GitEvolve

The repository for code in GitEvolve: Predicting the Evolution of GitHub Repositories. If you find this repository useful in your research, please consider citing:

@misc{zhou2020gitevolve,
      title={GitEvolve: Predicting the Evolution of GitHub Repositories}, 
      author={Honglu Zhou and Hareesh Ravi and Carlos M. Muniz and Vahid Azizi and Linda Ness and Gerard de Melo and Mubbasir Kapadia},
      year={2020},
      eprint={2010.04366},
      archivePrefix={arXiv},
      primaryClass={cs.SI}
}

Installation

Run 'conda env create -f environment.yml' to create a conda environment that satisfies the package requirement. Find (or modify) the conda environment name in the first line of 'environment.yml'.

How to run

models

We provide the following model variants:

multitask_lstm_user_cluster (proposed model that outputs event type, time and user cluster of the next time step)
multitask_lstm_user_cluster_true_event_only
multitask_lstm_branch_et
multitask_lstm_branch_td
multitask_lstm_branch_uc
multitask_random_constraints
multitask_repeat_last_action
multitask_repeat_no_action
multitask_lstm (this model variant outputs only event type and time.)

Please use the main script of the model variant that you selected.

train

python main_multitask_lstm_user_cluster.py train --create_dataset=1  --exp_name=myexp

test (and evaluation)

python main_multitask_lstm_user_cluster.py test --create_dataset=1 --load_model_epoch=40  --exp_name=myexp --given_gt=0

python main_multitask_lstm_user_cluster.py test --create_dataset=1 --load_model_epoch=40  --exp_name=myexp --given_gt=1

evaluation (only)

python main_multitask_lstm_user_cluster.py eval --load_model_epoch=40  --exp_name=myexp --given_gt=0

Please check the create_config.py inside the folder of each model variant, to change the data paths, hyper-parameters, and other settings.