GridSearchCV is not properly working with RandomBoosting*()

[tkrabel]

model = RandomBoostingRegressor()
rb = GridSearchCV(model, param_grid={'n_estimators':[75, 100, 150]}, cv=5)
rb = rb.fit(X_train, y_train)

This code fails with an IndexError

File ".../python_random_boost/random_boost/random_boost.py", line 1183, in _fit_stage self.depths_[i] = depth
IndexError: index 100 is out of bounds for axis 0 with size 100

The reason is that RandomBoostingRegressor() is initialized with n_estimators=100, and it seems GridSearchCV() is not able to properly overwrite n_estimators when searching the grid (hence leading to an error when it tries out n_estimators=150. This is not a problem with GradientBoostingRegressor(), which has the same default value for n_estimators.

In essence, the problem is related to the fact that I create a vector in which I save the drawn tree depth values. The length of this vector equals n_estimators and seems to be set to 100 independent from the actual value tried out by GridSearchCV.

[tkrabel]

There is no error when you initialize RandomBoostingRegressor with a high enough value for n_estimators. Is the issue related to the fact that RandomBoostinRegressor is a sub-sub-class from BaseGradientBoosting (level two inheritance)?