如何无需转换为数值，使用字符串特征与标签训练决策树分类器？

Awesome question—you don't need to manually map those string features and labels to numbers anymore! Scikit-learn has built-in tools that handle this automatically, making your code cleaner and less error-prone. Here's how to implement it:

Step 1: Understand the Tools We'll Use #

• ColumnTransformer: Lets us apply different preprocessing steps to specific columns (critical here, since we have one numeric column and one categorical string column).
• OrdinalEncoder: Converts categorical string values (like "smooth"/"bumpy") into numerical values consistently.
• Pipeline: Combines our preprocessing steps and classifier into a single, reusable object—this avoids manual data transformation and helps prevent data leakage.

Full Working Code #

from sklearn.tree import DecisionTreeClassifier
from sklearn.compose import ColumnTransformer
from sklearn.preprocessing import OrdinalEncoder
from sklearn.pipeline import Pipeline

# Your original string-containing data (no manual conversion needed!)
features = [[140,"smooth"],[130,"smooth"],[150,"bumpy"],[170,"bumpy"]]
labels = ["apple","apple","orange","orange"]

# Set up preprocessing: encode the 2nd column (index 1), leave 1st column as-is
preprocessor = ColumnTransformer(
    transformers=[
        ('categorical_encoder', OrdinalEncoder(), [1])
    ],
    remainder='passthrough'
)

# Create a pipeline that preprocesses data then trains the classifier
model = Pipeline(steps=[
    ('preprocessor', preprocessor),
    ('classifier', DecisionTreeClassifier())
])

# Train the model directly on your original data
model.fit(features, labels)

# Make a prediction using string values just like your input data
prediction = model.predict([[150, "bumpy"]])
print(prediction)  # Output: ['orange']

Key Notes #

1. No Manual Label Conversion: Scikit-learn's classifiers (including DecisionTreeClassifier) accept string labels directly—you don't need to convert "apple"/"orange" to numbers at all.
2. Consistent Encoding: OrdinalEncoder handles the string-to-number mapping for features automatically, and the pipeline ensures this same mapping is used for predictions (so you never have to remember if "smooth" was 0 or 1!).
3. Alternative: Manual Preprocessing (If You Prefer)
   If you don't want to use a pipeline, you can transform the features first then train the classifier:

   # Manually transform features
   transformed_features = preprocessor.fit_transform(features)
   
   # Train classifier on transformed features
   classifier = DecisionTreeClassifier()
   classifier.fit(transformed_features, labels)
   
   # Don't forget to transform new prediction data too!
   prediction = classifier.predict(preprocessor.transform([[150, "bumpy"]]))
   print(prediction)
   

内容的提问来源于stack exchange，提问作者Ashish.k