在执行这段代码时：

import matplotlib.pyplot as plt
import numpy as np

from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import LearningCurveDisplay, train_test_split

from imblearn.datasets import fetch_datasets


def load_dataset(dataset):
    data = fetch_datasets()[dataset]
    data.target = np.where(data.target < 0, 0, 1)
    return data.data, data.target

X, y = load_dataset("webpage")

X_train, X_test, y_train, y_test = train_test_split(
  X,
  y,
  test_size=0.3,
  random_state=0,
)

model = RandomForestClassifier(max_depth=3, random_state=0)

train_sizes = [0.55, 0.78, 1.0]

LearningCurveDisplay.from_estimator(
    model,
    X, y,
    train_sizes=train_sizes,
    cv=10,
    scoring='average_precision',
    line_kw = {"marker": "o"},
)

我得到以下错误：

ValueError: pos_label=1 is not a valid label: It should be one of [0]

我知道这是因为数据集不平衡，可能在验证折中没有足够的少数类样本？不过我随后尝试了带cv=5或 cv=10的分层交叉验证，结果仍然是同样的错误，因此我有点困惑。

它到底从哪里来，如何防止它？

解决方案

在 LearningCurveDisplay.from_estimator 调用中直接添加 shuffle=True 和一个 random_state：

LearningCurveDisplay.from_estimator(
    model,
    X, y,
    train_sizes=train_sizes,
    cv=10,
    scoring='average_precision',
    shuffle=True,          # <--- shuffles the training folds before slicing
    random_state=0,        # <--- Guarantees reproducibility of the shuffles
    line_kw={"marker": "o"},
)

另外，在你的代码中：

# Change this:
LearningCurveDisplay.from_estimator(model, X, y, ...)

# To this:
LearningCurveDisplay.from_estimator(model, X_train, y_train, ...)