•
sktime v1.0.2 (26 Jul 2026) adds many forecasting foundation model interfaces — Aurora, Cisco TSFM, Falcon-X, Moirai 2.0, Sundial, Toto-2.0, WindFM, MIRA, and more. Hook is the release list, not a Chronos/TimesFM mashup. Not investment advice.
sktime v1.0.2 (feature release, 26 July 2026) is a large add of forecasting foundation model wrappers under one familiar fit/predict interface. The hook for this Hub post is the release highlight list, not a mashup of our separate Chronos-2 / TimesFM / transformer paper notes. Not investment advice.
The only course CTA is Introduction to Python Programming.
From the v1.0.2 GitHub release (see also the changelog):
Also in the same release (not tonight’s rabbit hole): Arps decline curve forecasters, MiniRocketMultivariateCython, regression model_evaluation / RegressionBenchmark, EvoForestTSWM, Chen & Yang (2004) accuracy metrics, and more.
pip install -U "sktime==1.0.2"
python -c "import sktime; print(sktime.__version__)"
1. Confirm the install and browse estimators (names vary by optional extras):
import sktime
from sktime.registry import all_estimators
print(sktime.__version__)
# Filter locally for foundation / deep learning style tags if present in your build:
ests = all_estimators(filter_tags={"object_type": "forecaster"})
print("forecasters:", len(ests))
print([name for name, _ in ests if any(k in name.lower() for k in ("moirai", "aurora", "sundial", "toto", "wind", "falcon", "mira", "tsfm"))][:20])
2. Read one estimator’s docstring before downloading weights — many foundation models need extra packages / GPU / HF tokens. Start with the class that is actually importable in your env.
3. Keep classical baselines nearby. If a foundation wrapper fails to import, fall back to a classical sktime forecaster you already know (NaiveForecaster, AutoARIMA, etc.) so the night still ends with a printed forecast on a tiny synthetic series — not a live trading feed.
Stop after listing available names or one successful import. Do not bundle Chronos-2 / TimesFM Hub posts into this note.
If Python environments still fight you, Introduction to Python Programming, then pip install sktime==1.0.2 and print the filtered forecaster name list.