import numpy as np
import os

from activations import relu, sigmoid

# Fix the seed for reproducibility
np.random.seed(10)

class Layer:
    def __init__(self, n_inputs, n_neurons, activation_function):
        self.inputs = np.zeros((n_inputs, 1))
        self.outputs = np.zeros((n_neurons, 1))
        # 1. Initialize the weight matrix and the bias vector with random values
        self.weights = np.random.uniform(-1, 1, (n_neurons, n_inputs))
        self.biases  = np.random.uniform(-1, 1, (n_neurons, 1))
        self.activation = activation_function
    
    def forward(self, inputs):
        self.inputs = np.array(inputs).reshape(-1, 1)
        # 2. Compute the raw output values of the neurons
        self.outputs = np.dot(self.weights, self.inputs) + self.biases
        # 3. Apply the activation function
        return self.activation(self.outputs)

class Perceptron:
    def __init__(self, layers):
        self.layers = layers

input_size  = 2
hidden_size = 6
output_size = 1


# 4. Define three layers: 2 hidden layers and 1 output layer
hidden_1     = Layer(input_size,  hidden_size, relu)
hidden_2     = Layer(hidden_size, hidden_size, relu)
output_layer = Layer(hidden_size, output_size, sigmoid)

layers     = [hidden_1, hidden_2, output_layer]
perceptron = Perceptron(layers)

print("Weights of the third neuron in the second hidden layer:")
print(np.round(perceptron.layers[1].weights[2], 2))

print("Weights of the neuron in the output layer:")
print(np.round(perceptron.layers[2].weights[0], 2))

'''
How it works:
# The perceptron consists of multiple layers, each with its own weights and biases. The forward method computes the output of each
# layer by applying the activation function to the weighted sum of inputs. The weights and biases are initialized randomly, and 
# the perceptron can be used to process inputs through its layers.
# The output of the second hidden layer and the output layer can be accessed through the perceptron's layers attribute.
# The weights of the third neuron in the second hidden layer and the output layer are printed, showing how the perceptron is structured.

'''