[ENH] Add v2 interface support for RecurrentNetwork (RNN) by Meet-Ramjiyani-10 · Pull Request #2136 · sktime/pytorch-forecasting

Meet-Ramjiyani-10 · 2026-03-04T12:18:23Z

Reference Issues/PRs

Closes #2128. Part of #1992.

What does this implement/fix? Explain your changes.

Adds v2 interface support for the RecurrentNetwork (RNN) model.

Added _rnn_v2.py implementing RecurrentNetwork_v2 using TslibBaseModel
Added _rnn_pkg_v2.py with tags, datamodule, and test parameters
Supports both LSTM and GRU cell types
Supports QuantileLoss for probabilistic forecasting

I followed the v2 pattern established by DLinear and SAMformer and have made no changes to existing v1 implementation.

What should a reviewer concentrate their feedback on?

forward pass input preparation
Whether the output shape is consistent with other v2 models
Tags in the pkg file

Did you add any tests for the change?

Test parameters added in get_test_train_params covering LSTM, GRU, single layer, and multi-layer configurations.

Any other comments?

Happy to iterate based on feedback.

PR checklist

[✓ ] The PR title starts with either [ENH], [MNT], [DOC], or [BUG]. [BUG] - bugfix, [MNT] - CI, test framework, [ENH] - adding or improving code, [DOC] - writing or improving documentation or docstrings.
[ ✓] Used pre-commit hooks when committing to ensure that code is compliant with hooks. Install hooks with pre-commit install.
To run hooks independent of commit, execute pre-commit run --all-files

codecov · 2026-03-04T13:12:30Z

Codecov Report

❌ Patch coverage is 98.66667% with 1 line in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (main@cf46d07). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
pytorch_forecasting/models/rnn/_rnn_v2.py	98.03%	1 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #2136   +/-   ##
=======================================
  Coverage        ?   86.72%           
=======================================
  Files           ?      167           
  Lines           ?     9806           
  Branches        ?        0           
=======================================
  Hits            ?     8504           
  Misses          ?     1302           
  Partials        ?        0

Flag	Coverage Δ
cpu	`86.72% <98.66%> (?)`
pytest	`86.72% <98.66%> (?)`

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

phoeenniixx

Thanks!
Added some comments

phoeenniixx · 2026-03-04T13:51:33Z

pytorch_forecasting/models/rnn/_rnn_pkg_v2.py

+        return TslibDataModule
+
+    @classmethod
+    def _get_test_datamodule_from(cls, trainer_kwargs):


I dont think you need this method anymore? Please have a look at other pkgs (like timexer) for more info

thanks for pointing. After checking _timexer_pkg_v2.py. this method is not needed . Will remove it

@phoeenniixx done 👍

phoeenniixx · 2026-03-04T13:51:58Z

pytorch_forecasting/models/rnn/_rnn_pkg_v2.py

+        """Get the underlying DataModule class."""
+        from pytorch_forecasting.data._tslib_data_module import TslibDataModule
+
+        return TslibDataModule


Should it be EncoderDecoderDataModule?

@phoeenniixx After checking _timexer_pkg_v2., it also uses TslibDataModule in get_datamodule_cls. Should I still switch to EncoderDecoderDataModule, or is TslibDataModule correct here?

Please switch to EncoderDecoderDataModule for models with v1 native implementations. The rule of thumb is to stick to TslibDataModule only for direct migrations of models from thuml(tslib).

@PranavBhatP
thanks for clarifying!. But it seems EncoderDecoderDataModule doesn't exist in the codebase yet. Should I keep TslibDataModule for now, or is there a different existing class I should use?

EncoderDecoderTimeseriesDataModule is the exact name of the D2 data module I am referring, it is interchangeably referred to as EncoderDecoderDataModule (u get what i mean?).

yes @PranavBhatP got it . I was searching for EncoderDecoderDataModule as a single class and couldn't find it , making the changes.. EncoderDecoderTimeSeriesDataModule

phoeenniixx · 2026-03-04T13:53:58Z

pytorch_forecasting/models/rnn/_rnn_pkg_v2.py

+        params : list of dict
+            Parameters to create testing instances of the class.
+        """
+        from pytorch_forecasting.metrics import MAE, SMAPE, QuantileLoss


Did you mean to use them in the params?
I think it would be good if we could chekc which type of losses this can support

i will update get_test_train_params to include MAE, SMAPE, and QuantileLoss directly in the params to verify loss compatibility, as in _timexer_pkg_v2.py

@phoeenniixx done 👍

…ate test params with losses

…ani-10/pytorch-forecasting into enh/rnn-v2-interface

… module

…ani-10/pytorch-forecasting into enh/rnn-v2-interface

Meet-Ramjiyani-10 · 2026-03-05T16:27:37Z

@phoeenniixx and @PranavBhatP
I’ve updated the RNN v2 package so that info:name is "RecurrentNetwork_v2", which fixes the first part of test_pkg_linkage. However, I’m still seeing failures from the second assertion:

assert object_pkg.name == object_class.name + "_pkg_v2", msg
For RNN v2 we currently have:

object_class.name == "RecurrentNetwork_v2"
object_pkg.name == "RecurrentNetwork_v2_pkg"
So the test expects "RecurrentNetwork_v2_pkg_v2", which doesn’t match the current naming. For the other v2 models (e.g. TFT, DLinear, Samformer), the class name stays the same and only the package class gets the _pkg_v2 suffix, so this rule fits them but not RNN v2.

Could you please advise what naming you’d prefer here? I see two options:

Rename the model class back to RecurrentNetwork and keep the package as RecurrentNetwork_pkg_v2 (matching the convention used for the other v2 models), or
Keep RecurrentNetwork_v2 / RecurrentNetwork_v2_pkg and relax/special‑case the test for this model.
Once I know which direction you’d like, I can adjust the code and tests accordingly.

phoeenniixx · 2026-03-05T17:57:59Z

Yes this is a known issue, see #2080 by @PalakB09. You can:

Either wait for some time and update once [ENH] Relax naming convention in test_pkg_linkage #2080 is merged
Or maybe name the model as RNN in place of RecurrentNetwork_v2 so that the names doesnt class with v1

The decision is depends on you
FYI @PranavBhatP @fkiraly

Meet-Ramjiyani-10 · 2026-03-05T18:13:51Z

okay, for now i will rename the model class to RNN

…onvention

[MNT] fix end of file and ruff formatting

da9b21a

Meet-Ramjiyani-10 requested review from PranavBhatP, benHeid, fkiraly, fnhirwa, jdb78, phoeenniixx and yarnabrina as code owners March 4, 2026 12:18

Merge branch 'main' into enh/rnn-v2-interface

d47c878

phoeenniixx requested changes Mar 4, 2026

View reviewed changes

phoeenniixx added module:models enhancement New feature or request labels Mar 4, 2026

Meet-Ramjiyani-10 added 12 commits March 4, 2026 22:25

[ENH] address review comments - remove _get_test_datamodule_from, upd…

825b12d

…ate test params with losses

Merge branch 'enh/rnn-v2-interface' of https://github.com/Meet-Ramjiy…

8e26e9d

…ani-10/pytorch-forecasting into enh/rnn-v2-interface

Merge branch 'main' into enh/rnn-v2-interface

3498d10

Merge remote-tracking branch 'upstream/main' into enh/rnn-v2-interface

e32bb6f

[ENH] export RecurrentNetwork_v2 and RecurrentNetwork_pkg_v2 from rnn…

2c3d47e

… module

Merge branch 'enh/rnn-v2-interface' of https://github.com/Meet-Ramjiy…

57721fb

…ani-10/pytorch-forecasting into enh/rnn-v2-interface

[BUG] fix info:name tag in RecurrentNetwork_pkg_v2 to match class name

ddb218d

[BUG] fix info:name tag and switch to EncoderDecoderTimeSeriesDataModule

e15c963

[ENH] update RecurrentNetwork_v2 to use BaseModel and encoder_cont keys

d56dab1

[BUG] fix _pkg method to use renamed package class

fc53b3e

[BUG] fix info:name tag in RecurrentNetwork_pkg_v2 to match class name

df73720

[BUG] fix info:name tag in RecurrentNetwork_pkg_v2 to match class name

014f63d

Meet-Ramjiyani-10 added 2 commits March 5, 2026 23:44

[ENH] rename RNN v2 class to RNN and pkg to RNN_pkg_v2 following v2 c…

da1025b

…onvention

Merge branch 'main' into enh/rnn-v2-interface

e0b007a

Meet-Ramjiyani-10 requested a review from phoeenniixx March 6, 2026 19:03

Conversation

Meet-Ramjiyani-10 commented Mar 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Reference Issues/PRs

What does this implement/fix? Explain your changes.

What should a reviewer concentrate their feedback on?

Did you add any tests for the change?

Any other comments?

PR checklist

Uh oh!

codecov bot commented Mar 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

phoeenniixx 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!

Meet-Ramjiyani-10 Mar 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

PranavBhatP Mar 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Meet-Ramjiyani-10 Mar 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

PranavBhatP Mar 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

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!

Meet-Ramjiyani-10 commented Mar 5, 2026

Uh oh!

phoeenniixx commented Mar 5, 2026

Uh oh!

Meet-Ramjiyani-10 commented Mar 5, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Meet-Ramjiyani-10 commented Mar 4, 2026 •

edited

Loading

codecov bot commented Mar 4, 2026 •

edited

Loading

Meet-Ramjiyani-10 Mar 4, 2026 •

edited

Loading

PranavBhatP Mar 4, 2026 •

edited

Loading

Meet-Ramjiyani-10 Mar 4, 2026 •

edited

Loading

PranavBhatP Mar 5, 2026 •

edited

Loading