Artificial_Neural_Network_Library_using_NumPy from scratch

how to use ANN.py to create Artificial_Neural_Network

1. first choose network layers and neurons per layer from input unit to output unit

   • Example network =[2,50,20,3] input unit=2 ,first hidden=50 ,second hidden=20 ,output softmax layer=3

2. then create network object ANN and choose

   • Optimizer=['adam','momentum','SGD']
   • Regularization='[L2',dropout','none']
   • Activation_function=['relu','sigmoid']
   • Then choose hyperparameter
   • Learning_rate
   • Lambd, for L2 regularziation
   • Keep_prop for dropout
   • Beta for optimizer ='momentum'
   • Batch_size for minibatch
   • Example:

     net=ANN(network,iteration=300,optimizer='adam',regularization='L2',activation_function='relu',
                     learning_rate=0.1,lambd=0.2,keep_prop=0.9,beta=0.9,batch_size=64)    
     

3. Then call fit function and feed it with input and output (hot encoded)

   • loss=net.fit(X,y_hot) #fit function returns losses you can plot it
     

4. you can use to check training acuracy *

    net.predict(X,y)
   

5. to make prediction use *

   scores,zs = net.forward_prop(X) # scores[-1] is the softmax output unit