tsfeatures adapter

[ltsaprounis]

Reference Issues/PRs

Fixes #905

What does this implement/fix? Explain your changes.

Implementing a tsfeatures adapter for sktime.
Will be a panel-to-tabular transformer, similar to tsfresh and catch22

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

Adding tsfeatures as a soft depeendency

What should a reviewer concentrate their feedback on?

Nothing yet, will update this section in the near future

Any other comments?

PR checklist

For all contributions

• I've added myself to the list of contributors.
• Optionally, I've updated sktime's CODEOWNERS to receive notifications about future changes to these files.
• I've added unit tests and made sure they pass locally.

For new estimators

• I've added the estimator to the online documentation.
• I've updated the existing example notebooks or provided a new one to showcase how my estimator works.

[ltsaprounis]

@mloning / @fkiraly - Some dependency issues with tsfeatures here.
Looks like the requirements in tsfeatures are too strict:

arch==4.14
git+https://github.com/raphaelvallat/entropy@v0.1.1
pandas==1.0.5
scikit-learn==0.23.1
statsmodels==0.11.1
supersmoother==0.4

(Also entropy is depreciated)

I've raised this as an issue in tsfeatures but it might take a while till it's resolved. Another option is to create the individual features as series-to-primitives transformers as suggested by @fkiraly in #905 (comment) and then create a panel-to-dataframe from those.

A complete checklist for the individual series to primitives transformers (copied from Combination of Forecast Methods by Feature-based Learning) would be:

• x_acf
• x_acf10
• diff1_acf1
• diff1_acf10
• diff2_acf1
• diff2_acf10
• seas_acf1
• ARCH.LM
• crossing_point
• entropy
• flat_spots
• arch_acf
• garch_acf
• arch_r2
• garch_r2
• alpha
• beta
• hurst
• lumpiness
• nonlinearity
• x_pacf5
• diff1x_pacf5
• diff2x_pacf5
• seas_pacf
• nperiods
• seasonal_period
• trend
• spike
• linearity
• curvature
• e_acf1
• e_acf10
• se
• peak
• trough
• stability
• hw_alpha
• hw_beta
• hw_gamma
• unitroot_kpss
• unitroot_pp
• series_length

[mloning]

Hey @ltsaprounis I'm not sure if the dependency issues have been resolved? Are you still thinking of finishing this PR? I may close this, of course you can always reopen if you want to pick it up again.

[ltsaprounis]

@mloning - the maintainers solved the dependencies shortly after I raised the issue but I forgot to get back to it. I'm not planning to work on this in the near future so makes sense to close it.

[AJarman]

@fkiraly I'd like to pick this up if possible

[fkiraly]

feel free to go ahead, @AJarman! Please post in the original issue #905, and have a look at the new transformers extension template. There´s also a rough tutorial for transformers, still work in progress, in this PR: #1705 (feedback is appreciated)

[fkiraly]

PS @AJarman, I see you´re helping out in pycaret with interval forecasts, so you may also be interested to contribute to work in sktime on interval and probabilistic forecasting?

[AJarman]

PS @AJarman, I see you´re helping out in pycaret with interval forecasts, so you may also be interested to contribute to work in sktime on interval and probabilistic forecasting?

All really did was implement croston inside a container, as it only had one parameter it wasn't a complicated implementation! I am still quite new to open source so working through 'first issues' at the moment, I've also applied to the mentorship program.a

[AzulGarza]

Hi @AJarman! Are you still working on this issue? Let me know if I can help in any way :)