Cost computational

Module 6. Ensemble of models Intuitions on ensemble of tree-based models
#1

Can you explaine. How can see the computational cost is reduced in comparison of seeking for the optimal hyperparameters?

decision tree regressor with default parameters.

R2 score obtained by cross-validation: 0.354 +/- 0.087
CPU times: user 4.04 ms, sys: 36.7 ms, total: 40.8 ms
Wall time: 1.03 s

Making a grid-search to tune the hyperparameters

R2 score obtained by cross-validation: 0.523 +/- 0.107
CPU times: user 9.27 ms, sys: 4.67 ms, total: 13.9 ms
Wall time: 3.62 s

Using 50 decision trees

R2 score obtained by cross-validation: 0.642 +/- 0.083
CPU times: user 15 ms, sys: 4.31 ms, total: 19.3 ms
Wall time: 3.82 s

What is the “tree in depth” notebook which you have mentioned?

Thanks for your time

#2

I already stated the problem in that post : Potentially misleading comment on computational cost of BaggingRegressor
In short you can see the gain in computational cost locally on a computer but something is wrong on the server.
for “tree in depth” I think it is this notebook: Importance of decision tree hyperparameters on generalization

#3

Thank you Echidne!