Dataset for Towards Understanding Performance Bugs in Popular Data Science Libraries

In our paper, we conducted a large-scale empirical study to characterize performance bugs in seven popular data science libraries. We identified 202 performance bugs. By analyzing these bugs (including bug description, patches, and project development history), we analyzed their impacts, proposed a taxonomy of the root causes and summarized three challenges for locating root causes and four challenges for fixing these performance bugs. We found there are about 20% of fixes has LOC not larger than 10, which indicates they can be fixed through simple changes. We then manually checked the patch and found several fixing strategies with small LOC that can be automated. We believe that this study can facilitate future research and the development of data science ecosystems. Both data science libraries' developers and users can receive useful guidance from our study. This dataset contains 202 performance bugs in data science core libraries, and their impacts, root causes, location and fixing challenge, and fixing strategy. Our replication package consists of three main folders:RQ1&2_Impacts_and_Root_Causes, RQ3_Root_Causes_Locating_Fixing_Effort_Challenge and RQ4_Fixing_Strategy. RQ1&2_Impacts_and_Root_Causes In this folder we first placed the identified impact (Explicit and Implicit). Then we gave the identified symptoms and root cause taxonomy. In each file (corresponding to each iteration), we provided the repo name, issue number, and the label (symptom and root cause). RQ3_Root_Causes_Locating_Fixing_Effort_Challenge The challenge in locating and fixing these bugs in data science libraries are identified here. RQ4_Fixing_Strategy We provided the identified fixing strategy with small LOC. In the file, we provided the repo name, issue number, and the label (fixing strategy).