[Time series] Add PatchTSMixer

[ajati]

PatchTSMixer (KDD 2023) is a lightweight time-series modeling approach based on the MLP-Mixer architecture. In this HuggingFace implementation, we provide PatchTSMixer's capabilities to effortlessly facilitate lightweight mixing across patches, channels, and hidden features for effective multivariate time-series modeling. It also supports various attention mechanisms starting from simple gated attention to more complex self-attention blocks that can be customized accordingly. The model can be pretrained and subsequently used for various downstream tasks such as forecasting, classification and regression.

@kashif

Done: TODOs

• Add generate method
• Make pretrained dataset publicly available
• Make pretrained weights publicly available

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint.

[vijaye12]

@amyeroberts - Greetings! We have enabled all the changes suggested in PatchTST and also the corrections suggested in this PR from the past reviewers. Requesting your review and approval.

[ArthurZucker]

Thanks this looks a lot better than last time. Left some small nits but should be good to go otherwise

[ArthurZucker]

still not resolved

[vijaye12]

Hi @ArthurZucker -Thanks for approving this PR. We have resolved all the final changes you mentioned. Please review and help with merge, if all good.

PS: change with respect to adding segment names to docstring is the only one comment pending. When we add segment names in docstring following the syntax ssuggested - it getting auto removed during make fix-ups. Other than this - all other comments are resolved.

[ArthurZucker]

@vijaye12 thanks for bearing with us in this long review! and congrats for the merge! 🚀