README

Train model

Train LSTM model for OSS sustainability prediction.

Create a conda environment and install required dependencies

conda create -n LSTM python=3.8.5
conda activate LSTM
pip install -r train_requirement.txt

Specify experiment configuration in LSTM_config.yaml before running.
Train and evaluate LSTM model on dataset at ../mvts_dataset by running:
```
python LSTM.py LSTM_config.yaml
```
The program will automatically train repeat_times*(start_timestep-start_timestep+1) number of LSTM-based timeseries classification models. And the models and corresponding evaluation results will be generated under output_path. (all these parameters are specified in step 3)

Explain the prediction behavior of a specified model on specified data samples using SHAP algorithm.

Go to shap_explain directory
Create a new conda environment and install required dependencies.(note that it should be a separated environment from the model-training environment, due to some SHAP’s compatibility issues)
```
conda create -n SHAP python=3.8.5
conda activate SHAP
pip install -r explain_requirement.txt
```
Specify experiment configuration in shap4LSTM_config.yaml before running.
Explain LSTM model’s behavior by running:
```
python shap4LSTM.py shap4LSTM_config.yaml
```
The program will automatically use the model at model_path to inference. And SHAP algorithm will be used to observe the inference results and assign weights to each feature at each timestep of each project. Corresponding explanatory and analysis results (specified by mode) will be generated under output_path(all these parameters are specified in step 3)