การถดถอยแบบ Robust

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24

import japanize_matplotlib
import matplotlib.pyplot as plt
import numpy as np

def huber_loss(r: np.ndarray, delta: float = 1.5):
    half_sq = 0.5 * np.square(r)
    lin = delta * (np.abs(r) - 0.5 * delta)
    return np.where(np.abs(r) <= delta, half_sq, lin)

delta = 1.5
r_vals = np.arange(-2, 2, 0.01)
h_vals = huber_loss(r_vals, delta=delta)

plt.figure(figsize=(8, 6))
plt.plot(r_vals, np.square(r_vals), "red", label=r"quadratic loss $r^2$")
plt.plot(r_vals, np.abs(r_vals), "orange", label=r"absolute loss $|r|$")
plt.plot(r_vals, h_vals, "green", label=fr"Huber loss ($\delta={delta}$)")
plt.axhline(0, color="k", linewidth=0.8)
plt.grid(True, alpha=0.3)
plt.legend()
plt.xlabel("residual $r$")
plt.ylabel("loss value")
plt.title("เปรียบเทียบ quadratic / absolute / Huber loss")
plt.show()

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18

np.random.seed(42)

N = 30
x1 = np.arange(N)
x2 = np.arange(N)
X = np.c_[x1, x2]
epsilon = np.random.rand(N)
y = 5 * x1 + 10 * x2 + epsilon * 10

y[5] = 500  # แทรก outlier หนึ่งจุด

plt.figure(figsize=(8, 6))
plt.plot(x1, y, "ko", label="data")
plt.xlabel("$x_1$")
plt.ylabel("$y$")
plt.legend()
plt.title("ข้อมูลที่มี outlier")
plt.show()

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23

from sklearn.linear_model import HuberRegressor, Ridge, LinearRegression

plt.figure(figsize=(8, 6))

huber = HuberRegressor(alpha=0.0, epsilon=3.0)
huber.fit(X, y)
plt.plot(x1, huber.predict(X), "green", label="Huber regression")

ridge = Ridge(alpha=1.0, random_state=0)
ridge.fit(X, y)
plt.plot(x1, ridge.predict(X), "orange", label="Ridge (α=1.0)")

ols = LinearRegression()
ols.fit(X, y)
plt.plot(x1, ols.predict(X), "r-", label="OLS")

plt.plot(x1, y, "kx", alpha=0.7)
plt.xlabel("$x_1$")
plt.ylabel("$y$")
plt.legend()
plt.title("ผลกระทบของ outlier ต่อเส้นประมาณ")
plt.grid(alpha=0.3)
plt.show()