Arimancy

Updates

Follow along in a feed reader: Atom feed

  1. 2026-07-22

    Five datasets public and free, and an ERCOT correction

    All five Arimancy datasets are now public and free on Kaggle and Hugging Face under CC BY 4.0.

    ERCOT Hourly Demand + Population-Weighted Weather: Texas grid demand joined to population weighted weather, hourly, July 2015 to July 2026.

    CAISO Hourly Demand + Population-Weighted Weather: the same schema for the California grid, hourly, July 2015 to July 2026.

    PJM Hourly Demand + Population-Weighted Weather: the same schema for the PJM Interconnection, hourly, July 2015 to July 2026.

    CFTC Commitments of Traders Weekly Panel: every COT report family in one weekly panel, 1986 to the most recent release, updated weekly.

    Observed Weather for Every US Flight: every US domestic flight from January 2019 to July 2025 joined to observed weather at both airports.

    ERCOT changed in this release, and anyone holding the earlier version should re-download it. The weather provenance was corrected to pinned ERA5 rather than a blended model selection, which changed both values and the schema from the originally published version.

    The measured accuracy moved from 4.66% to 4.84% MAPE as a result, on a LightGBM baseline trained through 2024-12-31 and tested from 2025-01-01 onward. The number got slightly worse because the correction removed a defect that had been flattering the model. We publish the worse number because checkable accuracy claims are the point.

  2. 2026-07-05

    ERCOT Hourly Demand + Weather — v1 release

    The first Arimancy dataset is live: ERCOT hourly demand joined to population-weighted weather, spanning 2015–2026 (96,537 rows, 27 columns). A LightGBM baseline reaches 4.66% MAPE against an 8.80% seasonal-naive benchmark. Available on Hugging Face and Kaggle under CC BY 4.0, with every imputed value flagged in the data and full source attribution to EIA, Catalyst Cooperative's PUDL, and Open-Meteo.