This software package accompanies the manuscript Exhaustive Reassignment Reveals Identity Dependence in MOS Gas Sensor Correction Tables by Yibo Wang, Zhixian Zhao, and Yu Tang. Software and reproduction The package reproduces the numerical results for exhaustive reassignment of fixed metal-oxide-semiconductor (MOS) gas-sensor correction tables. It includes a Python analysis module, compact derived numerical tables covering 5,040 internal development assignments, 5,040 internal reserved assignments, and 24 external assignments, and a focused reproduction test suite. The analysis produces assignment summaries, group-level bootstrap intervals and sign counts, displacement correlations, external Path 1/3 pair statistics, and summaries of deterministic bound checks. The reproduction workflow starts from the frozen derived tables bundled in the archive. The underlying raw SGP40 datasets are available separately at 10.5281/zenodo.15423482 and 10.5281/zenodo.6821340. Software: mos-correction-reassignment, version 1.0.0. Requirements: Python 3.10 or later and NumPy 2.x. Run from the extracted package root with python -m mos_correction_reassignment --output-dir results. Associated manuscript abstract Unit-specific correction tables for metal-oxide semiconductor (MOS) gas sensors pair numerical offsets with the device or signal path for which they were estimated. We asked whether the same offsets remain useful after reassignment. Using two public SGP40 datasets for ethanol concentration prediction, we estimated each offset once from designated calibration records. We then tested every possible assignment while holding the Ridge regression model, features, parameter count, and evaluation records fixed. In data from two calibrations about one year apart, the matched four-pixel device vector was the unique minimum-error assignment among 5,040 possibilities in both evaluation sets. Across manufacturing batches, the matched path assignment outperformed 22 of 23 alternatives and reduced normalized mean absolute error by 9.67% relative to Ridge regression. At the median reassignment, the same correction table increased error above Ridge. These results show that device or path identity and channel order are part of the calibration information carried by a correction table. Those identifiers should remain attached to unit-specific corrections during storage, transfer, and deployment. Author contact information Yibo Wang: wangyibo2026@buaa.edu.cnZhixian Zhao: zhixianzhaosjtu@163.comYu Tang (corresponding author): y.tang@rioh.cn Funding This work was supported by the Key R&D Program of Shandong Province, China (Project No. 2025CXGC010110). Competing interests The authors declare no competing interests. License Creative Commons Attribution 4.0 International (CC BY 4.0).
This software package accompanies the manuscript Exhaustive Reassignment Reveals Identity Dependence in MOS Gas Sensor Correction Tables by Yibo Wang, Zhixian Zhao, and Yu Tang. Software and reproduction The package reproduces the numerical results for exhaustive reassignment of fixed metal-oxide-semiconductor (MOS) gas-sensor correction tables. It includes a Python analysis module, compact derived numerical tables covering 5,040 internal development assignments, 5,040 internal reserved assignments, and 24 external assignments, and a focused reproduction test suite. The analysis produces assignment summaries, group-level bootstrap intervals and sign counts, displacement correlations, external Path 1/3 pair statistics, and summaries of deterministic bound checks. The reproduction workflow starts from the frozen derived tables bundled in the archive. The underlying raw SGP40 datasets are available separately at 10.5281/zenodo.15423482 and 10.5281/zenodo.6821340. Software: mos-correction-reassignment, version 1.0.0. Requirements: Python 3.10 or later and NumPy 2.x. Run from the extracted package root with python -m mos_correction_reassignment --output-dir results. Associated manuscript abstract Unit-specific correction tables for metal-oxide semiconductor (MOS) gas sensors pair numerical offsets with the device or signal path for which they were estimated. We asked whether the same offsets remain useful after reassignment. Using two public SGP40 datasets for ethanol concentration prediction, we estimated each offset once from designated calibration records. We then tested every possible assignment while holding the Ridge regression model, features, parameter count, and evaluation records fixed. In data from two calibrations about one year apart, the matched four-pixel device vector was the unique minimum-error assignment among 5,040 possibilities in both evaluation sets. Across manufacturing batches, the matched path assignment outperformed 22 of 23 alternatives and reduced normalized mean absolute error by 9.67% relative to Ridge regression. At the median reassignment, the same correction table increased error above Ridge. These results show that device or path identity and channel order are part of the calibration information carried by a correction table. Those identifiers should remain attached to unit-specific corrections during storage, transfer, and deployment. Author contact information Yibo Wang: wangyibo2026@buaa.edu.cnZhixian Zhao: zhixianzhaosjtu@163.comYu Tang (corresponding author): y.tang@rioh.cn Funding This work was supported by the Key R&D Program of Shandong Province, China (Project No. 2025CXGC010110). Competing interests The authors declare no competing interests. License Creative Commons Attribution 4.0 International (CC BY 4.0).
