Add Radial Basis Function Neural Network (RBFNN)

[JeninaAngelin]

Feature description

Radial Basis Function Neural Networks (RBFNNs) are a type of neural network that combines elements of clustering and function approximation, making them powerful for both regression and classification tasks. RBFNNs can efficiently model non-linear relationships with fewer parameters than traditional multilayer perceptrons, thanks to their unique architecture that uses radial basis functions (typically Gaussian) in the hidden layer.

This structure allows RBFNNs to approximate functions and decision boundaries with high accuracy while maintaining a relatively simple network structure.

Goals:

Implement an RBFNN class with train and predict functionalities.

Use Gaussian radial basis functions as hidden layer activations.

Include KMeans clustering for initializing RBF centers and least-squares fitting for output weights.

Requirements:

Initialization: RBFNN class should initialize with num_centers and gamma for Gaussian spread.

Training: Add a train(x_data, y_data) method to:

Find RBF centers using KMeans.

Compute RBF activations for each input.

Calculate output weights through least-squares fitting.

Prediction: Implement a predict(x) method to calculate and return predictions based on learned weights.

[Tahmid019]

I would like to contibute

[JeninaAngelin]

I've submitted a PR but some tests are not passed