I am currently searching a grid with learning_rate between 0.001 and 0.02 and n_estimators between 200 and 4,000. But I know that all the good results are with learning_rate * n_estimators close to 4. Is there a way to specify only the learning_rate and a range for their product? I am guessing that I need to derive my own "model" class from the true model which will accept learning_rate and "product" and unbundle them to learning_rate and n_estimators and then call the true model. My base model is xgboost.XGBRegressor. I am not knowledgeable enough in Python to do this without some hints.
Note: I am actually using BayesSearchCV with a "space" parameter but I figure the solution will be the same as for GridSearchCV and a param_grid
