import numpy as np
from scipy.io.wavfile import read
from scipy.signal import convolve2d  # Imported for spreading the update

# 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)
        
        # --- NEW CONVOLUTIONAL UPDATE LOGIC ---
        # 1. Compress the 784x100 gradient into a compact 5x5 kernel using block averaging
        x_bins = np.linspace(0, dW1.shape[0], 6, dtype=int)
        y_bins = np.linspace(0, dW1.shape[1], 6, dtype=int)
        dW1_small = np.zeros((5, 5))
        
        for i in range(5):
            for j in range(5):
                dW1_small[i, j] = np.mean(dW1[x_bins[i]:x_bins[i+1], y_bins[j]:y_bins[j+1]])
        
        # 2. Convolve the 5x5 update kernel over self.W1. 
        # mode='same' guarantees the output retains 100% coverage at 784x100.
        dW1_spread = convolve2d(dW1, dW1_small, mode='full')[:784,:100]
        
        # 3. Apply the smoothly blended updates back to W1
        self.W1 -= self.learning_rate[0] * dW1_spread
        # --------------------------------------

        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)

# Initialize and train the MLP Classifier
if __name__ == "__main__":
    learning_rate = np.random.rand(4)
    f = MLPClassifier(input_size=784, hidden_size=100, output_size=10, learning_rate=learning_rate)

    i = 0
    while True:
        idx = np.random.randint(0, 60000, 100)
        X = X_train[idx]
        yt = y_train[idx]
        print(i, f.score(X,yt))
        f.update(X, np.eye(10)[yt])
        i += 1