การมอง Gradient Boosting แบบทีละขั้น | ดูว่าแต่ละต้นช่วยอะไร

import numpy as np
import matplotlib.pyplot as plt
import japanize_matplotlib
from sklearn.ensemble import GradientBoostingRegressor

n_samples = 500
X = np.linspace(-10, 10, n_samples)[:, np.newaxis]
noise = np.random.rand(X.shape[0]) * 10
y = (np.sin(X).ravel()) * 10 + 10 + noise

n_estimators = 10
learning_rate = 0.5
reg = GradientBoostingRegressor(
    n_estimators=n_estimators,
    learning_rate=learning_rate,
)
reg.fit(X, y)

y_pred = reg.predict(X)

plt.figure(figsize=(20, 10))
plt.scatter(X, y, c="k", marker="x", label="train")
plt.plot(X, y_pred, c="r", label="final prediction")
plt.axhline(y=np.mean(y), color="gray", linestyle=":", label="initial model")
plt.legend(); plt.show()

fig, ax = plt.subplots(figsize=(20, 10))
temp = np.zeros(n_samples) + np.mean(y)

for i in range(n_estimators):
    res = reg.estimators_[i][0].predict(X) * learning_rate
    ax.bar(X.flatten(), res, bottom=temp, label=f"tree {i+1}", alpha=0.05)
    temp += res

plt.scatter(X.flatten(), y, c="k", marker="x", label="train")
plt.plot(X, y_pred, c="r", linewidth=1, label="final prediction")
plt.legend(); plt.show()

for i in range(5):
    fig, ax = plt.subplots(figsize=(20, 10))
    plt.title(f"หลังเพิ่ม {i+1} ต้น")
    temp = np.zeros(n_samples) + np.mean(y)

    for j in range(i + 1):
        res = reg.estimators_[j][0].predict(X) * learning_rate
        ax.bar(X.flatten(), res, bottom=temp, alpha=0.05)
        temp += res

    plt.scatter(X.flatten(), y, c="k", marker="x")
    plt.plot(X, temp, c="r", linewidth=1.2, label="prediction")
    plt.legend(); plt.show()