如何在xgboost.XGBClassifier()中设置正类别？是否可指定xgboost.XGBClassifier()的正类别？

Hey there! Awesome questions about customizing the positive class in xgboost.XGBClassifier()—let me walk you through this step by step.

Can we decide which class to use as the positive class in XGBClassifier? #

Absolutely! XGBoost doesn’t lock you into a default positive class (though it does have a default behavior: for binary classification, it treats the numerically larger label as the positive class). You have full control over which class you want to designate as positive.

How to set the positive class in XGBClassifier? #

There are two straightforward ways to do this, depending on whether you want to adjust your labels or use XGBoost’s built-in parameters:

1. Re-encode your target labels (most direct approach) #

This is the simplest method. For binary classification:

• Map the class you want to be positive to 1
• Map all other classes to 0

For example, if your original labels are ["spam", "ham"] and you want "ham" to be the positive class:

import pandas as pd
from xgboost import XGBClassifier

# Sample data
df = pd.DataFrame({"text": ["buy now", "hello there"], "label": ["spam", "ham"]})

# Re-encode labels: ham = 1 (positive), spam = 0 (negative)
df["label"] = df["label"].map({"ham": 1, "spam": 0})

# Initialize and train the classifier
model = XGBClassifier(objective="binary:logistic")
model.fit(df[["text"]], df["label"])

2. Use the scale_pos_weight parameter (for class balancing + explicit positive class) #

If you want to keep your original labels but still designate a specific class as positive (and handle class imbalance at the same time), use scale_pos_weight. This parameter is set to the ratio of negative class samples to positive class samples.

For example, if your labels are [0, 1, 0, 0, 1] and you want 0 to be the positive class:

• First, count the number of positive (0) and negative (1) samples: let’s say 3 positives, 2 negatives
• Set scale_pos_weight = (number of negative samples) / (number of positive samples) = 2/3

from xgboost import XGBClassifier
import numpy as np

# Sample features and labels (0 is our desired positive class)
X = np.array([[1], [2], [3], [4], [5]])
y = np.array([0, 1, 0, 0, 1])

# Calculate scale_pos_weight
neg_count = (y == 1).sum()
pos_count = (y == 0).sum()
scale_pos_weight = neg_count / pos_count

# Initialize model with the parameter
model = XGBClassifier(objective="binary:logistic", scale_pos_weight=scale_pos_weight)
model.fit(X, y)

A quick note: When using scale_pos_weight, make sure you’re consistent with which class you’re treating as positive when interpreting predictions (since the model will still output probabilities corresponding to the label 1 by default—so if you flipped the positive class to 0, you’ll need to subtract the predicted probability from 1 to get the positive class probability).

内容的提问来源于stack exchange，提问作者Sara