def fit():
x, y = sample(5)
poly = polyreg.PolynomialRegression(4)
poly.fit(x, y)
lin = linearreg.LinearRegression()
lin.fit(x, y)
return poly, lin
import numpy as np
import matplotlib.pyplot as plt
# データ生成
np.random.seed(0)
def f(x):
return 1 + 2 * x
x = np.random.random(10) * 10
y = f(x) + np.random.randn(10)
# 多項式回帰
model = polyreg.PolynomialRegression(10)
model.fit(x, y)
plt.scatter(x, y, color="k")
plt.ylim([y.min() - 1, y.max() + 1])
xx = np.linspace(x.min(), x.max(), 300)
yy = np.array([model.predict(u) for u in xx])
plt.plot(xx, yy, color="k")
# 線形回帰
model = linearreg.LinearRegression()
model.fit(x, y)
b, a = model.w_
x1 = x.min() - 1
x2 = x.max() + 1
plt.plot([x1, x2], [f(x1), f(x2)], color="k", linestyle="dashed")
return x, y
xx = np.arange(0, 5, 0.01)
np.random.seed(0)
y_poly_sum = np.zeros(len(xx))
y_poly_sum_sq = np.zeros(len(xx))
y_lin_sum = np.zeros(len(xx))
y_lin_sum_sq = np.zeros(len(xx))
y_true = f(xx)
n = 100000
warnings.filterwarnings("ignore")
for _ in range(n):
x, y = sample(5)
poly = polyreg.PolynomialRegression(4)
poly.fit(x, y)
lin = linearreg.LinearRegression()
lin.fit(x, y)
y_poly = poly.predict(xx)
y_poly_sum += y_poly
y_poly_sum_sq += (y_poly - y_true)**2
y_lin = lin.predict(xx)
y_lin_sum += y_lin
y_lin_sum_sq += (y_lin - y_true)**2
fig = plt.figure()
ax1 = fig.add_subplot(121)
ax2 = fig.add_subplot(122)
ax1.set_title("Linear reg.")
ax2.set_title("Polynomial reg.")
xx = np.arange(0, 5, 0.01)
np.random.seed(0)
y_poly_sum = np.zeros(len(xx))
y_poly_sum_sq = np.zeros(len(xx))
y_lin_sum = np.zeros(len(xx))
y_lin_sum_sq = np.zeros(len(xx))
y_true = f(xx)
n = 1000
warnings.filterwarnings("ignore")
for _ in range(n):
# 1. 適当な5点をとる
x, y = sample(5)
# 多項式回帰学習
poly = polyreg.PolynomialRegression(4) # 多項式の次数(5点だから4次関数)
poly.fit(x, y)
# 線形回帰学習
lin = linearreg.LinearRegression()
lin.fit(x, y)
# 多項式回帰予測(0~5の全ての点で)
y_poly = poly.predict(xx)
# 多項式回帰予測：結果を用意した配列に格納してやる
y_poly_sum += y_poly
# 真の値と予測値の差の2乗を配列に格納
y_poly_sum_sq += (y_poly - y_true)**2
# 線形回帰予測
y_lin = lin.predict(xx.reshape(-1, 1))
y_lin_sum += y_lin
# 真の値と予測値の差の2乗を配列に格納
y_lin_sum_sq += (y_lin - y_true)**2