When is AutoML worth it? Data and code for the temporal validation and metrological assessment of low-cost greenhouse sensor calibration.

Data and analysis pipeline for the manuscript "When is AutoML worth it? Temporal validation and metrological assessment of low-cost greenhouse sensor calibration" (Rocha, dos Santos Safre, de Sousa, Guerrini; submitted to Measurement, 2026). Five low-cost air-temperature and relative-humidity sensor types (BMP180, BME280, DHT11, DHT22, LM35; three units each) were collocated with a Vaisala HMP45C reference probe in a commercial greenhouse (Botucatu, SP, Brazil, 11-15 November 2018). The archive contains the raw records of the reference logger and of the three low-cost units (with SHA-256 hashes), the 5-min datasets (original and clock-corrected), every script of the analysis (calibration models, validation protocols, nested AutoML search with TPOT, block-bootstrap and Diebold-Mariano tests, equivalence analysis, replica ablation, dynamic response, energy and cost budgets) and the result files from which the tables and figures of the article are generated. Paths are relative; see README.md for the execution order and requirements.txt for the pinned software versions. Code: MIT License. Data and results: CC BY 4.0.