Changelog

0.2.3

  • Add probabilistic forecasting via ProbForecastExp.

  • Add probabilistic metrics in torch_timeseries.metrics: CRPS, CRPSSum, QICE, PICP, ProbMSE, ProbMAE, and ProbRMSE.

  • Document WindowConfig.fast_val and fast_test for non-overlapping validation/test windows.

  • Refactor forecast batch handling to structured TSBatch inputs.

0.2.2

  • Add build_dataset(csv=..., freq=...) for local CSV datasets.

  • Infer dataset num_features and length from loaded data.

  • Remove the runtime torchvision dependency.

  • Add README guidance for custom datasets and compact curated leaderboard entries.

0.2.1

  • Add per-task leaderboard reproduce scripts for anomaly detection, imputation, long-term forecast, short-term forecast, and UEA classification across DLinear, NLinear, PatchTST, iTransformer, TimesNet, Autoformer, and FEDformer.

  • Add NLinear model and experiment wrappers.

  • Add TimesNet imputation and anomaly detection experiment wrappers.

  • Add README examples for researcher workflows using custom dataloaders, custom training loops, and default experiment registration.

  • Add default ForecastExp batch handling for simple models that map batch_x to forecast predictions.

  • Normalize anomaly metric names for leaderboard rendering.