We implement a rule-based static checker, named DeepPerf, to detect PPs in DL systems. DeepPerf is implemented with two static analysis tools, AST and Jedi. It currently supports three types of PPs whose detection rules are manually derived from our empirical study.
- Python >= 3.8.0
- jedi >= 0.18.0
Run python graph_detect.py [project_path] to check all python files in [project_path].
You could specify [project_path] as ./test_code/graph to check the default examples.
We have implemented Checker 2 and Checker 3 at the same time. Run python data_detect.py [project_path] to check all python files in [project_path].
You could specify [project_path] as ./test_code/data to check the default examples.