Usage

Please refer to Requirements.md and Installation.md.

To detect version incompatibility issues in a given deep learning project, one can create a folder named by the project's name in DECIDE/benchmark and follows the format of files in DECIDE/benchmark/examples to specify the project-required libraries and the local stack information. Then run:

$ python main.py -n PROJ_NAME

Parameters:

• -n: the name of detect project.
• -m: the mode of collecting information in local stack. Use 'file' to collect from DECIDE/benchmark/PROJ_NAME/local_info.json, use 'local' to directly gather the version information by scanning the local machine.
• -l (Optional): the specified file path of local stack information, default path is DECIDE/benchmark/PROJ_NAME/local_info.json .
• -r (Optional): the specified file path of version information of libraries required by the project, default path is DECIDE/benchmark/PROJ_NAME/requirements.txt.
• -p (Optional): The root path of the target project.

Replay Experiments

To replay the evaluation in our benchmark, please run:

$ chmod +x experiment.sh
$ ./experiment.sh

The results will be generated in DECIDE/detection_result.