cyberneticlibrary

Forecast time-series data without training

timesfm-forecastingskillsetup L227,559
K-Dense-AI/scientific-agent-skills
What it does

Forecast time-series data using Google's TimesFM foundation model

Best for

Zero-shot forecasting without manual feature engineering or hyperparameter tuning.

Inputs
  • · historical time-series data
  • · forecast horizon
Outputs
  • · point forecasts
  • · confidence intervals
  • · backtest metrics
Requires
  • · TimesFM
  • · NumPy
  • · Hugging Face Transformers
Preconditions

Historical data minimum 512 points; stationarity check recommended

Failure modes

Out-of-distribution extrapolation; low forecast quality on volatile data

Trust signals
  • · Google foundation model
  • · pretrained weights
  • · zero-shot capability