随机森林特征重要性的交叉验证数据集折叠

Question

我正在尝试使用交叉验证折叠来生成随机森林的特征重要性图。当只使用特征(X)和目标(Y)数据时，实现很简单，例如：

rfc = RandomForestClassifier()
rfc.fit(X, y)
importances = pd.DataFrame({'FEATURE':data_x.columns,'IMPORTANCE':np.round(rfc.feature_importances_,3)})
importances = importances.sort_values('IMPORTANCE',ascending=False).set_index('FEATURE')

print(importances)
importances.plot.bar()
plt.show()

这会产生：

​

然而，我如何转换这段代码，以便为我将要创建的每个交叉验证折叠(k-折叠图)创建类似的图？

我现在拥有的代码是：

# Empty list storage to collect all results for displaying as plots
mylist = []

from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import KFold
kf = KFold(n_splits=3)
for train, test in kf.split(X, y):
    train_data = np.array(X)[train]
    test_data = np.array(y)[test]
for rfc = RandomForestClassifier():
    rfc.fit(train_data, test_data)

例如，上面的代码使用交叉验证技术创建(3个折叠)，我的目标是为所有3个折叠创建特征重要性图，从而生成3个特征重要性图。目前，它给了我循环错误。

我不确定使用创建的每个(k-折叠)分别通过随机森林为每个(k-折叠)生成特征重要性图的最有效的技术是什么。

Answer 1

这是为我工作的代码：

import numpy as np
import matplotlib.pyplot as plt
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import KFold
from sklearn.datasets import make_classification

# classification dataset
data_x, data_y = make_classification(n_features=9)

# feature names must be declared outside the function
# feature_names = list(data_x.columns)

kf = KFold(n_splits=10)
rfc = RandomForestClassifier()
count = 1
# test data is not needed for fitting
for train, _ in kf.split(data_x, data_y):
    rfc.fit(data_x[train, :], data_y[train])
    # sort the feature index by importance score in descending order
    importances_index_desc = np.argsort(rfc.feature_importances_)[::-1]
    feature_labels = [feature_names[-i] for i in importances_index_desc]

    # plot
    plt.figure()
    plt.bar(feature_labels, rfc.feature_importances_[importances_index_desc])
    plt.xticks(feature_labels, rotation='vertical')
    plt.ylabel('Importance')
    plt.xlabel('Features')
    plt.title('Fold {}'.format(count))
    count = count + 1
plt.show()

​

​

Answer 2

错误的原因之一是代码rfc.fit(train_data, test_data)。你应该把训练标签作为第二个参数，而不是测试数据。

至于绘图，您可以尝试执行类似以下代码的操作。我假设你知道，在这种情况下，k-折叠式CV仅用于选择不同的训练数据集。测试数据被忽略，因为没有做出预测：

import numpy as np
import matplotlib.pyplot as plt
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import KFold
from sklearn.datasets import make_classification

# dummy classification dataset
X, y = make_classification(n_features=10)
# dummy feature names
feature_names = ['F{}'.format(i) for i in range(X.shape[1])]

kf = KFold(n_splits=3)
rfc = RandomForestClassifier()
count = 1
# test data is not needed for fitting
for train, _ in kf.split(X, y):
    rfc.fit(X[train, :], y[train])
    # sort the feature index by importance score in descending order
    importances_index_desc = np.argsort(rfc.feature_importances_)[::-1]
    feature_labels = [feature_names[i] for i in importances_index_desc]
    # plot
    plt.figure()
    plt.bar(feature_labels, rfc.feature_importances_[importances_index_desc])
    plt.xticks(feature_labels, rotation='vertical')
    plt.ylabel('Importance')
    plt.xlabel('Features')
    plt.title('Fold {}'.format(count))
    count = count + 1
plt.show()