According to the docs,
for the cross_val_score's scoring parameter:
If None, the estimator’s default scorer (if available) is used.
For a DecisionTreeRegressor, the default criterion is mse. So why am I getting different results here?
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_squared_error
from sklearn.metrics import r2_score
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3)
dt = DecisionTreeRegressor(max_depth=4, min_samples_leaf=0.26)
- cross_val_score(dt, X_train, y_train, cv=10, scoring='neg_mean_squared_error')
>>> array([ 46.94808341, 18.78121305, 18.19914701, 18.06935431,
17.19546733, 28.91247609, 39.41410887, 21.30453162,
31.96443414, 23.74191199])
cross_val_score(dt, X_train, y_train, cv=10)
>>> array([ 0.35723619, 0.75254466, 0.7181376 , 0.65718608, 0.72531937,
0.4752839 , 0.43169728, 0.63916363, 0.41406146, 0.68977882])
If I had to guess, it seems the default scoring is R2 instead of mse. Is my understanding of default scorer correct or is this a bug?
The default scorer of a DecisionTreeRegression is the r2-score, you can find it in the docs of the DecisionTreeRegression.
score(self, X, y, sample_weight=None)[source]
Return the coefficient of determination R^2 of the prediction.
The coefficient R^2 is defined as (1 - u/v), where u is the residual sum of squares ((y_true - y_pred) ** 2).sum() and v is the total sum of squares ((y_true - y_true.mean()) ** 2).sum(). The best possible score is 1.0 and it can be negative (because the model can be arbitrarily worse). A constant model that always predicts the expected value of y, disregarding the input features, would get a R^2 score of 0.0.