timesfm-forecasting

Zero-shot time series forecasting is performed using Google's TimesFM foundation model, which generates predictions on univariate time series data directly without requiring you to train a custom model. It leverages pre-trained models with various parameter sizes and contexts to infer future values.

Yes, probabilistic forecasting with prediction intervals is supported using TimesFM's calibrated quantile-based predictions. This allows you to generate forecasts that include probability intervals, providing a range of likely future values rather than just single point estimates for your univariate time series data.

Yes, you need PyTorch installed because TimesFM requires it for model inference. Along with the TimesFM dependency, PyTorch handles the underlying tensor computations and model loading required to execute zero-shot forecasting on your univariate time series data.

The best way to forecast future sales from a CSV dataset is to apply zero-shot forecasting using TimesFM on your historical univariate sales data. This Skill processes the historical data and outputs future forecasts complete with quantile prediction intervals for probabilistic accuracy.

Yes, a mandatory system preflight check is required to verify your system resources before loading the TimesFM model. This preflight script ensures your environment has sufficient memory and compute capacity to handle the pre-trained model's various parameter sizes and contexts during inference.