Forecast your time series without model training.
Produces zero-shot time-series forecasts with quantile bands and guides held-out evaluation using TimesFM.
When to use it
Use for forecasting an ordered temporal target without training a task-specific model, not for causal estimation, clinical detection or physics-based climate modeling.
Give it a time-series CSV and forecast horizon; it writes forecasts and a metadata sidecar to local files.
What you provide
Requires Python 3.10+ for the main workflow; the optional XReg workflow with current JAX requires Python 3.12+.
Uses uv to create the separate environment and install the reviewed packages.
This skill
google/timesfm-2.5-200m-pytorch
Downloads pretrained TimesFM weights
uv pip install
Installs the forecasting dependencies