import numpy as np
from sklearn.datasets import make_blobs

def make_blocks(n_samples=20, centers=2, n_features=4):  # Change n_features to 4
    x, y = make_blobs(n_samples=n_samples, centers=centers, n_features=n_features, random_state=42)
    y = y[:, np.newaxis]
    return x, y

def sigmoid(x):
    return 1 / (1 + np.exp(-x))

def sigmoid_derivative(x):
    return x * (1 - x)

input_neurons = 4
hidden_neurons1 = 3
hidden_neurons2 = 2
output_neurons = 1

np.random.seed(42)

weights_input_hidden1 = np.random.rand(input_neurons, hidden_neurons1)
bias_hidden1 = np.zeros((1, hidden_neurons1))

weights_hidden1_hidden2 = np.random.rand(hidden_neurons1, hidden_neurons2)
bias_hidden2 = np.zeros((1, hidden_neurons2))

weights_hidden2_output = np.random.rand(hidden_neurons2, output_neurons)
bias_output = np.zeros((1, output_neurons))


learning_rate = 0.001
epochs = 100


x, y = make_blocks(n_samples=20, centers=2, n_features=4)

for epoch in range(epochs):
    # Forward pass
    hidden_layer_input1 = np.dot(x, weights_input_hidden1) + bias_hidden1
    hidden_layer_output1 = sigmoid(hidden_layer_input1)

    hidden_layer_input2 = np.dot(hidden_layer_output1, weights_hidden1_hidden2) + bias_hidden2
    hidden_layer_output2 = sigmoid(hidden_layer_input2)

    output_layer_input = np.dot(hidden_layer_output2, weights_hidden2_output) + bias_output
    predicted_output = sigmoid(output_layer_input)

    error = y - predicted_output

    output_error = error * sigmoid_derivative(predicted_output)
    hidden_layer_error2 = output_error.dot(weights_hidden2_output.T) * sigmoid_derivative(hidden_layer_output2)
    hidden_layer_error1 = hidden_layer_error2.dot(weights_hidden1_hidden2.T) * sigmoid_derivative(hidden_layer_output1)

    weights_hidden2_output += hidden_layer_output2.T.dot(output_error) * learning_rate
    bias_output += np.sum(output_error, axis=0, keepdims=True) * learning_rate

    weights_hidden1_hidden2 += hidden_layer_output1.T.dot(hidden_layer_error2) * learning_rate
    bias_hidden2 += np.sum(hidden_layer_error2, axis=0, keepdims=True) * learning_rate

    weights_input_hidden1 += x.T.dot(hidden_layer_error1) * learning_rate
    bias_hidden1 += np.sum(hidden_layer_error1, axis=0, keepdims=True) * learning_rate

hidden_layer_input1 = np.dot(x, weights_input_hidden1) + bias_hidden1
hidden_layer_output1 = sigmoid(hidden_layer_input1)

hidden_layer_input2 = np.dot(hidden_layer_output1, weights_hidden1_hidden2) + bias_hidden2
hidden_layer_output2 = sigmoid(hidden_layer_input2)

output_layer_input = np.dot(hidden_layer_output2, weights_hidden2_output) + bias_output
predicted_output = sigmoid(output_layer_input)

predicted_output = np.round(predicted_output)

accuracy = np.mean(predicted_output == y)

print("Predicted Output:")
print(predicted_output)
print("\nTrue Output:")
print(y)
print("\nAccuracy: {:.2%}".format(accuracy))