[MRG] Doctest with print change only adjusts default options for doctest

[thomasjpfan]

Reference Issues/PRs

Fixes #13985
Fixes #13564

What does this implement/fix? Explain your changes.

This PR changes:

1. Makes NORMALIZE_WHITESPACE and ELLIPSIS the default options.
2. Changes formatting for estimators by setting print_changed_only=True for doctest.

Any other comments?

This PR only changes print_changed_only on doctests, and not normal tests.

[amueller]

why did you change two things :-/ I assume the global doctest options are non-controversial while the change only might be.

Also: I really hope you did this automatically ;)

Personally I think this is much more informative. A possible source of confusion could be that locally people get different outputs.

[jnothman]

There still appear to be some flags used in doctests in this PR:

doc/modules/linear_model.rst:    >>> reg.intercept_ #doctest: +ELLIPSIS
doc/modules/model_evaluation.rst:  >>> hinge_loss(y_true, pred_decision, labels)  #doctest: +ELLIPSIS
setup.cfg:doctest_optionflags = NORMALIZE_WHITESPACE ELLIPSIS
sklearn/datasets/descr/twenty_newsgroups.rst:  >>> vectors.nnz / float(vectors.shape[0])       # doctest: +ELLIPSIS
sklearn/datasets/descr/twenty_newsgroups.rst:  >>> metrics.f1_score(newsgroups_test.target, pred, average='macro')  # doctest: +ELLIPSIS
sklearn/datasets/descr/twenty_newsgroups.rst:  >>> metrics.f1_score(pred, newsgroups_test.target, average='macro')  # doctest: +ELLIPSIS
sklearn/datasets/descr/twenty_newsgroups.rst:  >>> metrics.f1_score(newsgroups_test.target, pred, average='macro')  # doctest: +ELLIPSIS
sklearn/metrics/classification.py:    >>> hinge_loss(y_true, pred_decision, labels)  #doctest: +ELLIPSIS
sklearn/metrics/pairwise.py:    >>> manhattan_distances([[3]], [[3]])#doctest:+ELLIPSIS
sklearn/metrics/pairwise.py:    >>> manhattan_distances([[3]], [[2]])#doctest:+ELLIPSIS
sklearn/metrics/pairwise.py:    >>> manhattan_distances([[2]], [[3]])#doctest:+ELLIPSIS
sklearn/metrics/pairwise.py:         [[1, 2], [0, 3]])#doctest:+ELLIPSIS
sklearn/metrics/pairwise.py:    >>> manhattan_distances(X, y, sum_over_features=False)#doctest:+ELLIPSIS
sklearn/neighbors/unsupervised.py:      >>> neigh.fit(samples)  #doctest: +ELLIPSIS
sklearn/neighbors/unsupervised.py:      ... #doctest: +ELLIPSIS
sklearn/preprocessing/label.py:    >>> le.transform([1, 1, 2, 6]) #doctest: +ELLIPSIS
sklearn/preprocessing/label.py:    >>> le.transform(["tokyo", "tokyo", "paris"]) #doctest: +ELLIPSIS
sklearn/utils/__init__.py:    >>> list(gen_even_slices(10, 10))                     #doctest: +ELLIPSIS
sklearn/utils/__init__.py:    >>> list(gen_even_slices(10, 5))                      #doctest: +ELLIPSIS
setup.cfg:doctest_optionflags = NORMALIZE_WHITESPACE ELLIPSIS
sklearn/linear_model/coordinate_descent.py:    ... #doctest: +NORMALIZE_WHITESPACE
sklearn/linear_model/coordinate_descent.py:    ... #doctest: +NORMALIZE_WHITESPACE
sklearn/linear_model/sag.py:    ... #doctest: +NORMALIZE_WHITESPACE
sklearn/linear_model/sag.py:    ... #doctest: +NORMALIZE_WHITESPACE
sklearn/linear_model/stochastic_gradient.py:    ... #doctest: +NORMALIZE_WHITESPACE
sklearn/linear_model/stochastic_gradient.py:    ... #doctest: +NORMALIZE_WHITESPACE
sklearn/svm/classes.py:    >>> clf.fit(X, y) #doctest: +NORMALIZE_WHITESPACE
sklearn/svm/classes.py:    >>> clf.fit(X, y) #doctest: +NORMALIZE_WHITESPACE
sklearn/svm/classes.py:    >>> clf.fit(X, y) #doctest: +NORMALIZE_WHITESPACE
sklearn/svm/classes.py:    >>> clf.fit(X, y)  #doctest: +NORMALIZE_WHITESPACE

[jnothman]

nice

[jnothman]

I like this, btw... I think it makes doctest maintenance much more straightforward.

[rth]

Thanks @thomasjpfan , this is very nice!

(Fixed a last few occurrences of +ELLIPSIS)

[rth]

Opened a follow-up issue in #14008 about using print_changed_only=True by default for meta-estimators.

[amueller]