Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[ENH] forecasting: ensure compositors (pipeline, tuning) work for multivariate forecasters #1364

Closed
2 of 6 tasks
fkiraly opened this issue Aug 30, 2021 · 6 comments · Fixed by #2458
Closed
2 of 6 tasks

[ENH] forecasting: ensure compositors (pipeline, tuning) work for multivariate forecasters #1364

fkiraly opened this issue Aug 30, 2021 · 6 comments · Fixed by #2458
Labels
feature request New feature or request good first issue Good for newcomers implementing algorithms Implementing algorithms, estimators, objects native to sktime implementing framework Implementing or improving framework for learning tasks, e.g., base class functionality module:forecasting forecasting module: forecasting, incl probabilistic and hierarchical forecasting
Projects
Workstream: forecasting and series tr...

Comments

@fkiraly
Copy link
Collaborator

fkiraly commented Aug 30, 2021 •
edited

From the Aug 27 developer meeting.

We need to ensure that the following compositors work for multivariate forecasters, in descending order of priority

Some of these might be slightly challenging, but still good first issues.
As usual, we use this issue to track who is working on what, please indicate your pick before starting work on it.
@fkiraly fkiraly added feature request New feature or request good first issue Good for newcomers implementing algorithms Implementing algorithms, estimators, objects native to sktime implementing framework Implementing or improving framework for learning tasks, e.g., base class functionality module:forecasting forecasting module: forecasting, incl probabilistic and hierarchical forecasting labels Aug 30, 2021
@aiwalter
Copy link
Contributor

I added this also to the wish list, but lets keep track here then. I will pick the TransformedTargetForecaster as a start :). After my test refactor contained in #1347 is merged.

@aiwalter
Copy link
Contributor

aiwalter commented Sep 2, 2021 •
edited

can we add NaiveForecaster to this list? It could internally just apply a ColumnEnsembleForecaster, so would be nice to have this as a shorthand

@aiwalter
Copy link
Contributor

aiwalter commented Sep 2, 2021

What about ForecastingPipeline? Do we need to adjust it? Not sure if there is a use-case where y and X are both pd.DataFrame?

@fkiraly
Copy link
Collaborator Author

fkiraly commented Sep 3, 2021

Yes, I think that´s not an uncommon use case.

Btw, it looks like TransformedTargetForecaster works for multivairate out of the box?

@mloning
Copy link
Contributor

mloning commented Sep 3, 2021 •
edited

Btw, it looks like TransformedTargetForecaster works for multivairate out of the box?

No, still only works for univariate input data. The Theta discussion wasn't about the input data to the TransformedTargetForecaster. But should be relatively easy to refactor (I hope).

@aiwalter
Copy link
Contributor

aiwalter commented Sep 4, 2021 •
edited

I have a PR open for TransformedTargetForecaster and BaseGridSearch, see #1376 . Need to wait merging #1301 . I found sth that we could refactor: The check_y_X, check_y and check_X are not so handy now bcz we might rather directly call check_series as we can pass there

enforce_univariate=False,
enforce_multivariate=False,

separately, which we need now sometimes. maybe we can deprecate sth here, but would best be a separate PR.

@fkiraly fkiraly mentioned this issue Apr 13, 2022
Merged
Workstream: forecasting and series transformers automation moved this from To do to Done Apr 28, 2022
fkiraly added a commit that referenced this issue Apr 28, 2022
@fkiraly
[ENH] MultiplexForecaster compatibility with multivariate, probabil…
e77249f 
…istic and hierarchical forecasting (#2458)

This PR partially resolves #1364 and #2088, it enables `MultiplexForecaster` to work with multivariate, hierarchical, and/or probabilistic forecasters.

This is achieved by replacing manual loopthrough with:

* tag cloning
* delegation via the general `_DelegatedForecaster` (which includes the probabilistic methods and will also pick up any future method that is added)

This PR also contains:
* docstring improvements in the `_HeterogeneousMetaEstimator`
* a docstring fix, added missing default value for parameter `deep` in `get_params` throughout the code base
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
feature request New feature or request good first issue Good for newcomers implementing algorithms Implementing algorithms, estimators, objects native to sktime implementing framework Implementing or improving framework for learning tasks, e.g., base class functionality module:forecasting forecasting module: forecasting, incl probabilistic and hierarchical forecasting
Projects
Workstream: forecasting and series tr...
Done
Milestone
No milestone
3 participants
@fkiraly @mloning @aiwalter