Rename random state to random seed by freddyaboulton · Pull Request #1798 · alteryx/evalml

freddyaboulton · 2021-02-08T21:51:52Z

Pull Request Description

Fixes #1737 by adding random_seed as an argument and keeping random_state but setting it to None. A warning will show if random_state is used.

After creating the pull request: in order to pass the release_notes_updated check you will need to update the "Future Release" section of docs/source/release_notes.rst to include this pull request by adding :pr:123.

codecov · 2021-02-08T22:30:15Z

Codecov Report

Merging #1798 (5778ed2) into main (cb09781) will increase coverage by 0.2%.
The diff coverage is 100.0%.

@@           Coverage Diff            @@
##            main    #1798     +/-   ##
========================================
+ Coverage   99.9%   100.0%   +0.2%     
========================================
  Files        252      252             
  Lines      20117    20250    +133     
========================================
+ Hits       20086    20242    +156     
+ Misses        31        8     -23

Impacted Files	Coverage Δ
...components/ensemble/stacked_ensemble_classifier.py	`100.0% <ø> (ø)`
.../components/ensemble/stacked_ensemble_regressor.py	`100.0% <ø> (ø)`
...components/transformers/encoders/target_encoder.py	`100.0% <ø> (ø)`
...elines/components/transformers/imputers/imputer.py	`100.0% <ø> (ø)`
...mponent_tests/test_delayed_features_transformer.py	`100.0% <ø> (ø)`
evalml/utils/__init__.py	`100.0% <ø> (ø)`
evalml/automl/automl_algorithm/automl_algorithm.py	`100.0% <100.0%> (ø)`
...lml/automl/automl_algorithm/iterative_algorithm.py	`100.0% <100.0%> (ø)`
evalml/automl/automl_search.py	`99.7% <100.0%> (ø)`
evalml/automl/utils.py	`100.0% <100.0%> (ø)`
... and 90 more

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update cb09781...5778ed2. Read the comment docs.

freddyaboulton · 2021-02-09T16:52:10Z

                 tuner_class=None,
                 text_columns=None,
-                 random_state=0,
+                 random_seed=0,


I'm replacing random_state with random_seed here, rather than adding it as an argument, because this is not a user-facing class.

freddyaboulton · 2021-02-09T16:53:45Z

+        random_seed = deprecate_arg("random_state", "random_seed", random_state, random_seed)
        if n_components:
-            pca = SkPCA(n_components=n_components, **kwargs)
+            pca = SkPCA(n_components=n_components, random_state=random_seed, **kwargs)


Not sure why we were not passing random_state to PCA before.

Ooo, good catch 👀

bchen1116

Nice job fixing everything and raising the warnings! LGTM

bchen1116 · 2021-02-09T22:04:17Z

    automl.search()
    assert automl._pipelines_per_batch == 2
    assert automl._automl_algorithm.pipelines_per_batch == 2
+    assert automl._automl_algorithm.pipelines_per_batch == 2


repeated line?

bchen1116 · 2021-02-09T22:14:42Z

        return filepath
+
+
+def deprecate_arg(old_arg, new_arg, old_value, new_value):


Should we add a test for this helper since it's public?

bchen1116 · 2021-02-09T22:17:02Z

        * Sped up permutation importance for some pipelines :pr:`1762`
    * Fixes
    * Changes
+        * Added ``random_seed`` as an argument to our automl/pipeline/component api. Using ``random_state`` will raise a warning :pr:`1798`


We probably can't add this to breaking_changes, but is there another category similar that we can put it in to warn users? Just wondering since it'll soon be a breaking change once we officially deprecate.

I think this is fine for now? Here we're telling users we'll raise a warning and the warning itself will tell them about the impending deprecation.

angela97lin

Left a few comments but nice and thorough. Thank you, LGTM!! 🚢

angela97lin · 2021-02-10T22:22:34Z

        self.ensembling = ensembling and len(self.allowed_pipelines) > 1
        self._pipeline_params = pipeline_params or {}
-        self._random_state = random_state
+        self._random_state = random_seed


Do we want to update this to self._random_seed too?

angela97lin · 2021-02-10T22:23:49Z

        next_batch = []
        if self._batch_number == 0:
-            next_batch = [pipeline_class(parameters=self._transform_parameters(pipeline_class, {}), random_state=self.random_state)
+            next_batch = [pipeline_class(parameters=self._transform_parameters(pipeline_class, {}), random_seed=self.random_seed)


Confused, what is self.random_seed set to here? 👀 I ask because I made the previous comment and noticed that we have self._random_state right now... Is this supposed to be self.random_seed, and then we need to update self._random_seed --> self.random_seed?

self.random_seed == self._random_state. We get self.random_seed from the parent class. self._random_state is used to sample params in the first batch if the user provides pipelines with frozen hyperparams. I'll get rid of self._random_state since we don't need it anymore!

angela97lin · 2021-02-10T22:29:05Z

@@ -36,8 +36,8 @@ class CatBoostClassifier(Estimator):
    SEED_MAX = SEED_BOUNDS.max_bound


More out of curiosity but does this mean we dont need SEED_BOUNDS anymore? Or maybe we need to check that the seed provided is within those bounds?

I think we can get rid of them actually!

angela97lin · 2021-02-10T22:34:06Z

+        random_seed = deprecate_arg("random_state", "random_seed", random_state, random_seed)
        if n_components:
-            pca = SkPCA(n_components=n_components, **kwargs)
+            pca = SkPCA(n_components=n_components, random_state=random_seed, **kwargs)


Ooo, good catch 👀

angela97lin · 2021-02-10T22:36:55Z

                Will override default objectives for problem type if not empty.

-            random_state (int): Seed for the random number generator. Defaults to 0.
+            random_state (None, int): Deprecated - use random_seed instead.


Really nit-picking (for no reason loool) but since the other places that were touched are now random_state (int), let's match that hehe

angela97lin · 2021-02-10T23:13:43Z

        return filepath
+
+
+def deprecate_arg(old_arg, new_arg, old_value, new_value):


Nice, this will be useful for any future dep patterns we follow 🥳

freddyaboulton added the refactor Work being done to refactor code. label Feb 8, 2021

freddyaboulton self-assigned this Feb 8, 2021

freddyaboulton changed the title ~~1737 rename random state to random seed~~ Rename random state to random seed Feb 8, 2021

freddyaboulton force-pushed the 1737-rename-random-state-to-random-seed branch from 0fad373 to 69756a3 Compare February 9, 2021 16:30

freddyaboulton commented Feb 9, 2021

View reviewed changes

freddyaboulton marked this pull request as ready for review February 9, 2021 17:38

freddyaboulton requested review from ParthivNaresh, angela97lin, bchen1116, chukarsten, dsherry and jeremyliweishih February 9, 2021 17:38

bchen1116 approved these changes Feb 9, 2021

View reviewed changes

angela97lin approved these changes Feb 10, 2021

View reviewed changes

freddyaboulton added 9 commits February 11, 2021 14:32

Renaming random_state to random_seed in pipelines/transformers.

21b8aaf

Fixing setting random_seed instead of random_state everywhere.

2633697

Updating docs and linting.

639f874

Adding tests to check the warning is raised.

45b53d3

Adding PR 1798 to release notes.

38c86d9

Rewrittign skopt tuner test because circleci raises more warnings.

2225edb

Deleting more places where random_state is used.

f5f3f43

Fixing docstring in deprecate arg.

66f5ded

Adding test for deprecate_arg and misc review.

7c37f3e

freddyaboulton force-pushed the 1737-rename-random-state-to-random-seed branch from a881926 to 7c37f3e Compare February 11, 2021 20:11

Adding deprecation warning for split_data.

5778ed2

freddyaboulton merged commit 5cba8a9 into main Feb 11, 2021

freddyaboulton deleted the 1737-rename-random-state-to-random-seed branch February 11, 2021 21:26

chukarsten mentioned this pull request Feb 23, 2021

v0.19.0 Release #1884

Merged

dsherry mentioned this pull request Mar 10, 2021

v0.20.0 #1961

Merged

		return filepath


		def deprecate_arg(old_arg, new_arg, old_value, new_value):

		@@ -36,8 +36,8 @@ class CatBoostClassifier(Estimator):
		SEED_MAX = SEED_BOUNDS.max_bound

Uh oh!

Conversation

freddyaboulton commented Feb 8, 2021

Pull Request Description

Uh oh!

codecov Bot commented Feb 8, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

bchen1116 left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

angela97lin left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

codecov Bot commented Feb 8, 2021 •

edited

Loading