v0.0.31

Features

• foundationforecast library: Foundation model implementations now live in the standalone foundationforecast package on PyPI (requires foundationforecast>=0.1.2 for Tafsut). TimeCopilot re-exports them at timecopilot.models.foundation.* for backward compatibility. Model docstrings and core forecaster logic are maintained in the foundationforecast repository. See #368.

• Tafsut foundation model: Added support for Tafsut, a zero-shot probabilistic univariate time series foundation model. Use it via the Tafsut class. See the model card.

  import pandas as pd
  from timecopilot.models.foundation.tafsut import Tafsut
  
  df = pd.read_csv(
      "https://timecopilot.s3.amazonaws.com/public/data/air_passengers.csv",
      parse_dates=["ds"],
  )
  
  model = Tafsut()
  fcst_df = model.forecast(df, h=12)

Breaking Changes

• GIFT-Eval moved to standalone package: The in-repo timecopilot.gift_eval module was removed. GIFT-Eval integration now lives in the timecopilot-gift-eval PyPI package (timecopilot_gift_eval import path). Install via pip install "timecopilot[gift-eval]" or pip install timecopilot-gift-eval>=0.3.0. See #369.

  Migration:

  from timecopilot.gift_eval.eval import GIFTEval
  from timecopilot.gift_eval.gluonts_predictor import GluonTSPredictor
  from timecopilot.gift_eval.utils import DATASETS_WITH_TERMS
  

  from timecopilot_gift_eval import GIFTEval, GluonTSPredictor
  from timecopilot_gift_eval.utils import DATASETS_WITH_TERMS

  See the GIFT-Eval API docs for details.

---

Full Changelog: v0.0.30...v0.0.31