replace Boston dataset with diabetes dataset in examples #357
Conversation
| scoring=make_scorer(mean_squared_error), | ||
| cv=n_folds) | ||
|
|
||
| rgf_score = sum(rgf_scores)/n_folds | ||
| print('RGF Regressor MSE: {0:.5f}'.format(rgf_score)) | ||
| # >>>RGF Regressor MSE: 11.79409 | ||
| # >>> RGF Regressor MSE: 3377.46076 |
There was a problem hiding this comment.
MSE is large?
https://www.researchgate.net/figure/MSE-Convergence-of-used-algorithms-for-Diabetes-dataset_fig3_346454502
How about increse max_leaf and n_iter?
There was a problem hiding this comment.
@fukatani Thanks a lot for your feedback!
Yeah, I tried to increase those hyperparameters, of course. Unfortunately, increasing them makes results only worse.
I believe that this example is to show that RGF can outperform RF in some cases. And this goal is already achieved
RGF Regressor MSE: 3377.46076
Random Forest Regressor MSE: 3441.01988
Also, I have a great doubt that the article you've mentioned is referring the same dataset used in scikit-learn. MSE less than 0.55 for y scaled from 25 to 346 looks unbelievable.
To prove my words, here are some examples of MSE for this dataset over the GitHub:
https://github.com/Akshaykumarcp/ML-ensemble-learning/blob/b142d7d2ac32554fe8dba545bfcb0a2fb31e02da/0.2_generative_ensemble_methods/0.2.2_boosting_method/0.2.2.3_adaboost_scikit_regression.py#L37-L40
There was a problem hiding this comment.
OK, now I'm sure these datasets are different:
Source URL: https://www4.stat.ncsu.edu/~boos/var.select/diabetes.html
https://scikit-learn.org/stable/datasets/toy_dataset.html#diabetes-dataset
Size: 11x442.
The diabetes dataset containing multivariate characteristicsthat are extracted from the University of California machine learn-ing repository contained 768 specimens of adult females. This depicts chemical changes occurs from initial to peak stages in the female body that result in diabetes (Smith et al., 1988).
There was a problem hiding this comment.
Thx!
LGTM.
Sorry for very late response 🙏
There was a problem hiding this comment.
No problem at all! Thank you very much for the thoughtful review!
Boston dataset will be removed from scikit-learn in next version due to ethical reasons. Details: scikit-learn/scikit-learn#16155.