Non-functional bugs (e.g., performance-or accuracy-related bugs) in Deep Learning (DL) frameworks can lead to some of the most devastating consequences. Reporting those bugs on a repository such as GitHub is a standard route to fix them. Yet, given the growing number of new GitHub reports for DL frameworks, it is intrinsically difficult for developers to distinguish those that reveal non-functional bugs among the others, and assign them to the right contributor for investigation in a timely manner. In this paper, we propose MHNurf -an end-to-end tool for automatically identifying non-functional bug related reports in DL frameworks. The core of MHNurf is a Multifaceted Hierarchical Attention Network (MHAN) that tackles three unaddressed challenges: (1) learning the semantic knowledge, but doing so by (2) considering the hierarchy (e.g., words/tokens in sentences/statements) and focusing on the important parts (i.e., words, tokens, sentences, and statements) of a GitHub report, while (3) independently extracting information from different types of features, i.e., content, comment, code, command, and label.To evaluate MHNurf, we leverage 3,721 GitHub reports from five DL frameworks for conducting experiments. The results show that MHNurf works the best with a combination of content, comment, and code, which considerably outperforms the classic HAN where only the content is used. MHNurf also produces significantly more accurate results than nine other state-of-the-art classifiers with strong statistical significance, i.e., up to 71% AUC improvement and has the best Scott-Knott rank on four frameworks while 2nd on the remaining one. To facilitate reproduction and promote future research, we have made our dataset, code, and detailed supplementary results publicly available at: https://github.com/ideas-labo/APSEC2022-MHNurf.

The tremendous success of Deep Learning (DL) has significantly boosted the number of open-sourced DL frameworks hosted on GitHub. Among others, performance and accuracy bugs are critical factors that affect the reputation of these DL frameworks, therefore understanding the practice of discovering and investigating them for DL is important. In this paper, we conduct an exploratory study on the nature of reporting performance and accuracy bugs for DL frameworks, aiming to improve our knowledge on this topic. Our study covers 10 most popular open-sourced DL frameworks on GitHub (e.g., TensorFlow, Keras, and PyTorch), based on which we sample 664 representative performance and accuracy bug reports out of a total population of 22,522. Through systematic analysis, we found that: (1) low speed is the primary reason that a performance bug related report is submitted but we see no consistent pattern for accuracy related ones; (2) most of the reports are about issues encountered in the training stage; (3) only a small proportion of the reports provide insufficient information to investigate; (4) the majority of the performance and accuracy bug reports (from 69% to 100%) are not related to the actual bug or regarded as unclassified; (5) around 50% of the performance and accuracy bug reports, which indeed reveal bugs, are not resolved by direct patches. Deriving from the above, we discuss a set of actionable implications to the researchers, maintainers, and report submitters. To promote open science, the labeled dataset has been made publicly available at https://zenodo.org/record/6371676 .