报错：'DecisionTreeClassifier'对象无'transform'属性，求解决方法

Got it, let's break down why you're hitting this error and how to fix it quickly.

Why This Error Occurs #

The core issue here is that DecisionTreeClassifier is a classification model, not a feature selection transformer. It doesn't come with a built-in transform() method. Your code tries to use it for entropy-based feature selection by chaining .fit(X, y).transform(X), but that's not what the DT classifier is designed to do.

When you fit a DecisionTreeClassifier, it does calculate feature importance scores based on entropy, but it can't directly "transform" your dataset to keep only the important features. That's a job for dedicated feature selection tools in scikit-learn.

Fixes to Resolve the Error #

We have two simple, scikit-learn-native ways to fix this, both leveraging the DT's feature importance scores for selection:

Option 1: Use SelectFromModel (Cleanest Approach) #

SelectFromModel is a transformer that wraps any estimator with a feature_importances_ attribute (like DecisionTreeClassifier) to automatically select features based on importance thresholds. Here's how to update your feature selection block:

from sklearn.feature_selection import SelectFromModel

# Replace your existing feature selection code with this
if doFeatureSelection:
    print 'Performing Feature Selection:'
    print 'Shape of dataset before feature selection: ' + str(X.shape)
    # Wrap the DT classifier in SelectFromModel
    selector = SelectFromModel(DecisionTreeClassifier(criterion='entropy'))
    X = selector.fit_transform(X, y)
    print 'Shape of dataset after feature selection: ' + str(X.shape) + '\n'

By default, this keeps features whose importance is above the mean of all importances. You can tweak the threshold with the threshold parameter (e.g., threshold='median' or a numeric value like 0.05) if you want more control over how many features are retained.

Option 2: Manually Filter Features (Full Control) #

If you want to set a custom threshold or have more visibility into which features are kept, you can extract the importance scores from the fitted DT and filter the features yourself:

if doFeatureSelection:
    print 'Performing Feature Selection:'
    print 'Shape of dataset before feature selection: ' + str(X.shape)
    clf = DecisionTreeClassifier(criterion='entropy')
    clf.fit(X, y)
    # Get feature importances and set a threshold (e.g., mean importance)
    importances = clf.feature_importances_
    threshold = importances.mean()
    # Create a mask to keep only features above the threshold
    mask = importances > threshold
    X = X[:, mask]
    print 'Shape of dataset after feature selection: ' + str(X.shape) + '\n'

This lets you adjust the threshold to be higher (fewer features) or lower (more features) based on your needs.

Quick Side Note #

If you're using Python 3, remember that print is a function, so you should use parentheses (e.g., print('Encoding feature...')) instead of the Python 2 syntax in your code. No big deal if you're still on Python 2, but it's a heads-up for future upgrades.

内容的提问来源于stack exchange，提问作者Aparna