Python运行XGBoost报xgboost.core.XGBoostError 0特征输入错误求助

错误根本原因 #

你的数据集仅包含Posts Frequency、Likes Count两列，构造训练特征时把两列都排除了，导致输入XGBoost的训练特征集为空，触发了0 feature is supplied的报错。

具体问题点 #

• 构造X_train时使用了~train_data.columns.isin(['Posts Frequency','Likes Count'])的过滤规则，直接把仅有的两列都过滤掉，训练特征维度为0
• 重复执行了两次GroupShuffleSplit拆分逻辑，属于冗余代码
• 构造y_test时列名多输入了前置空格，写为[' Likes Count']，会导致测试集标签为空
• 预测函数中过滤规则写为~data.columns.isin( ['Posts Frequency'] )，如果后续特征构造正确的话，也会导致预测时传入空特征

修复方案 #

1. 修正训练/测试集的特征构造逻辑，仅过滤标签列Likes Count，保留特征列Posts Frequency
2. 删除重复的拆分代码，修正y_test的列名空格问题
3. 同步修正预测函数的特征过滤规则

修正后的核心代码片段 #

# 仅保留一次拆分逻辑即可
gss = GroupShuffleSplit(test_size=.20, n_splits=1, random_state = 7).split(data, groups=data['Posts Frequency'])
X_train_inds, X_test_inds = next(gss)

train_data= data.iloc[X_train_inds]
# 仅排除标签列，保留特征列
X_train = train_data.loc[:, ~train_data.columns.isin(['Likes Count'])]
y_train = train_data.loc[:, train_data.columns.isin(['Likes Count'])]
groups = train_data.groupby('Posts Frequency').size().to_frame('size')['size'].to_numpy()

test_data= data.iloc[X_test_inds]
X_test = test_data.loc[:, ~test_data.columns.isin(['Likes Count'])]
# 修正列名的空格问题
y_test = test_data.loc[:, test_data.columns.isin(['Likes Count'])]

# 训练代码不变
model = xgb.XGBRanker(
    tree_method='gpu_hist',
    booster='gbtree',
    objective='rank:pairwise',
    random_state=42,
    learning_rate=0.1,
    colsample_bytree=0.9,
    eta=0.05,
    max_depth=6,
    n_estimators=110,
    subsample=0.75
    )
model.fit(X_train, y_train, group=groups, verbose=True)

# 修正预测函数的特征过滤规则
def predict(model, data):
    return model.predict( data.loc[:, ~data.columns.isin( ['Likes Count'] )] )

predictions = (data.groupby( 'Posts Frequency' )
               .apply( lambda x: predict( model, x ) ))

内容的提问来源于stack exchange，提问作者Sofia Vlachou