main.py

import random
import numpy as np
from data import mnist_dataloader  # Import the dataloader class
from nn import *
import matplotlib.pyplot as plt

(x_train, y_train), (x_test, y_test) = mnist_dataloader.load_data()

# Normalize and preprocess data
x_train = np.array(x_train) / 255.0
x_test = np.array(x_test) / 255.0

x_train = x_train.reshape(x_train.shape[0], -1).T  # Flatten and transpose
x_test = x_test.reshape(x_test.shape[0], -1).T

# Convert labels to one-hot encoding
def one_hot_encode(labels, num_classes):
    return np.eye(num_classes)[labels].T

y_train = one_hot_encode(y_train, 10)
y_test = one_hot_encode(y_test, 10)

# Dimensions
input_neurons = x_train.shape[0]  # 784
hidden_neurons = 30
output_neurons = 10

dimensions = (input_neurons, hidden_neurons, output_neurons)


dimensions = (input_neurons, hidden_neurons, output_neurons)

# Initialize parameters
total_params = (dimensions[0] + 1) * dimensions[1] + (dimensions[1] + 1) * dimensions[2]
params = np.random.randn(total_params) * 0.01

# Hyperparameters
learning_rate = 1
epochs = 3000
batch_size = 10
rate_decay = 0.1
# Train the neural network
trained_params = sgd(x_train, y_train, params, dimensions, learning_rate, epochs)


np.save('trained_params_with_mini_batch16.npy', trained_params)
print("Parameters saved to 'trained_params_with_mini_batch16.npy'!")


# Plot the graph
plt.plot(x, y, marker='o')
plt.xlabel("epochs")
plt.ylabel("loss")
plt.title("Graph of Loss")

# Show the graph
plt.show()