ai4chems/nist-logv

该数据集名为NIST对数粘度（基础数据 + Redlich-Kister LOOCV拟合），是一个用于化学混合物粘度回归分析的两层结构数据集。基础数据层（points）包含约15万行测量数据，每行代表一个二元混合物的粘度测量点，列包括：mixture_T_id（连接键）、value（对数粘度）、T（温度，单位K）、P（压力，单位bar）、cmp_ids（长度为2的化合物ID列表）和cmp_mole_fractions（与cmp_ids对齐的摩尔分数列表）。数据集还提供compounds.csv文件，用于将化合物ID映射到SMILES表示，以及fits/rk_loocv.csv文件，存储了基于Redlich-Kister模型和留一交叉验证（LOOCV）方法的拟合结果，包括拟合参数和标准列（如pure_at_x0、pure_at_x1、nonlinearity）。数据集来源于NIST对数粘度基准，通过chemixhub处理，并过滤了多来源的测量箱以确保数据一致性。它适用于化学信息学、机器学习回归任务和混合物性质预测研究。

This dataset is named NIST log-viscosity (base + Redlich-Kister LOOCV fit) and features a two-layer structure for viscosity regression analysis of chemical mixtures. The base data layer (points) contains approximately 150,000 measurement rows, each representing a viscosity measurement point for binary mixtures. Columns include: mixture_T_id (join key), value (log viscosity), T (temperature in K), P (pressure in bar), cmp_ids (length-2 list of compound IDs), and cmp_mole_fractions (length-2 list of mole fractions aligned with cmp_ids). The dataset also includes a compounds.csv file for mapping compound IDs to SMILES representations, and a fits/rk_loocv.csv file storing fit results based on the Redlich-Kister model with leave-one-out cross-validation (LOOCV), including fit parameters and canonical columns (e.g., pure_at_x0, pure_at_x1, nonlinearity). Derived from the NIST log-viscosity benchmark and processed via chemixhub, it filters out bins with measurements from multiple references to ensure consistency. It is suitable for cheminformatics, machine learning regression tasks, and mixture property prediction research.