PCR-SAITS

PCR-SAITS is a lightweight residual-correction layer for multivariate time-series imputation. The public package exposes the same audited PCR core with SAITS and BRITS backbones through a small user-facing API.

Paper:

PCR-SAITS: A Lightweight Disagreement-Based Residual Corrector for SAITS-Based Multivariate Time Series Imputation
Sawet Somnugpong, Expert Systems with Applications (2026)
DOI: https://doi.org/10.1016/j.eswa.2026.134510

Install

From this source tree:

python -m pip install .

Core dependencies are numpy, torch, and pypots.

Public API

from pcrsaits import PCRSAITS, PCRBRITS

Both are thin wrappers around the legacy-equivalence-tested PCRCorrector. A trained SAITS/BRITS backbone is supplied to the PCR wrapper; the PCR layer does not retrain or own the backbone checkpoint.

For a new dataset, explicit feature metadata is the default:

model = PCRSAITS(
    backbone=trained_saits,
    feature_names=["PM2.5", "TEMP", "WSPM"],
    feature_groups=["pollutant", "meteorological", "meteorological"],
)
model.fit(train_values, val_values, seed=7)
imputed = model.impute(masked_values)

Allowed public feature groups are:

• pollutant
• sensor
• meteorological

To reproduce legacy paper-suite name inference, use:

model = PCRSAITS(
    backbone=trained_saits,
    feature_names=["CO(GT)", "PT08.S1(CO)", "T"],
    metadata_mode="paper_legacy_inference",
)

Important inference behavior

PCR correction is applied to every cell that is missing in the supplied input. Originally observed cells are restored exactly.

Save / load

Backbone and PCR checkpoints are intentionally separate:

trained_saits.save("saits_backbone.pypots")
model.save("pcrsaits.pt")

After restoring the backbone:

model = PCRSAITS.load("pcrsaits.pt", backbone=restored_saits)

Examples

• examples/quickstart_pcrsaits.py
• examples/quickstart_pcrbrits.py

Paper reproduction

Historical experiment programs supplied by the author are preserved under paper_reproduction/legacy_scripts/ with a SHA256 manifest.

python paper_reproduction/verify_sources.py

Final reviewer-specific Table 17 and patient-aware PhysioNet source identities are recorded separately so that older protocols are not silently presented as the final paper protocol.

See:

• REPRODUCIBILITY.md
• paper_reproduction/README.md
• paper_reproduction/PHYSIONET_PROTOCOL.md

Citation

See CITATION.cff.

Software archive DOI: https://doi.org/10.5281/zenodo.22973879

Release status

This tree is the v1.0.0 release payload prepared for final pre-tag audit. The Git tag and archival deposit should be created only after the clean ZIP, wheel, and sdist hashes are independently verified.

License

MIT License. See LICENSE.