[ENH] fABBA Transform #8838

poopsiclepooding · 2025-09-22T20:11:39Z

What does this implement/fix? Explain your changes.

This code implements fABBA transform as found in https://github.com/nla-group/fABBA/tree/master .
fABBA is a time series symbolic representation transform, it compresses and digitizes the time series
into series of symbols.
Have added a python implementation of fABBA along with example usage.

Depends on #8959 for the unequal length output handling, #8959 should be merged first.

Does your contribution introduce a new dependency? If yes, which one?

No new dependency.

What should a reviewer concentrate their feedback on?

Any formatting or major code issues. Any errors with particular combination of hyperparameters

Did you add any tests for the change?

Haven't added any particular test. The example provided illustrates the working of algorithm. If any test suggestions are there I can add them.

Any other comments?

Original fABBA implementation was really helpful to write this - https://github.com/nla-group/fABBA/tree/master

For all contributions

I've added myself to the list of contributors with any new badges I've earned :-)
How to: add yourself to the all-contributors file in the sktime root directory (not the CONTRIBUTORS.md). Common badges: code - fixing a bug, or adding code logic. doc - writing or improving documentation or docstrings. bug - reporting or diagnosing a bug (get this plus code if you also fixed the bug in the PR).maintenance - CI, test framework, release.
See here for full badge reference
Optionally, for added estimators: I've added myself and possibly to the maintainers tag - do this if you want to become the owner or maintainer of an estimator you added.
See here for further details on the algorithm maintainer role.
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.

For new estimators

I've added the estimator to the API reference - in docs/source/api_reference/taskname.rst, follow the pattern.
I've added one or more illustrative usage examples to the docstring, in a pydocstyle compliant Examples section.

…to fABBA

review-notebook-app · 2025-09-22T20:11:44Z

Check out this pull request on

See visual diffs & provide feedback on Jupyter Notebooks.

Powered by ReviewNB

sktime/transformations/series/fabba.py

fkiraly

Nice! Welcome to open source!

Some comments above.

Further questions/requests:

code formatting seems to fail - if you want to know how to ensure code formatting is done automatically on your computer, follow this guide: https://www.sktime.net/en/latest/developer_guide/coding_standards.html
this looks like a new implementation of fabba - I think there is one by the authors? This could be used, through an import: https://github.com/nla-group/fABBA/blob/master/fABBA/fabba.py no? It is of course legitimate to carry out a new implementation, I am just wondering why you chose to reimplement? Or is this partly copied code?

poopsiclepooding · 2025-09-22T20:35:45Z

Will look into code formatting.
The implementation in original branch doesn't work for multiple time series of different lengths so I had changed that.
Also I wanted option to export symbols as int(labels of clusters) rather than just char so implemented that as well.
I used the original repo as reference for main algorithms.

fkiraly · 2025-09-22T21:02:30Z

I used the original repo as reference for main algorithms.

I see - we should cite it and the paper then.

poopsiclepooding · 2025-09-22T21:05:19Z

Should I add the repo citation below the paper citation or somewhere else as well?

fkiraly · 2025-09-22T21:08:14Z

Should I add the repo citation below the paper citation or somewhere else as well?

Once in References, and then from the main text, I would say.

chenxinye · 2025-09-25T19:59:11Z

Hi @fkiraly and @poopsiclepooding,

Thanks @poopsiclepooding for implementing it, later I will also update my repo to support length variability for time series.

I think my paper [1] has mentioned this method can be easily extended to arbtrary length time series:

I suggest the citations can be:
[1] X. Chen, 2024. Joint symbolic aggregate approximation of time series
[2] X. Chen and S. Güttel., 2024. fABBA: A Python library for the fast symbolic approximation of time series. Journal of Open Source Software
[3] X. Chen and S. Güttel. 2023. An Efficient Aggregation Method for the Symbolic Representation of Temporal Data. ACM Trans. Knowl. Discov. Data
[4] S. Elsworth and S. Güttel. 2023. ABBA: adaptive Brownian bridge-based symbolic aggregation of time serie

Thanks,
Xinye

poopsiclepooding · 2025-09-26T08:07:04Z

Thanks a lot @chenxinye . I will add those citations

fkiraly · 2025-09-27T23:49:46Z

The failures seem to indicate that a single time series is getting transformed intio multiple. Is this expeted behaviour? If yes, we need to set tags differently.

poopsiclepooding · 2025-09-28T12:13:23Z

Yes that is expected behaviour. If we set partition hyperparameter then time series is broken into parts before apply fABBA and each part's symbolic representation is returned. What should I change the tags to?

fkiraly · 2025-09-30T18:18:36Z

there is currently no valid tag for this behaviour, we will have to look into it on framework level. May I suggest to skip the failing tests for now via tests:skip_by_namd and raise an issue that explicitly refers to the issue:

if an mtype is passed that can only represent equal length, and the transform changes it to unequal length, the transform tries to convert it back to the original mtype which fails. We need to have a design decision on this.

…to pr/8838

fkiraly · 2025-10-10T19:08:47Z

@poopsiclepooding, I have updated the framework to handle transformers that turn series into unequal length, here: #8959

I will merge #8959 into this Pr and the tests for the fabba transformer should pass now.

FYI, I changed the name to FABBA since classes need to start with capital letters.

sktime/transformations/series/fabba.py

fkiraly · 2025-10-10T22:22:20Z

I think the example is now failing, see the failing tests.

poopsiclepooding · 2025-10-10T22:33:11Z

I think the example is now failing, see the failing tests.

Yes fixed

poopsiclepooding and others added 8 commits September 22, 2025 13:01

addded draft implementation of fABBA alogrithm

314c260

converted transform to series-to-series type

a74c436

Added examples, contribution and docs

0bf38cf

Merge branch 'main' into fABBA

a9c206b

[AUTOMATED] update CONTRIBUTORS.md

97aed3b

Added raise error and init docstring

06f87bd

Merge branch 'fABBA' of https://github.com/poopsiclepooding/sktime in…

ebc0f64

…to fABBA

Removed todos

f62138f

poopsiclepooding requested review from benHeid, felipeangelimvieira, fkiraly and yarnabrina as code owners September 22, 2025 20:11

fkiraly changed the title ~~[ENH] Added fABBA Transform~~ [ENH] fABBA Transform Sep 22, 2025

fkiraly reviewed Sep 22, 2025

View reviewed changes