoch=100, # number of epochs, default is 1000
              #log=True, #if you want see some log in training
              )

4. Train NN

"Inputs" are inputs data for neural network and "Outputs" are target neural network must reach as it output for each input

NN.Train(Inputs=Inputs, Outputs=Outputs, learning_factor=1e+10, thershold=1e-10)

There is an example in main part of code if you need

Good luck!

P.S.

you can use google colab and numba or cupy to speed up running