bermaneh/pde-var-logprob-evolution-canary-v1
收藏资源简介:
变量对数概率演化数据集:记录了在NoComm_CorrVar和NoComm_Valid PDE代码片段中每个变量出现时,真实变量名(gt_var)和错误变量名(corrupt_var)的条件对数概率(logP)及其差异。数据集包含442个样本,每个样本有20个特征,详细记录了代码片段信息、变量信息、位置信息以及模型对变量名的概率评估结果。该数据集旨在研究代码理解中变量名称的概率演化特性。
Variable log-probability evolution: logP(gt_var | prefix) vs logP(corrupt_var | prefix) at each variable occurrence in NoComm_CorrVar and NoComm_Valid PDE code snippets. The dataset contains 442 rows with 20 columns, detailing code snippet identifiers, condition types, PDE classes, variable information, occurrence indices, character offsets, line numbers, normalized code positions, and the models log-probability assessments of variable names. The dataset is designed to study the probability evolution characteristics of variable names in code understanding.




