Scikit-Learn 管道中的新功能 - 两个现有功能之间的交互

Question

我的数据集中有两个特征：高度和面积。我想通过在 scikit-learn 中使用管道交互面积和高度来创建一个新特征。

任何人都可以指导我如何实现这一目标吗？

谢谢

Answer 1

您可以使用自定义转换器实现此目的，实现适合和转换方法。 Optionnaly 你可以让它从 sklearn TransformerMixin 继承来进行 bullet-profing。

from sklearn.base import TransformerMixin

class CustomTransformer(TransformerMixin):
    def fit(self, X, y=None):
        """The fit method doesn't do much here, 
           but it still required if your pipeline
           ever need to be fit. Just returns self."""
        return self

    def transform(self, X, y=None):
        """This is where the actual transformation occurs.
           Assuming you want to compute the product of your feature
           height and area.
        """
        # Copy X to avoid mutating the original dataset
        X_ = X.copy()
        # change new_feature and right member according to your needs
        X_["new_feature"] = X_["height"] * X_["area"]
        # you then return the newly transformed dataset. It will be 
        # passed to the next step of your pipeline
        return X_

您可以使用此代码进行测试：

import pandas as pd
from sklearn.pipeline import Pipeline

# Instantiate fake DataSet, your Transformer and Pipeline
X = pd.DataFrame({"height": [10, 23, 34], "area": [345, 33, 45]})
custom = CustomTransformer()
pipeline = Pipeline([("heightxarea", custom)])

# Test it
pipeline.fit(X)
pipeline.transform(X)

对于这样一个简单的处理，这似乎有点矫枉过正，但将任何数据集操作放入 Transformers 中是一个很好的做法。这样它们的重现性更高。

Scikit-Learn 管道中的新功能 - 两个现有功能之间的交互

New Feature in Scikit-Learn Pipeline - Interaction between two existing Features

feature-engineering

scikit-learn-pipeline