Neural Networks Models in Python

Demonstration of several neural networks with the use of Python (Without Third-Party Frameworks)

ANN (Back Propagation Neural Network)

Architecture of this package:

This back-propagation (ANN) implementation is designed to handle two inputs and produce a single output.

The width (number of nodes) of the single hidden layer can be configured accordingly.

Specifications:

The example demo of this ANN is to predict the function output of: $$ 4\cdot(5\sin{x}) + 2y^2 $$ With a range from 1 to 10 for both inputs.

Prediction of different target functions can also be made available provided that the updated training data is feed correctly to the model.

Requried packages include:

• Numpy
• Pandas
• csv_writer

Normalization used in this specific implementation is the min-max method.

Code Architecture:

Several functions declared in the Generic_Neural_Network class are depicted below:

forward(self, X):
//Take in X and progpagate through the network in a feed-forward fashion

backward(self, X, y, o): 
//Calculate deltas and adjust weights.

sigmoid(self, s): 
//Calculate the sigmoid value of parameter s.

sigmoidPrime(self, s): 
//Calculate the derivative of the sigmoid function with parameter s.

train(self, X, y): 
//Define the flow of the ANN network.

For more info, check the Word document in the same directory for details on function declaration and data processing methods.

SOM (Self organizing map)

LVQ (Learning vector quantization)

RBF (Radial Basis Functions)

ART (Adaptive resonance theory)

CPN (Counter propagation network)