Make StackingCVRegressor capable of replacing regressors in GridSearchCV

[qiagu]

Description

Current named_regressors and the fitting regressors are disconnected. When replacing the one of regressors, either base or meta, in GridSearchCV, the functioning regressors remain unchanged.
The nice naming of the meta_regressor is lost after this PR, but it seems sklearn processes many single estimator parameters in such way.
After this PR, all keys in the get_params are able to set_params and functional in SearchCV.

Related issues or pull requests

Pull Request Checklist

• Added a note about the modification or contribution to the ./docs/sources/CHANGELOG.md file (if applicable)
• Added appropriate unit test functions in the ./mlxtend/*/tests directories (if applicable)
• Modify documentation in the corresponding Jupyter Notebook under mlxtend/docs/sources/ (if applicable)
• Ran nosetests ./mlxtend -sv and make sure that all unit tests pass (for small modifications, it might be sufficient to only run the specific test file, e.g., nosetests ./mlxtend/classifier/tests/test_stacking_cv_classifier.py -sv)
• Checked for style issues by running flake8 ./mlxtend

[qiagu]

@rasbt If you like this PR, I can add a test for the regressor replacement.

[coveralls]

Coverage increased (+0.03%) to 91.543% when pulling cf2db67 on qiagu:stackingcv into 3f0935a on rasbt:master.

[rasbt]

Thanks a lot, I really like this PR. I think the naming change is in the worst case a bit unfortunate, but not a big deal, because it's a substantial improvement overall.

I can add a test for the regressor replacement.

that would be great!

Also, we may have to clean up the documentation a little bit (the documentation files are in https://github.com/rasbt/mlxtend/tree/master/docs/sources/user_guide from which the HTML docs are auto-generated upon version updates)

I think this will allow us to have much better compatibility with GridSearchCV like you mentioned -- there were a couple of questions about this in the past. (Probably related to #511)

Also, but this would be a bit of extra work as we haven't finished the refactoring, it would be nice to bring this to the other stacking classes as well (StackingClassifier, StackingCVClassifier, and StackingRegressor).

[qiagu]

Thanks, @rasbt. I'll make a test and update the documentation file, but my work for this PR, probably, will be limited to StackingCVRegressor. I can come back later for other stacking estimators after I finish current tasks on hand, if those estimators still need love at that time.
For SequentialFeatureSelector in #511, I see that GridSearchCV is able to change the attributes estimator and named_est, but not est_, whereas the last one carries the real estimator doing the fit and calculating scores. So, yes, they are similar issues.

[rasbt]

probably, will be limited to StackingCVRegressor. I can come back later for other stacking estimators after I finish current tasks on hand, if those estimators still need love at that time.

No worries, Rome wasn't built in one day, I will add it to the "Issues" to do list :)

in #511, I see that GridSearchCV is able to change the attributes estimator and named_est, but not est_, whereas the last one carries the real estimator doing the fit and calculating scores. So, yes, they are similar issues.

Good catch, yeah, I think there are lot of these things where using more of scikit-learn's paradigms would be useful to mitigate these issues.

[qiagu]

@rasbt Finally, I make a base utility which could be applied to other stacking estimators. Also, ordered parameter replacement seems to be possible now.
I often feel struggling in English and naming stuff, please feel free to correct my doc, comment, naming and other things.
If there is still a missing piece for StackingCVRegressor, please let me know.

[rasbt]

Thanks a lot, this is great! Only made a few minor tweaks and looking forward to merging once the tests pass.

[qiagu]

haha, typo.

[rasbt]

Oh, didn't catch that one, thanks again!