An Empirical Study of Logging Practice in CUDA-based Deep Learning Systems

This project is a replication package for the paper titled "An Empirical Study of Logging Practice in CUDA-based Deep Learning Systems" . The paper proposes an empirical study to examine the characteristics and rationales of logging practices in CUDA-based deep learning (DL) systems.

Purpose

In this repository, we propose a pipeline that extracts both general and DL-specific logs from selected projects.

Folders Overview

dataset

Contains method level python code files for each selected subjects.

DL-logging

Includes Python scripts for:

• Downloading projects that imported selected libraries by using GitHub Search API
• Extracting logging statements