train.py

import csv
import json

with open("params.json", "r") as f:
    params = json.load(f)
b = float(params["bias"])
w = params["weights"] 

with open("Housing_ls.csv", "r") as f:
    dataset = [[float(x) for x in row] for row in csv.reader(f)]

n = len(dataset)
epochs = 10000
learn_rate = 1e-5

for i, row in enumerate(dataset):
    if len(row) != len(w) + 1: 
        raise ValueError(f"Row {i} has {len(row)} values, expected {len(w)+1}: {row}")

old_mse = None

for epoch in range(epochs):
    errors = []
    mse = 0.0

    for row in dataset:
        y = row[0]
        x = row[1:]  
        y_p = sum(x_i * w_i for x_i, w_i in zip(x, w)) + b
        error = y_p - y
        errors.append(error)
        mse += error ** 2

    mse /= n

    if epoch == 0:
        old_mse = mse

    grad_w = [0.0 for _ in range(len(w))]
    grad_b = 0.0

    for i, row in enumerate(dataset):
        x = row[1:]
        for j in range(len(w)):
            grad_w[j] -= 2 * errors[i] * x[j] / n
        grad_b -= 2 * errors[i] / n

    w = [w_j + learn_rate * grad_j for w_j, grad_j in zip(w, grad_w)]
    b += learn_rate * grad_b

    if epoch % 100 == 0 or epoch == epochs - 1:
        print("--------------------")
        print(f"Epoch: #{epoch}")
        print(f"MSE: {mse}")

print("\n--Training--")
print(f"Old Loss: {old_mse}")
print(f"New Loss: {mse}")
print(f"diff: {old_mse - mse}")

with open("params.json", "w") as f:
    json.dump({"bias": b, "weights": w}, f)