manasviaggarwal / hybridpool

HybridPool

A Deep Hybrid Pooling Architecture for Graph Classification with Hierarchical Attention

This is a tensorflow based implementation of Hybrid Pooling as discussed in the paper.

Dataset:

1. The dataset_graph folder contains all the datasets which we used in experiments of graph classification.

How to run:

1. For Graph Classification: (Default dataset is set to MUTAG) python graph_classification.py

Requirements:

1. python (version 3.6 or above)
2. tensorflow (version 1.14)
3. networkx
4. keras
5. numpy
6. pickle
7. scipy
8. pandas
9. collections

Parameters:

1. For Graph Classification: 1.1 dataset: The name of the dataset 1.2 epoch: Number of epochs to train the model 1.3 learning_rate: Learning rate 1.4 embd_dim: Final Embedding dimension 1.5 gcn_layer: Number of GCN layers 1.6 gcn_dim: GCN Embedding dimension 1.7 dropout: Dropout rate 1.8 batch_size: Batch size

We can specify these parameters while running python file. For eg: To specify any other dataset, run following command: python graph_classification.py --dataset NCI1