scikit-learn RandomForestClassifier 中的子样本大小

Question

如何控制用于训练森林中每棵树的子样本的大小？根据 scikit-learn 的文档：

A random forest is a meta estimator that fits a number of decision tree classifiers on various sub-samples of the dataset and use averaging to improve the predictive accuracy and control over-fitting. The sub-sample size is always the same as the original input sample size but the samples are drawn with replacement if bootstrap=True (default).

所以bootstrap允许随机性但找不到如何控制子样本的数量

Answer 1

Scikit-learn 不提供此选项，但您可以使用（较慢的）版本结合树和装袋元分类器轻松获得此选项：

from sklearn.ensemble import BaggingClassifier
from sklearn.tree import DecisionTreeClassifier

clf = BaggingClassifier(base_estimator=DecisionTreeClassifier(), max_samples=0.5)

附带说明一下，Breiman 的随机森林确实没有将子样本视为参数，完全依赖于 bootstrap，因此大约 (1 - 1 / e) 个样本用于构建每棵树.

Answer 2

你实际上可以修改forest.py中的_generate_sample_indices函数来每次改变子样本的大小，感谢fastai lib 来实现一个函数 set_rf_samples 为此目的，它看起来像

def set_rf_samples(n):
    """ Changes Scikit learn's random forests to give each tree a random sample of
    n random rows.
    """
    forest._generate_sample_indices = (lambda rs, n_samples:
        forest.check_random_state(rs).randint(0, n_samples, n))

您可以将此函数添加到您的代码中

scikit-learn RandomForestClassifier 中的子样本大小

Subsample size in scikit-learn RandomForestClassifier

machine-learning

scikit-learn

random-forest

data-science