Fix feature importance for time series baseline

[freddyaboulton]

Pull Request Description

Fixes #3284 , Fixes #3282

---

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]

Codecov Report

Merging #3285 (4a0e79c) into main (e2bca54) will increase coverage by 0.1%.
The diff coverage is 100.0%.

@@           Coverage Diff           @@
##            main   #3285     +/-   ##
=======================================
+ Coverage   99.8%   99.8%   +0.1%     
=======================================
  Files        322     322             
  Lines      31600   31614     +14     
=======================================
+ Hits       31510   31524     +14     
  Misses        90      90             

Impacted Files	Coverage Δ
evalml/tests/component_tests/test_components.py	99.3% <ø> (ø)
...ators/regressors/time_series_baseline_estimator.py	100.0% <100.0%> (ø)
evalml/tests/automl_tests/test_automl.py	99.6% <100.0%> (+0.1%)	⬆️
...peline_tests/test_time_series_baseline_pipeline.py	100.0% <100.0%> (ø)

---

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 e2bca54...4a0e79c. Read the comment docs.

[freddyaboulton]

We typically don't set fields like this in predict but since the feature_name is not used until predict, it seemed better to keep this logic together.

[bchen1116]

LGTM! Left one comment but nothing blocking.

[bchen1116]

Personally, I'm hesitant on putting numeric values for these since if we change/add components or change the way things are calculated, these numbers will likely change. Not blocking, but something to keep in mind!

[tamargrey]

Ran this branch with my time series demo, and I'm seeing these feature importances for the baseline

Should target_rolling_mean be there? And if so, how does it have 0 importance when, in the other pipelines, it's one of the most important features?

[ParthivNaresh]

Ran this branch with my time series demo, and I'm seeing these feature importances for the baseline Should target_rolling_mean be there? And if so, how does it have 0 importance when, in the other pipelines, it's one of the most important features?

We include the target_rolling_mean from the TimeSeriesFeaturizer if the target is numeric.

Also with the baseline pipeline, predictions are based entirely off of the delayed feature, so we'd have to set that to 1 I believe.

[ParthivNaresh]

Looks solid!

[tamargrey]

We include the target_rolling_mean from the TimeSeriesFeaturizer if the target is numeric.

Also with the baseline pipeline, predictions are based entirely off of the delayed feature, so we'd have to set that to 1 I believe.

So If I'm understanding correctly: The target_rolling_mean feature shows up in the baseline pipeline importances because its added via the TimeSeriesFetureizer even though it's not actually included in the model and should always have zero importance?

[freddyaboulton]

@tamargrey Exactly. The estimator technically "sees" the rolling mean feature, but it doesn't use it, so it gets an importance of 0.