Scikit-Multiflow报错：replace=False时样本量不能大于总体量求解决

我尝试运行以下代码，其中x是维度为(2381,)的特征向量，转换为Numpy数组后的y是维度为(1,)的标签。

from skmultiflow.meta import AdaptiveRandomForestClassifier
import numpy as np
import data

np.random.seed(1)


def main() -> None:
    dataset = data.get_full_dataset()
    metadata = data.get_metadata()
    training_batch = data.get_windows(dataset, metadata, data.get_initial_training_groups())
    streaming_batch = data.get_windows(dataset, metadata, data.get_incremental_learning_groups())
    initial_features = np.concatenate([dataset.feature_vectors for group, dataset in training_batch])
    initial_labels = np.concatenate([dataset.labels for group, dataset in training_batch])
    model = AdaptiveRandomForestClassifier()
    correct_count = 0
    n_samples = 0
    for x, y in zip(initial_features, initial_labels):
        y = np.asarray([y])
        y_prediction = model.predict(x)
        if y_prediction[0] == y:
            correct_count += 1
        model.partial_fit(x, y)
        n_samples += 1

    print(f"Accuracy: {correct_count / n_samples}")


if __name__ == "__main__":
    main()

运行后出现如下错误：

Traceback (most recent call last):
  File "/home/nathan/Documents/Research/BodmasOnline/main.py", line 31, in <module>
    main()
  File "/home/nathan/Documents/Research/BodmasOnline/main.py", line 24, in main
    model.partial_fit(x, y)
  File "/home/nathan/Documents/Research/BodmasOnline/venv/lib/python3.10/site-packages/skmultiflow/meta/adaptive_random_forests.py", line 313, in partial_fit
    self._partial_fit(X[i], y[i], self.classes, weight[i])
  File "/home/nathan/Documents/Research/BodmasOnline/venv/lib/python3.10/site-packages/skmultiflow/meta/adaptive_random_forests.py", line 328, in _partial_fit
    self.ensemble[i].partial_fit(np.asarray([X]), np.asarray([y]),
  File "/home/nathan/Documents/Research/BodmasOnline/venv/lib/python3.10/site-packages/skmultiflow/meta/adaptive_random_forests.py", line 569, in partial_fit
    self.classifier.partial_fit(X, y, classes=classes, sample_weight=sample_weight)
  File "/home/nathan/Documents/Research/BodmasOnline/venv/lib/python3.10/site-packages/skmultiflow/trees/hoeffding_tree.py", line 394, in partial_fit
    self._partial_fit(X[i], y[i], sample_weight[i])
  File "/home/nathan/Documents/Research/BodmasOnline/venv/lib/python3.10/site-packages/skmultiflow/trees/hoeffding_tree.py", line 424, in _partial_fit
    learning_node.learn_from_instance(X, y, sample_weight, self)
  File "/home/nathan/Documents/Research/BodmasOnline/venv/lib/python3.10/site-packages/skmultiflow/trees/nodes/random_learning_node_nb_adaptive.py", line 54, in learn_from_instance
    super().learn_from_instance(X, y, weight, ht)
  File "/home/nathan/Documents/Research/BodmasOnline/venv/lib/python3.10/site-packages/skmultiflow/trees/nodes/random_learning_node_classification.py", line 58, in learn_from_instance
    self.list_attributes = self._sample_features(get_dimensions(X)[1])
  File "/home/nathan/Documents/Research/BodmasOnline/venv/lib/python3.10/site-packages/skmultiflow/trees/nodes/random_learning_node_classification.py", line 72, in _sample_features
    return self.random_state.choice(
  File "mtrand.pyx", line 965, in numpy.random.mtrand.RandomState.choice
ValueError: Cannot take a larger sample than population when 'replace=False'

问题原因 #

1. 输入维度错误：partial_fit和predict期望输入是二维数组（形状为(n_samples, n_features)），但你传入的x是一维数组(2381,)，导致模型误判特征数量。
2. 特征采样参数冲突：AdaptiveRandomForestClassifier默认按sqrt(n_features)采样特征，若模型内部逻辑误判特征数（比如把一维数组当成1个特征），会出现采样数超过特征总数的情况，触发报错。
3. 未指定类别信息：流式模型需要提前知道类别集合，缺失会导致内部逻辑异常。

解决方案 #

修改后的代码 #

from skmultiflow.meta import AdaptiveRandomForestClassifier
import numpy as np
import data

np.random.seed(1)


def main() -> None:
    dataset = data.get_full_dataset()
    metadata = data.get_metadata()
    training_batch = data.get_windows(dataset, metadata, data.get_initial_training_groups())
    streaming_batch = data.get_windows(dataset, metadata, data.get_incremental_learning_groups())
    initial_features = np.concatenate([dataset.feature_vectors for group, dataset in training_batch])
    initial_labels = np.concatenate([dataset.labels for group, dataset in training_batch])
    
    # 提前获取所有类别，传入模型初始化
    classes = np.unique(initial_labels)
    # 显式设置特征采样参数，确保采样数不超过特征总数
    model = AdaptiveRandomForestClassifier(max_features='sqrt', classes=classes)
    
    correct_count = 0
    n_samples = 0
    for x, y in zip(initial_features, initial_labels):
        # 将一维特征向量转为二维数组（符合模型输入要求）
        x_2d = x.reshape(1, -1)
        y = np.asarray([y])
        y_prediction = model.predict(x_2d)
        # 修正比较逻辑，避免数组与标量直接对比
        if y_prediction[0] == y[0]:
            correct_count += 1
        model.partial_fit(x_2d, y)
        n_samples += 1

    print(f"Accuracy: {correct_count / n_samples}")


if __name__ == "__main__":
    main()

关键修改点 #

1. 输入维度转换：将一维的x通过reshape(1, -1)转为二维数组，确保模型识别正确的特征数量。
2. 显式设置模型参数：初始化时指定max_features='sqrt'（计算后约48个特征，远小于2381），避免采样数超出特征总数；同时传入classes参数，符合流式模型使用规范。
3. 修正标签比较逻辑：将y改为y[0]，避免数组与标量的直接对比问题。

内容的提问来源于stack exchange，提问作者Nate Waltz