Skip to content

v2.0.0

Choose a tag to compare

@Ladbaby Ladbaby released this 04 Apr 12:32
· 20 commits to main since this release

We are excited to introduce PyOmniTS v2.0.0, featuring our new model ReIMTS (ICLR 2026)!
Thanks to improved hyperparameter search capabilities, nearly all models now deliver superior performance compared to v1.0.0.

Due to recently identified and corrected bug in the USHCN dataset, we have rerun all experimental results related to USHCN. Therefore, these results in our HyperIMTS paper are no longer accurate, and we have updated them in our ReIMTS paper. You can also check the latest results in our time series analysis leaderboard.

These resources are newly added:

✨ New Features

Core Framework

  • main: PyOmniTS now supports classification tasks in addition to forecasting and imputation. commit
  • main: UVM (Unified Virtual Memory) can be disabled when batch size is not 1 for better memory management. commit
  • configs: --scaleformer_scales option added for ScaleFormer architecture configuration. commit
  • configs: ReIMTS-related arguments added for the new ReIMTS model. commit

Datasets

  • datasets: Imputation task support added to data loading pipeline. commit
  • datasets: sample_ID support added for tracking individual samples. commit
  • datasets: Imputation support for irregular time series datasets. commit
  • datasets: collate_fn_fractal added for Fractal dataset processing. commit

Models - Imputation Support

The following models now support imputation tasks:

  • Ada_MSHyper: Imputation and classification support. commit
  • AutoformerMS: New model with multi-scale autoencoding architecture. commit
  • BigST: Imputation and classification support. commit
  • CRU: Imputation support. commit
  • DLinear: Imputation support. commit
  • ETSformer: Imputation support. commit
  • FEDformer: Imputation support. commit
  • FiLM: Imputation support. commit
  • FourierGNN: Imputation and classification support. commit
  • FreTS: Imputation support. commit
  • GRU_D: Imputation support. commit
  • HD_TTS: New model for hierarchical temporal representation learning. commit
  • Hi_Patch: Imputation support. commit
  • higp: Classification support. commit
  • iTransformer: Imputation support. commit
  • Koopa: Imputation support. commit
  • Latent_ODE: Imputation support. commit
  • Leddam: Imputation support. commit
  • LightTS: Imputation support. commit
  • MambaSimple: Imputation support. commit
  • MICN: Imputation support. commit
  • MOIRAI: Imputation support. commit
  • mTAN: Imputation and classification support. commit
  • NeuralFlows: Imputation support. commit
  • Nonstationary_Transformer: Imputation support. commit
  • PrimeNet: Imputation support. commit
  • Reformer: Imputation support. commit
  • ReIMTS: New model with representation learning for imputation tasks. commit
  • SegRNN: Imputation support. commit
  • TimeCHEAT: New model with temporal cheat architecture. commit
  • TimeMixer: Imputation support. commit
  • TSMixer: Imputation support. commit
  • Transformer: Imputation support. commit
  • Warpformer: Imputation support. commit

Models - Classification Support

The following models now support classification tasks:

  • Ada_MSHyper: Classification support. commit
  • BigST: Classification support. commit
  • FourierGNN: Classification support. commit
  • higp: Classification support. commit
  • mTAN: Classification support. commit
  • tPatchGNN: Classification support. commit

Models - New


🐛 Bug Fixes

Core Framework

  • main: Fixed --test_train_time and --test_gpu_memory to handle CUDA OOM errors properly. commit
  • main: Fixed set_per_process_memory_fraction configuration for better GPU memory management. commit
  • main: Fixed --wandb 1 integration issues. commit
  • exp: Fixed NaN detection during training. commit
  • exp: Fixed --load_checkpoints_test 0 parameter handling. commit
  • utils: Fixed EarlyStopping callback logic. commit

Models

  • Ada_MSHyper: Speed optimization for faster training. commit
  • Autoformer: Fixed model implementation issues. commit
  • Crossformer: Fixed model implementation issues. commit
  • GraFITi: Fixed model implementation issues. commit
  • GRU_D: Fixed imputation task handling. commit
  • HyperIMTS: Reverted representation learning changes and fixed comments. commit
  • Informer: Fixed imputation task handling. commit
  • PatchTST: Fixed model implementation issues. commit
  • PrimeNet: Fixed --sweep 1 parameter handling. commit
  • Pyraformer: Fixed imputation task handling. commit
  • Raindrop: Fixed imputation task handling. commit
  • ReIMTS: Fixed logging issues. commit
  • SeFT: Fixed model implementation issues. commit
  • Warpformer: Fixed model implementation issues. commit

Scripts & Datasets

  • scripts: Fixed SegRNN script for irregular time series datasets. commit
  • scripts: Fixed HyperIMTS training script. commit
  • scripts: Fixed P12 dataset script. commit
  • scripts: Fixed MIMIC_IV dataset script. commit
  • scripts: Fixed MIMIC_III dataset script. commit
  • scripts: Fixed HumanActivity dataset script. commit
  • scripts: Fixed ETTh1.sh script. commit
  • scripts: Fixed FourierGNN training script. commit
  • scripts: Fixed Hi_Patch training scripts (multiple fixes). commits, commit
  • scripts: Fixed tPatchGNN training script. commit
  • scripts: Fixed CRU training script. commit
  • scripts: Fixed NeuralFlows training script. commit
  • scripts: Fixed SeFT training script. commit
  • scripts: Fixed TimeMixer training script. commit
  • scripts: Fixed Crossformer training script. commit
  • scripts: Fixed Leddam training script. commit
  • scripts: Fixed PatchTST training script. commit
  • scripts: Fixed Pyraformer training script. commit
  • scripts: Fixed TimesNet training script. commit
  • scripts: Fixed Autoformer training script. commit
  • scripts: Fixed Ada_MSHyper training script. commit
  • scripts: Fixed GraFITi training script. commit
  • scripts: Fixed Raindrop training script. commit
  • scripts: Fixed GRU_D training script. commit
  • scripts: Fixed mTAN training script. commit
  • scripts: Fixed PrimeNet training script. commit

📚 Documentation

  • README: Updated README with framework and agent details. commit
  • README: Multiple README updates for improved clarity and completeness. commits, commit, commit, commit
  • docs: New documentation added. commit
  • images: Updated project images and icons. commit