LightGBM 警告：没有有意义的特征，因为所有特征值都是常量

Question

我试过下面的简单代码：

import lightgbm, pandas
params = {'objective': 'multiclass', 'num_classes': 4}
train_df = pandas.DataFrame({'f0': [0, 1, 2, 3] * 5, 'f1': [0, 0, 1] * 6 + [1, 2]}, dtype=float)
train_target = pandas.Series([0, 1, 2, 3] * 5)
train_set = lightgbm.Dataset(train_df, train_target)
model = lightgbm.train(params=params, train_set=train_set)

输出结果如下：

[LightGBM] [Warning] There are no meaningful features, as all feature values are constant.

[LightGBM] [Info] Total Bins 0

[LightGBM] [Info] Number of data: 20, number of used features: 0

[LightGBM] [Info] Start training from score -1.386294

[LightGBM] [Info] Start training from score -1.386294

[LightGBM] [Info] Start training from score -1.386294

[LightGBM] [Info] Start training from score -1.386294

[LightGBM] [Warning] Stopped training because there are no more leaves that meet the split requirements

我的特征显然不是一成不变的。

怎么了？

我运行 Python 3.5.2 Ubuntu 16.04.

Answer 1

我想通了。

问题是min_data_in_leaf的默认值是20，我没改。

我的数据只有 20 行。为此，LightGBM 报告无法拆分，因为每次拆分的最小样本数为 20。

（事实上，它不需要分裂它，因为解决方案是一棵只有一片叶子的树。但显然 LightGBM 正在检查分裂的可能性。）

我增加了行数，LightGBM 训练得很好。

LightGBM 警告：没有有意义的特征，因为所有特征值都是常量

LightGBM Warning: There are no meaningful features, as all feature values are constant

lightgbm