So, I have mainly focused on tuning n_estimators and max_depth hyperparameter of RandomForestRegressor (as pointed out in the literature and also Varada's lecture notes) and the it is safe to say that the model does not give R2 scores better than R2 = 0.359 and RMSE = 1.201. So it is obviously performing worse than SVR(). The GridSearch is quite slow here with RandomForestRegressor but is the right choice for our case. The relatively poor performance of the model is in accordance with what I have observed in literature and the notes.
Increasing the n_estimators hyperparameter improves the scores but that is insignificant. The same story is with max_depth hyperparameter.
For RMSE , playing with the hyperparameters, I have not obtained a value better than 1.201.
arashshams
commented
3 years ago
So, I have mainly focused on tuning n_estimators and max_depth hyperparameter of RandomForestRegressor (as pointed out in the literature and also Varada's lecture notes) and the it is safe to say that the model does not give R2 scores better than R2 = 0.359 and RMSE = 1.201. So it is obviously performing worse than SVR(). The GridSearch is quite slow here with RandomForestRegressor but is the right choice for our case. The relatively poor performance of the model is in accordance with what I have observed in literature and the notes. Increasing the n_estimators hyperparameter improves the scores but that is insignificant. The same story is with max_depth hyperparameter. For RMSE , playing with the hyperparameters, I have not obtained a value better than 1.201.