Untitled

import numpy as np
import random
from sklearn.linear_model import LinearRegression
from sklearn.model_selection import KFold
np.random.seed(101)
random.seed(123)

hold_out_errors = []
loocv_errors = []
cv5_errors = []

for _ in range(100):
    n = 100
    x = np.random.uniform(-3, 3, n)
    ones = np.repeat(1, n)
    X = np.array([x, ones]).T
    y = 2*x+1 + np.random.normal(0, 1, n)
    
    # Hold out
    indices = np.array(random.sample(range(n), int(n/2)))
    reg = LinearRegression().fit(X[indices], y[indices])
    preds = reg.predict(X[~indices])
    error = np.mean((preds - y[~indices])**2)
    hold_out_errors.append(error)

    # 5-fold
    kf = KFold(n_splits=5)

    errors = []
    for train_index, test_index in kf.split(x):
        reg = LinearRegression().fit(X[train_index], y[train_index])
        preds = reg.predict(X[test_index])
        error = np.mean((preds - y[test_index])**2)
        errors.append(error)

    errors = np.array(errors)
    cv5_errors.append(errors.mean())

    # LOOCV
    kf = KFold(n_splits=n)

    errors = []
    for train_index, test_index in kf.split(x):
        reg = LinearRegression().fit(X[train_index], y[train_index])
        preds = reg.predict(X[test_index])
        error = np.mean((preds - y[test_index])**2)
        errors.append(error)

    errors = np.array(errors)
    loocv_errors.append(errors.mean())

hold_out_errors = np.array(hold_out_errors)
cv5_errors = np.array(cv5_errors)
loocv_errors = np.array(loocv_errors)

print("Hold out")
print(hold_out_errors.mean())
print(hold_out_errors.std())

print("CV5")
print(cv5_errors.mean())
print(cv5_errors.std())
print("LOOCV")
print(loocv_errors.mean())
print(loocv_errors.std())