Generative Modeling of Labeled Graphs under Data Scarcity

Installation

The code has been tested over PyTorch 1.8.0 version with Python 3.7.0.

Change cuda version as per your GPU hardware support.

pip install -r requirements.txt

Boost (Version >= 1.70.0) and OpenMP are required for compling C++ binaries. Run build.sh script in the project's root directory.

./build.sh

Scripts to Run GShot:

• python meta_main_n.py leukemia : Training GShot Meta-Model
• python tune_main_n.py leukemia : Fine-Tuning GShot for Leukemia
• python evaluate_n.py leukemia : Evaluation on Leukemia

Here leukemia is the name of the config file present in configs/ folder. For AIDS-ca:aids_ca, Enzyme: enzyme , Spring: spring.

Code description

1. meta_main_n.py is the main script file of GShot, and specific arguments are set in configs/<Config_Filename>
2. tune_main_n.py is the fine-tuning script file of GShot and specific arguments are set in configs/<Config_Filename>
3. train.py model specific training files.
4. datasets/preprocess.py and util.py contain dataset processing functions.
5. datasets/process_dataset.py reads graphs from different data formats.

DFS code generation:

• dfscode/dfs_code.cpp calculates the minimum DFS code.
• dfscode/dfs_wrapper.py is a python wrapper for the cpp file.

Parameter setting:

• Configuration: Dataset specific parameters are present in configs/<Config_Filename>
• All the input arguments and hyper parameters setting are included in base_args.py.
• Dataset specific arguments which override the default arguments of base_args.py are present in configs folder.
• See the documentation in base_args.py for more detailed descriptions.

Outputs

base_args.py contains parameters to specify where the default directory etc. should be

• tensorboard/ contains tensorboard event objects which can be used to view training and validation graphs in real time.
• model_save/ stores the model checkpoints
• tmp/ stores all the temporary files generated during training and evaluation.

Evaluation

• evaluate_n.py: File to compute metrics between generated graphs and graphs from original dataset.

• We use GraphRNN implementation for structural metrics.

• NSPDK is evaluated using EDeN python package.

Baseline models:

GraphGen

• python main_n.py aids_ca_single Training GraphGen on AIDS-CA dataset
• python evaluate_n.py aids_ca_single Evaluating GraphGen on AIDS-CA dataset

GraphRNN

• python main_n.py aids_ca_single-grnn Training GraphRNN on AIDS-CA dataset
• python evaluate_n.py aids_ca_single-grnn Evaluating GraphRNN on AIDS-CA dataset

PreTrain GraphGen

• python multi_main_n.py aids_ca_multi Training PreTrain GraphGen
• python tune_main_n.py aids_ca_multi Fine-Tuning PreTrained GraphGen on AIDS-CA dataset
• python evaluate_n.py aids_ca_multi Evaluating PreTrain GraphGen on AIDS-CA dataset