Neural network implemented in python without libraries.

Properties

• One input-layer with two perceptron
• One output-layer with one perceptron
• Three hidden-layers with four, eight and four perceptron

• Least squares as cost function

• Sigmoid function for the nonlinearity of the model

• Gradiant descent for reduce cost function

• Learning rate applied to gradient descent = 0.1

Result

Graph showing the decrease in the cost function using the gradient descent algorithm.

Divide points into two factions.