Description of changes:
Adding support for including time features in the PachTST model. The original PatchTST paper does not talk about time-based features and there are several ways one could include them.
Here we go with this simple approach (and maintain the simplicity of the PatchTST model):
Similar to target time series, time series features, if provided, are converted to patches and appended to the target patches.
Since PatchTST is encoder only model, we combine past_time_feat and future_time_feat (this dichotomy is an artefact of gluonts) and (left) shift them by prediction_length so that the features are aligned with the target input.
Added tests for the PatchTSTEstimator with features enabled.
By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.
Please tag this pr with at least one of these labels to make our release process faster: BREAKING, new feature, bug fix, other change, dev setup
Description of changes: Adding support for including time features in the
PachTST
model. The originalPatchTST
paper does not talk about time-based features and there are several ways one could include them.Here we go with this simple approach (and maintain the simplicity of the
PatchTST
model):PatchTST
is encoder only model, we combinepast_time_feat
andfuture_time_feat
(this dichotomy is an artefact ofgluonts
) and (left) shift them byprediction_length
so that the features are aligned with thetarget
input.Added tests for the
PatchTSTEstimator
with features enabled.By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.
Please tag this pr with at least one of these labels to make our release process faster: BREAKING, new feature, bug fix, other change, dev setup