"""
'mlp_demo.py'

Classify the XOR data pattern 

David Pan, UAH
"""

import numpy as np


x1 = np.array([1, 1])
x2 = np.array([-1, -1])
x3 = np.array([-1, 1])
x4 = np.array([1, -1])
X = np.vstack([x1, x2, x3, x4]) # Features are along the row

y1 = np.array([1, 0])
y2 = np.array([1, 0])
y3 = np.array([0, 1])
y4 = np.array([0, 1])
y = np.vstack([y1, y2, y3, y4])

from sklearn.neural_network import MLPClassifier
clf = MLPClassifier(
    hidden_layer_sizes = (2), 
    activation='logistic', 
    random_state= 100, 
    alpha = 0.001, 
    solver='lbfgs', 
    max_iter=10000000
    )

clf.fit(X, y)

clf.n_layers_
clf.loss_
clf.predict(X)

# Weight matrix for the hidden layer
clf.coefs_[0]
# Bias vector for the hidden layer
clf.intercepts_[0]
# Output layer weights and biases
clf.coefs_[1]
clf.intercepts_[1]

# Verification of the final output 

Z2 = np.matmul(X,clf.coefs_[0]) + clf.intercepts_[0]
A2 = 1/(1 + np.exp(-Z2))

Z3 = np.matmul(A2,clf.coefs_[1]) + clf.intercepts_[1]
A3 = 1/(1 + np.exp(-Z3))

A3
clf.predict_proba(X)