This repository contains an implementation of a small Neural Network in Python, with the ability to specify the number of hidden layers, the number of neurons in each layer, and the activation function, and the optimizer to be used.

pip install miniMLP

The code requires the following dependencies:

• NumPy
• matplotlib (for visualization)

Create an instance of the MLP class, specifying the input size, output size, number of hidden layers, sizes of each hidden layer, and activation function.

import numpy as np
from miniMLP.engine import MLP

mlp = MLP(input_size=2, output_size=1, hidden_layers=3, hidden_sizes=[4, 6, 4], activation='relu')

Then, train the MLP using the train method, providing the training data (an array of inputs and corresponding labels), and optional parameters such as the learning rate, number of epochs, and optimizer.

X_train = np.array([[0, 0], [0, 1], [1, 0], [1, 1]])
y_train = np.array([[0], [1], [1], [0]])

mlp.train(X_train, y_train, epochs=2000, learning_rate=0.0001, optimizer='Adam')

Finally, use the predict method to obtain predictions for new data points.

y_pred = mlp.predict(X_new)

The following activation functions are currently supported:

• Sigmoid
• ReLU
• Tanh
• Softmax
• Leaky ReLU
• ELU
• GELU
• Softplus
• SeLU
• PReLu
• Swish
• Gaussian

The following optimizers are currently supported:

• Adam
• Stochastic Gradient Descent
• Momentum

This project is licensed under the MIT License. Feel free to use and modify this code for your own projects.