import numpy as np
from scipy.io.wavfile import read

# Load training and test data
X_train = read('../X_train.wav')[1].reshape(-1, 784)
y_train = (read('../y_train.wav')[1] * 9).astype(int)
X_test = read('../X_test.wav')[1].reshape(-1, 784)
y_test = (read('../y_test.wav')[1] * 9).astype(int)
X20 = X_test[:1000]
yt20 = y_test[:1000]

# Define the MLP Classifier
class MLPClassifier:
    def __init__(self, input_size, hidden_size, output_size, learning_rate=np.random.rand(4)):
        self.W1 = np.random.randn(input_size, hidden_size) * 0.01
        self.b1 = np.zeros((1, hidden_size))
        self.W2 = np.random.randn(hidden_size, output_size) * 0.01
        self.b2 = np.zeros((1, output_size))
        self.learning_rate = learning_rate
    
    def relu(self, x):
        return np.maximum(0, x)
    
    def relu_derivative(self, x):
        return np.where(x > 0, 1, 0)
    
    def softmax(self, x):
        exp_x = np.exp(x - np.max(x, axis=1, keepdims=True))
        return exp_x / np.sum(exp_x, axis=1, keepdims=True)
    
    def forward(self, X):
        self.z1 = np.dot(X, self.W1) + self.b1
        self.a1 = self.relu(self.z1)
        self.z2 = np.dot(self.a1, self.W2) + self.b2
        output = self.softmax(self.z2)
        return output
    
    def compute_loss(self, y_true, y_pred):
        m = y_true.shape[0]
        loss = -np.sum(y_true * np.log(y_pred + 1e-9)) / m
        return loss
    
    def backward(self, X, y_true, y_pred):
        m = y_true.shape[0]
        dz2 = y_pred - y_true
        dW2 = np.dot(self.a1.T, dz2) / m
        db2 = np.sum(dz2, axis=0, keepdims=True) / m
        da1 = np.dot(dz2, self.W2.T)
        dz1 = da1 * self.relu_derivative(self.z1)
        dW1 = np.dot(X.T, dz1) / m
        db1 = np.sum(dz1, axis=0, keepdims=True) / m
        return dW1, db1, dW2, db2
    
    def update(self, X, y_true):
        y_pred = self.forward(X)
        dW1, db1, dW2, db2 = self.backward(X, y_true, y_pred)
        self.W1 -= self.learning_rate[0] * dW1
        self.b1 -= self.learning_rate[1] * db1
        self.W2 -= self.learning_rate[2] * dW2
        self.b2 -= self.learning_rate[3] * db2
    
    def predict(self, X):
        probabilities = self.forward(X)
        return np.argmax(probabilities, axis=1)

    def score(self, X, y_true):
        y = self.predict(X)
        return np.mean(y == y_true)


def create_relative_features(X, block_size=7):
    """Create features as differences between neighboring pixels"""
    n_samples = X.shape[0]
    img = X.reshape(-1, 28, 28)
    
    # Horizontal differences
    h_diff = img[:, :, 1:] - img[:, :, :-1]
    # Vertical differences
    v_diff = img[:, 1:, :] - img[:, :-1, :]
    
    # Diagonal differences
    d1_diff = img[:, 1:, 1:] - img[:, :-1, :-1]
    d2_diff = img[:, 1:, :-1] - img[:, :-1, 1:]
    
    # Local mean differences (contrast)
    kernel = np.ones((block_size, block_size)) / (block_size ** 2)
    local_mean = np.zeros_like(img)
    for i in range(n_samples):
        from scipy.ndimage import uniform_filter
        local_mean[i] = uniform_filter(img[i], size=block_size)
    
    contrast = img - local_mean
    
    # Combine all features
    features = np.concatenate([
        h_diff.reshape(n_samples, -1),
        v_diff.reshape(n_samples, -1),
        d1_diff.reshape(n_samples, -1),
        d2_diff.reshape(n_samples, -1),
        contrast.reshape(n_samples, -1)
    ], axis=1)
    
    return features

def create_zscore_features(X):
    """Normalize each image to have zero mean, unit variance"""
    mean = X.mean(axis=1, keepdims=True)
    std = X.std(axis=1, keepdims=True) + 1e-8
    return (X - mean) / std
    
# Initialize and train the MLP Classifier
learning_rate = np.random.rand(4)
f = MLPClassifier(input_size=3754, hidden_size=100, output_size=10, learning_rate=learning_rate)
X20 = create_zscore_features(create_relative_features(X20))

i = 0
while True:
    idx = np.random.randint(0, 60000, 100)
    X = create_zscore_features(create_relative_features(X_train[idx]))
    yt = y_train[idx]
    print(i, f.score(X20,yt20))
    for _ in range(2):
        f.update(X, np.eye(10)[yt])
    i += 1