DECODER VARIABLE MISUSE TOOL

This project contains the last release of the code of the variable misuse tool deployment for the DECODER project. This tools allows to detect and repair bugs or errors corresponding to variables within source code files written in Java, C and C++.

More info about the project and its tools here.

First steps

The basic environment is created with the following command that only needs to be executed the first time the environment is setup:

conda env create -f environment.yml

Once the environment is created, it is preserved in between environment restarts, it can be activated as:

conda activate tensorflow-gpu

Train and evaluate models

Notebooks foulder contains the training and the evaluation of the models for the Java and the C/C++ use cases. Training phase is carried out by the hypeopt library, in order to obtain the best model possible for both programming languages.

The result models are stored in the models foulder.

Git LFS data

Be careful that the LFS files tracked are downloaded when you run the notebooks. If they are not downloaded, you can download all with:

git lfs pull

from the root directory of this repository.