|
| 1 | +--- |
| 2 | +#### Blog Post Template #### |
| 3 | + |
| 4 | +#### Post Information #### |
| 5 | +title: "Changes and development of scikit-learn's developer API" |
| 6 | +date: December 12, 2024 |
| 7 | + |
| 8 | +#### Post Category and Tags #### |
| 9 | +# Format in titlecase without dashes (Ex. "Open Source" instead of "open-source") |
| 10 | +categories: |
| 11 | + - Updates |
| 12 | +tags: |
| 13 | + - Open Source |
| 14 | + - Machine Learning |
| 15 | + - License |
| 16 | + |
| 17 | +#### Featured Image #### |
| 18 | +featured-image: BSD_watermark.svg |
| 19 | + |
| 20 | +#### Author Info #### |
| 21 | +# Can accomodate multiple authors |
| 22 | +# Add SQUARE Author Image to /assets/images/author_images/ folder |
| 23 | +postauthors: |
| 24 | + - name: Adrin Jalali |
| 25 | + website: https://adrin.info/ |
| 26 | + image: adrin-jalali.jpeg |
| 27 | +--- |
| 28 | +<div> |
| 29 | + <img src="/assets/images/posts_images/{{ page.featured-image }}" alt=""> |
| 30 | + {% include postauthor.html %} |
| 31 | +</div> |
| 32 | + |
| 33 | +Historically, scikit-learn's API has been divided into public and private. Public API is |
| 34 | +intended to be used by users, and private API is used internally in scikit-learn to |
| 35 | +develop new features and estimators. However, many of those functionalities have become |
| 36 | +essential to develop scikit-learn estimators by third parties who develop them outside |
| 37 | +the scikit-learn codebase. |
| 38 | + |
| 39 | +When it comes to our public API, we have very strict and high standards on backward |
| 40 | +compatibility. The rule of thumb is that no change should cause a change in users' |
| 41 | +code unless we warn about it for two release cycles, which means we give users a year |
| 42 | +time to update their code. |
| 43 | + |
| 44 | +On the other hand, we have no such guarantees or constraints on our private API. This |
| 45 | +brings an issue to third party developers who would like to use methods used by |
| 46 | +scikit-learn developers to develop their estimators. Constantly changing private API |
| 47 | +without prior warning brings certain challenges to third party developers which is not |
| 48 | +ideal. |
| 49 | + |
| 50 | +As a result, we've been working on creating a developer API which would sit somewhere |
| 51 | +between our public and private API in terms of backward compatibility. That means we |
| 52 | +intend to try to keep that API stable, and if needed, introduce changes with one release |
| 53 | +cycle warning. |
| 54 | + |
| 55 | +In the past few releases, we've slowly introduced more functionalities under this |
| 56 | +umbrella. `__sklearn_clone__` and `__sklearn_is_fitted__` are two examples. |
| 57 | + |
| 58 | +In the 1.6 release, we focused on the testing infrastructure and estimator tag system. |
| 59 | +Estimator tags used to be private, and we were not sure about their design. In the 1.6 |
| 60 | +release, new tags are introduced and using them looks like the following: |
| 61 | + |
| 62 | +```python |
| 63 | +from sklearn.base import BaseEstimator, ClassifierMixin |
| 64 | + |
| 65 | +class MyEstimator(ClassifierMixin, BaseEstimator): |
| 66 | + |
| 67 | + ... |
| 68 | + |
| 69 | + def __sklearn_tags__(self): |
| 70 | + tags = super().__sklearn_tags__() |
| 71 | + # modify tags here |
| 72 | + tags.non_deterministic = True |
| 73 | + return tags |
| 74 | +``` |
| 75 | + |
| 76 | +The new tags mostly follow the same structure as the old tags, but there are certain |
| 77 | +changes to them. The main change is that the old `_xfail_checks` is no longer present |
| 78 | +in the new tags. That tag was used to tell the common testing tools about the tests |
| 79 | +which are known to fail and are to be skipped. That information is now directly passed |
| 80 | +to the test functionalities. The old way of skipping a test was the following: |
| 81 | + |
| 82 | +```python |
| 83 | +from sklearn.base import BaseEstimator, ClassifierMixin |
| 84 | + |
| 85 | +class MyEstimator(ClassifierMixin, BaseEstimator): |
| 86 | + |
| 87 | + ... |
| 88 | + |
| 89 | + def _more_tags(self): |
| 90 | + return { |
| 91 | + "_xfail_checks": { |
| 92 | + "check_to_skip_name": "this check is known to fail", |
| 93 | + ... |
| 94 | + } |
| 95 | + } |
| 96 | +``` |
| 97 | + |
| 98 | +And then when calling `check_estimator` or using `parametrize_with_checks` with `pytest` |
| 99 | +would automatically ignore those tests for the estimator. |
| 100 | + |
| 101 | +Instead, in this release, you pass that information directly to those methods: |
| 102 | + |
| 103 | +```python |
| 104 | +from sklearn.utils.estimator_checks import check_estimator, parametrize_with_checks |
| 105 | + |
| 106 | +CHECKS_EXPECTED_TO_FAIL = { |
| 107 | + "check_to_skip_name": "this check is known to fail", |
| 108 | + ... |
| 109 | +} |
| 110 | + |
| 111 | +# Using check_estimator |
| 112 | +def test_with_check_estimator(): |
| 113 | + check_estimator(MyEstimator(), expected_failed_checks=CHECKS_EXPECTED_TO_FAIL) |
| 114 | + |
| 115 | +# Using parametrize_with_checks |
| 116 | +@parametrize_with_checks( |
| 117 | + [MyEstimator()], |
| 118 | + expected_failed_checks=lambda est: CHECKS_EXPECTED_TO_FAIL |
| 119 | +) |
| 120 | +def test_with_parametrize_with_checks(estimator, check): |
| 121 | + check(estimator) |
| 122 | +``` |
| 123 | + |
| 124 | +While working on the testing infrastructure, we have also been working on improving our |
| 125 | +tests and that means in this release we had a particularly high number of changes in |
| 126 | +their names and what they do. The changes will make it easier for developers to fix |
| 127 | +issues with their estimators. Note that you can now pass `legacy=False` to both |
| 128 | +`check_estimator` and `parametrize_with_checks` to include only strictly API related |
| 129 | +tests. |
| 130 | + |
| 131 | +The above changes mean developers need to update their estimators and depending on |
| 132 | +what they use, write scikit-learn version specific code to handle supporting multiple |
| 133 | +scikit-learn versions. To make that process easier, we've worked on a package called |
| 134 | +[`sklearn_compat`](https://github.com/sklearn-compat/sklearn-compat/). You can either |
| 135 | +depend on it as a package dependency, or vendor a single file inside your project. At |
| 136 | +the moment this project is in its infancy and might change in the future. But hopefully |
| 137 | +it helps developers out there. |
| 138 | + |
| 139 | +If you think there are missing functionalities in the developer API, please let us know |
| 140 | +and give us feedback on our [issue tracker]( |
| 141 | +https://github.com/scikit-learn/scikit-learn/issues). |
0 commit comments