Skip to content

Separate pipeline_parameters from custom_hyperparameters - #2317

Merged
ParthivNaresh merged 72 commits into
mainfrom
Clean-API-Frozen-And-HyperParameters
Jun 7, 2021
Merged

ParthivNaresh merged 72 commits into
mainfrom
Clean-API-Frozen-And-HyperParameters

Conversation

@ParthivNaresh

@ParthivNaresh ParthivNaresh commented May 31, 2021 •

Copy link
Copy Markdown
Contributor

Fixes #2130

This PR deals with separating the custom_hyperparameters argument from its current pipeline implementation to AutoMLSearch instead. This is the first step before we replace allowed_pipelines with allowed_component_graphs and keep time series specifications within problem_configuration.

Document comparing the differences between the previous implementation and this one can be found here.

@ParthivNaresh
ParthivNaresh marked this pull request as ready for review June 4, 2021 18:43

@bchen1116 bchen1116 left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nice job, Parthiv! Left a few nits and testing suggestions.

Comment thread dev-requirements.txt Resolved
@freddyaboulton freddyaboulton mentioned this pull request Jun 4, 2021

@dsherry dsherry left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@ParthivNaresh this is really excellent! And comprehensive. I left a few code organization things, in particular moving the ValueError for skopt parameters sneaking into custom_parameters to a different location, but not much.

Comment thread docs/source/user_guide/automl.ipynb Resolved
Comment thread evalml/automl/automl_search.py Resolved
Comment thread evalml/automl/automl_search.py Outdated
self._frozen_pipeline_parameters = {}

parameters = copy.copy(self.pipeline_parameters)
custom_hyperparameters = copy.copy(self.custom_hyperparameters)

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Why do a copy here? We copy parameters on the previous line because we then create a modified version where we add the pipeline-level parameters (time series config etc). But we don't do that for the custom hyperparameters, right?

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ah thanks, initially I was considering modifying the custom_hyperparameters based on other criteria but forgot to change this back.

Comment thread evalml/automl/automl_search.py Resolved

@freddyaboulton freddyaboulton left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@ParthivNaresh Looks great to me! I left a minor comment that's similar to @dsherry `s suggestion but other than that this is awesome.

Comment thread evalml/automl/callbacks.py Resolved

@dsherry dsherry left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🚢 🙏

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

AutoMLSearch: clean up API for freezing parameters and constraining hyperparam ranges

6 participants