* fix having mixed types in a single passthrough
* use check_array
* add comment
* Reference the relevant issue in the test
* don't force finite on check_array
* add the whats_new entry
* modify whats_new entry for a [probably] better description.
* raise a value error if columns can't be stacked.
* test the custom ValueError
* take sparse.hstack out of try/catch
* improve whats_new entry
* add comment for the dtype conversion
* change the tests to use an np of dtype 'O' instead
<!--
Thanks for contributing a pull request! Please ensure you have taken a look at
the contribution guidelines: https://github.com/scikit-learn/scikit-learn/blob/master/CONTRIBUTING.md#pull-request-checklist
-->
#### Reference Issues/PRs
<!--
Example: Fixes#1234. See also #3456.
Please use keywords (e.g., Fixes) to create link to the issues or pull requests
you resolved, so that they will automatically be closed when your pull request
is merged. See https://github.com/blog/1506-closing-issues-via-pull-requests
-->
Fixes#12096.
#### What does this implement/fix? Explain your changes.
Converts the input list for ColumnTransformer to a numpy array.
Added a check inside `transform` and `fit_transform` to check if the input `X` is a list, if it is then it gets converted to a numpy array.
#### Any other comments?
Should this conversion be documented in the docstrings for ColumnTransfomer's `fit`, `transform` and `fit_transform`?
<!--
Please be aware that we are a loose team of volunteers so patience is
necessary; assistance handling other issues is very welcome. We value
all user contributions, no matter how minor they are. If we are slow to
review, either the pull request needs some benchmarking, tinkering,
convincing, etc. or more likely the reviewers are simply busy. In either
case, we ask for your understanding during the review process.
For more information, see our FAQ on this topic:
http://scikit-learn.org/dev/faq.html#why-is-my-pull-request-not-getting-any-attention.
Thanks for contributing!
-->
Reasoning: when eg OneHotEncoder is used as part of ColumnTransformer it would cause the final result to be a sparse matrix. As this is a typical case when we have mixed dtype data, it means that many pipeline will have to deal with sparse data implicitly, even if you have only some categorical features with low cardinality.
Idea was first to change default of `OneHotEncoder` sparse to False, but based on gitter discussion (https://gitter.im/scikit-learn/dev?at=5b4e5a69a94c5255523bc9fc) we decided to let ColumnTransformer switch between both based on a threshold. The user still has full control if he/she wants always or never sparse.
* define select_types callable factory
* include dtype column selector example
* support selector function for remainder=passthrough (already worked for drop)
* add docstring to example file. apparently sphinx fails if missing
* remove example and select_dtype factory
* generalize callable case (all specification types) + add tests
* add docstring
In order to fix#11408, this swaps `joblib` and `_joblib`. It however, allows users to access joblib's `Memory` or `Parallel` functionality without accessing `sklearn.externals._joblib` by importing `Memory`, `Parallel`, etc. into `sklearn.utils`.
pierretallotte
Hanmin Qin