为什么 LightGBM 中的 R2 分数为零？

import numpy as np
import pandas as pd
import lightgbm
from sklearn.metrics import r2_score,mean_squared_error as MSE
dataset = pd.read_excel("Prali Marble.xlsx")
X = dataset.iloc[:,2].values.reshape((-1, 1))
Y = dataset.iloc[:,3].values

from lightgbm import LGBMRegressor
lgb_r = LGBMRegressor()
lgb_r.fit(X,Y)
y_pred = lgb_r.predict(X)
print("LGBM R2_SCORE:", r2_score(Y, lgb_r.predict(X)))

给出R2分数ZERO，那么为什么 LGBMRegressor 会得到零值呢？ 我没有分割我的数据 train_test 因为我的数据集很小。

这个例子不能完全重现，因为内容"Prali Marble.xlsx"不包括在内。

However, I can reproduce a 0.0 R² with the following code that I think closely matches your example. Similar to your code, this trains a LightGBM regression model on a dataset with a single feature.

这段代码使用lightgbmPython 3.8 上的 3.1.1。

import numpy as np
import pandas as pd
import lightgbm as lgb
from sklearn.metrics import r2_score,mean_squared_error as MSE

X = pd.DataFrame({
    "feat1": np.append(np.repeat(0.5, 99), np.ones(1))
})
Y = np.random.random(100, )

lgb_r = lgb.LGBMRegressor()
lgb_r.fit(X,Y)

y_pred = lgb_r.predict(X)
print("LGBM R2_SCORE:", r2_score(Y, lgb_r.predict(X)))

LGBM R2_SCORE：0.0

In this case, the R² is 0 because the model is just predicting the mean of Y. You can see this by examining the structure of the model.

lgb_r.booster_.trees_to_dataframe()

这将返回 1 行数据帧，当 LightGBM 不添加任何树时会发生这种情况。

LightGBM 有一些参数用于防止过度拟合。这里有两个相关的：

• min_data_in_leaf https://lightgbm.readthedocs.io/en/latest/Parameters.html#min_data_in_leaf（默认=20）
• min_sum_hessian_in_leaf https://lightgbm.readthedocs.io/en/latest/Parameters.html#min_sum_hessian_in_leaf（默认=0.001）

您可以通过将这些参数设置为 0 来告诉 LightGBM 忽略这些过拟合保护。

import numpy as np
import pandas as pd
import lightgbm as lgb
from sklearn.metrics import r2_score

X = pd.DataFrame({
    "feat1": np.append(np.repeat(0.5, 99), np.ones(1))
})
Y = np.random.random(100, )

lgb_r = lgb.LGBMRegressor(
    min_data_in_leaf=0,
    min_sum_hessian_in_leaf=0.0
)
lgb_r.fit(X,Y)

y_pred = lgb_r.predict(X)
print("LGBM R2_SCORE:", r2_score(Y, lgb_r.predict(X)))