somfieao/tsf-tta-artifacts

TSF-TTA Artifacts

Artifacts for a time-series-forecasting test-time-adaptation research project. Total ~91 GB. Local filenames are preserved; see layout below.

Layout

datasets/
  real/           # 14 public benchmark datasets as CSVs (weather, ETh1/h2, ETm1/m2,
                  #   electricity, traffic, solar, exchange_rate, PEMS03/04/07/08)
  lorenz96/       # generated Lorenz-96 cache (.npz)
  hybrid_tafas/   # preprocessed PyTorch tensors (.pt) for the hybrid TAFAS experiments (54 GB)

checkpoints/
  sources/        # backbone source models (source_patchtST_*, fullb_source_{backbone}_*)
  methods/
    full_benchmark/  # fullb_{steps,modal,J}_{backbone}_{dataset}_* method artifacts
    reskoop/         # reskoop_* / longf_reskoop_* correctors
    directy/         # directy_patchtst_{e2e,y2e,y2y}_*
    longf/           # longf_patchtst_*
    variants/        # modal*, steps*, nokoop* method variants
  ablations/      # resablation_* architectural ablations
  experiments/    # hv14_*, hv4_*, hybrid_tafas_* hypothesis-validation checkpoints
    lorenz96/     # koopman_Lorenz96_*, reskoop_Lorenz96_* (early experiments)

results/
  main/           # full_benchmark campaign (MASTER.csv, pivot tables, per-{backbone,dataset} CSVs)
  experiments/    # exploratory analysis outputs (cr_lambda_*, reskoop_* diagnostics, tsne/adf_psd .npz)

Download

pip install -U "huggingface_hub[cli]"
# everything (use a persistent-allowlisted disk; ~91 GB):
hf download somfieao/tsf-tta-artifacts --repo-type model --local-dir .
# or just one subtree, e.g. main results:
hf download somfieao/tsf-tta-artifacts results/main --repo-type model --local-dir .