Multi-Location Software Model Completion

Paper

Additional information regarding the paper is given in the PDF file: Multi-Location_Software Model_Completion.pdf of the repository. More precisely, the paper has an appendix there; for more info, just scroll down to the appendix at the very end of the paper.

Code

Code to run is given in experiments/crosscutting_changes.

Preprocessing

To compute the embeddings, run the script preprocessing_add_node_embeddings.py,
which will embed the nodes of the graphs. Please provide your OpenAI key there.

For Experiment 5, the cross-project setting, run crossproject.py which divides the data in 10 folds.

Training

Run neural_network_analyser.py for training the neural network.

Run neural_network_hyperparametertuning.py for hyperparameter tuning.

Run historical_only_analyser.py, which builds the adjacency matrix for the historical baseline from the train set.

Testing

All scripts start with test_eval.
The testing scripts for the baseline semantics, historical, random, and neural network approach "NextFocus" are named test_eval_{approach}. For the baselines these files are located in the folder baselines The testing scripts for the baseline semantics, historical, random, and neural network approach "NextFocus" are named test_eval_{approach}. For the baselines these files are located in the folder baselines`.

These will be passed through the network.

Multi-location model completion

We integrated NextFocus with existing approach by Tinnes et al. and slighlty improved their approach for multi-location model completion. The corresponding files: call round_trip.py for the overall iterative procedure.

Final evaluation of the RQs and Experiments

Start with final_:

• final_comparison.py for statistics on Experiment 1 and Experiment 3

• final_eval_graph_radius.py for Experiment 2

• final_eval_dataset_size.py for Experiment 3

• Experiment 5 is done with the same scripts as experiment 1, training the neural network via neural_network_analyser.py in sum this is done 10 times, with different folds. These 10 networks are afterwards evaluated via test_eval_neuralnetwork and statistics are calculated via final_comparison.py

• for Experiment 6, the final statistics are computed by final_t-times_completion.py

Additional info

Files starting with Helper are called by the rest.
Modules follow the same pattern.

Data

The data and intermediate results of the approach are for simplicity provided in the data folder.