XGBoost 0.7二分类训练报错：base_score需处于(0,1)区间

Hey there! I’ve dealt with this exact error when working on binary classification tasks with XGBoost, so let’s walk through what’s going wrong and how to fix it quickly.

What’s Causing the Error? #

The error message clearly calls out: base_score must be in (0,1) for logistic loss.

XGBoost’s default base_score is 0.5 (which is safely in the (0,1) range), so this usually pops up for one of two reasons:

• Your training labels (y_train) aren’t in the expected format for binary logistic regression (i.e., they’re not 0/1, maybe -1/1 or another range)
• There’s a quirk with your XGBoost version where the default base_score isn’t being applied correctly

How to Fix It #

Here are three straightforward solutions to try:

• Explicitly set the base_score parameter
  Even though 0.5 is the default, explicitly defining it can override any unexpected environment or version-related issues. Update your code to:

  import xgboost as xgb
  XGB = xgb.XGBClassifier(base_score=0.5)
  model = XGB.fit(X_train, y_train)
  

• Verify your training labels
  For XGBoost’s default binary logistic loss (objective='binary:logistic'), your labels should be 0 and 1. If your y_train uses a different encoding (like -1 and 1), you have two options:

  1. Convert your labels to 0/1 (e.g., replace -1 with 0 using y_train = (y_train + 1) // 2)
  2. Switch to a different objective that supports your label format, like objective='binary:hinge'

• Upgrade XGBoost to the latest stable version
  Older versions of XGBoost had rare bugs related to base_score handling. Run this command to update:

  pip install --upgrade xgboost
  

Since you had success with scikit-learn’s GradientBoosting, making sure your labels are 0/1 and explicitly setting base_score should get your XGBoost model training smoothly.

内容的提问来源于stack exchange，提问作者Oussama Jabri