Add naive and vectorized implementations of Linear Regression using Gradient Descent

Author: somrita-banerjee

machine_learning/linear_regression_naive.py

@@ -0,0 +1,138 @@
+"""
+Naive implementation of Linear Regression using Gradient Descent.
+
+This version is intentionally less optimized and more verbose,
+designed for educational clarity. It shows the step-by-step
+gradient descent update and error calculation.
+
+Dataset used: CSGO dataset (ADR vs Rating)
+"""
+
+# /// script
+# requires-python = ">=3.13"
+# dependencies = [
+#     "httpx",
+#     "numpy",
+# ]
+# ///
+
+import httpx
+import numpy as np
+
+
+def collect_dataset() -> np.ndarray:
+    """Collect dataset of CSGO (ADR vs Rating)
+
+    :return: dataset as numpy matrix
+    """
+    response = httpx.get(
+        "https://raw.githubusercontent.com/yashLadha/The_Math_of_Intelligence/"
+        "master/Week1/ADRvsRating.csv",
+        timeout=10,
+    )
+    lines = response.text.splitlines()
+    data = [line.split(",") for line in lines]
+    data.pop(0)  # remove header row
+    dataset = np.matrix(data)
+    return dataset
+
+
+def run_steep_gradient_descent(
+    data_x: np.ndarray, data_y: np.ndarray, len_data: int, alpha: float, theta: np.ndarray
+) -> np.ndarray:
+    """Run one step of steep gradient descent.
+
+    :param data_x: dataset features
+    :param data_y: dataset labels
+    :param len_data: number of samples
+    :param alpha: learning rate
+    :param theta: feature vector (weights)
+
+    :return: updated theta
+
+    >>> import numpy as np
+    >>> data_x = np.array([[1, 2], [3, 4]])
+    >>> data_y = np.array([5, 6])
+    >>> len_data = len(data_x)
+    >>> alpha = 0.01
+    >>> theta = np.array([0.1, 0.2])
+    >>> run_steep_gradient_descent(data_x, data_y, len_data, alpha, theta)
+    array([0.196, 0.343])
+    """
+    prod = np.dot(theta, data_x.T)
+    prod -= data_y.T
+    grad = np.dot(prod, data_x)
+    theta = theta - (alpha / len_data) * grad
+    return theta
+
+
+def sum_of_square_error(
+    data_x: np.ndarray, data_y: np.ndarray, len_data: int, theta: np.ndarray
+) -> float:
+    """Return sum of square error for error calculation.
+
+    >>> vc_x = np.array([[1.1], [2.1], [3.1]])
+    >>> vc_y = np.array([1.2, 2.2, 3.2])
+    >>> round(sum_of_square_error(vc_x, vc_y, 3, np.array([1])), 3)
+    0.005
+    """
+    prod = np.dot(theta, data_x.T)
+    prod -= data_y.T
+    error = np.sum(np.square(prod)) / (2 * len_data)
+    return float(error)
+
+
+def run_linear_regression(data_x: np.ndarray, data_y: np.ndarray) -> np.ndarray:
+    """Run linear regression using gradient descent.
+
+    :param data_x: dataset features
+    :param data_y: dataset labels
+    :return: learned feature vector theta
+    """
+    iterations = 100000
+    alpha = 0.000155
+
+    no_features = data_x.shape[1]
+    len_data = data_x.shape[0] - 1
+
+    theta = np.zeros((1, no_features))
+
+    for i in range(iterations):
+        theta = run_steep_gradient_descent(data_x, data_y, len_data, alpha, theta)
+        error = sum_of_square_error(data_x, data_y, len_data, theta)
+        print(f"Iteration {i + 1}: Error = {error:.5f}")
+
+    return theta
+
+
+def mean_absolute_error(predicted_y: np.ndarray, original_y: np.ndarray) -> float:
+    """Return mean absolute error.
+
+    >>> predicted_y = np.array([3, -0.5, 2, 7])
+    >>> original_y = np.array([2.5, 0.0, 2, 8])
+    >>> mean_absolute_error(predicted_y, original_y)
+    0.5
+    """
+    total = sum(abs(y - predicted_y[i]) for i, y in enumerate(original_y))
+    return total / len(original_y)
+
+
+def main() -> None:
+    """Driver function."""
+    data = collect_dataset()
+
+    len_data = data.shape[0]
+    data_x = np.c_[np.ones(len_data), data[:, :-1]].astype(float)
+    data_y = data[:, -1].astype(float)
+
+    theta = run_linear_regression(data_x, data_y)
+    print("Resultant Feature vector:")
+    for value in theta.ravel():
+        print(f"{value:.5f}")
+
+
+if __name__ == "__main__":
+    import doctest
+
+    doctest.testmod()
+    main()

machine_learning/linear_regression_vectorized.py

@@ -0,0 +1,96 @@
+"""
+Vectorized implementation of Linear Regression using Gradient Descent.
+
+This version uses NumPy vectorization for efficiency.
+It is faster and cleaner than the naive version but assumes
+readers are familiar with matrix operations.
+
+Dataset used: CSGO dataset (ADR vs Rating)
+"""
+
+# /// script
+# requires-python = ">=3.13"
+# dependencies = [
+#     "httpx",
+#     "numpy",
+# ]
+# ///
+
+import httpx
+import numpy as np
+
+
+def collect_dataset() -> np.ndarray:
+    """Collect dataset of CSGO (ADR vs Rating).
+
+    :return: dataset as numpy array
+    """
+    response = httpx.get(
+        "https://raw.githubusercontent.com/yashLadha/The_Math_of_Intelligence/"
+        "master/Week1/ADRvsRating.csv",
+        timeout=10,
+    )
+    lines = response.text.splitlines()
+    data = [line.split(",") for line in lines]
+    data.pop(0)  # remove header row
+    return np.array(data, dtype=float)
+
+
+def gradient_descent(
+    x: np.ndarray, y: np.ndarray, alpha: float = 0.000155, iterations: int = 100000
+) -> np.ndarray:
+    """Run gradient descent in a fully vectorized form.
+
+    :param x: dataset features
+    :param y: dataset labels
+    :param alpha: learning rate
+    :param iterations: number of iterations
+    :return: learned feature vector theta
+    """
+    m, n = x.shape
+    theta = np.zeros((n, 1))
+
+    for i in range(iterations):
+        predictions = x @ theta
+        errors = predictions - y
+        gradients = (x.T @ errors) / m
+        theta -= alpha * gradients
+
+        if i % (iterations // 10) == 0:  # log occasionally
+            cost = np.sum(errors**2) / (2 * m)
+            print(f"Iteration {i+1}: Error = {cost:.5f}")
+
+    return theta
+
+
+def mean_absolute_error(predicted_y: np.ndarray, original_y: np.ndarray) -> float:
+    """Return mean absolute error.
+
+    >>> pred = np.array([3, -0.5, 2, 7])
+    >>> orig = np.array([2.5, 0.0, 2, 8])
+    >>> mean_absolute_error(pred, orig)
+    0.5
+    """
+    return float(np.mean(np.abs(original_y - predicted_y)))
+
+
+def main() -> None:
+    """Driver function."""
+    dataset = collect_dataset()
+
+    m = dataset.shape[0]
+    x = np.c_[np.ones(m), dataset[:, :-1]]  # add intercept term
+    y = dataset[:, -1].reshape(-1, 1)
+
+    theta = gradient_descent(x, y)
+    print("Resultant Feature vector:")
+    for value in theta.ravel():
+        print(f"{value:.5f}")
+
+
+if __name__ == "__main__":
+    import doctest
+
+    doctest.testmod()
+    main()
+