Jigyasu Krishnan

[ENH] Add `ClassificationBenchmark` Class and its Helper Functions

1

_Stacks on #8347 and #8347 stacks on #8267, please merge in that order._ This PR adds `ClassificationBenchmark` class for benchmarking of classifiers. The design for it is mostly taken from...

[ENH] Missing Trainable Layers in `TimeLLMForecaster`

7

Paper: [https://arxiv.org/abs/2310.01728](https://arxiv.org/abs/2310.01728) Official code: [https://github.com/KimMeen/Time-LLM](https://github.com/KimMeen/Time-LLM) The `sktime` interface to `TimeLLM` (vendor) currently misses some implementation from the original architecture. * Reprogramming and mapping layers from the original implementation needs to...

This PR fixes the development-dependencies installation command so that it works universally across all terminals and operating systems. While setting up a development environment on macOS, I discovered that the...

**Which URL are your requesting to add?** https://jgyasu.github.io **Is the download size of the site you are adding smaller than 256kB in total?** - [X] Yes, I checked the size...

Stacks on #8749 Partially fixes #9077 This PR implements a catalogue (collection) of datasets, algorithms, metrics, and CV splitters used in the Bake Off 2017 study.

[ENH] `catalogues` in sktime

10

This PR introduces a new catalogues API for organising and accessing related groups of sktime objects (datasets, estimators, metrics, etc.). To understand the need and origin idea of catalogues, please...

[ENH] Inherit Splitters from `BaseSplitter`

3

Currently `PresplitFilesCV` and `SingleSplit` doesn't inherit from `BaseSplitter` which leads to them not getting discovered by the registry. This PR fixes it.

[ENH] Catalogues for Different Benchmarking Studies and Competitions

5

PR #8749 adds catalogue API, it includes the `BaseCatalogue` class and a dummy forecasting and classification catalogue along with tests. Details about the API can be found on the above...

[ENH] Reduce HF API calls when downloading datasets for a benchmarking study

4

sktime's older dataset downloader used to download datasets from URLs, with 1 dataset/URL. We then moved to use HuggingFace as our dataset hosts, to be more precise, we now use...

[ENH] Add Overall Weighted Average (OWA) Score

1

This PR adds overall weighted average score which was used in the M4 forecasting competition. The metric is defined as: `OWA = 0.5 * ( MASE / MASE_Naive2 + sMAPE...