Work on this project is ongoing and is described in detail in the original paper.

If you use this work, please cite the following paper:

Meltzer, P., Mallea, M. D. G., & Bentley, P. J. (2019). PiNet: Attention Pooling for Graph Classification, NeurIPS 2019 Graph Representation Learning Workshop

If this work is of interest to you, or you use the code, I'd love your feedback. Please email me at [email protected] with any comments, criticisms or suggestions! :D

A Pytorch implementation of Pinet can be found here.

• Example Usage
  • Building the model
  • Data Format
  • Evaluation
  • Prediction Only
• GCN Layer
• Data Generator
• Experiments
  • Experiment 1: Message Passing Mechanisms
  • Experiment 2: Isomorphism
  • Experiment 3: Benchmark Against Existing Methods

• out_dim_a2: output dimension for attention
• out_dim_x2: output dimension for features
• learn_pqr: learn message passing parametrisation during training (default is True)
• preprocess_A: List of options as Strings (for manual pre-processing - should not be used with learn_pqr=True)
  • 'add_self_loops'
  • 'sym_normalise_A'
  • 'laplacian'
  • 'sym_norm_laplacian'
  • could include multiple, i.e.: ['add_self_loops', 'sym_normalise_A']
• tensor_board_logging: enable logging for TensorBoard
• reduce_lr_callback: reduce learning rate based on validation set

from model.PiNet import PiNet
from analysis.experiment2 import generate
from sklearn.model_selection import StratifiedKFold

num_classes = 3
batch_size = 5

pinet = PiNet()

folds = list(StratifiedKFold(n_splits=10, shuffle=True).split(X, Y))

• A: List of adjacency matrices as ndarrays
• X: List of features matrices as ndarrays
• Y: (n x 1) ndarray containing class no.

• num_classes: number of classes
• epochs: default 200
• batch_size: default 50
• folds: Folds or splits of train/test ids
• dataset_name: default is 'dataset_name'
• verbose: default is 1

accs, times = pinet.fit_eval(A, X, Y, num_classes=num_classes,
                                  epochs=50, batch_size=batch_size, folds=folds, verbose=0)

preds = pinet.get_predictions(A, X, Y, batch_size=batch_size)

• GCN: model/MyGCN.py
• MyGCN layer takes a list of A_ and X^{l} as input, and gives a single output of X^{l+1}

• analysis/experiment2/generate.py

params:

num_nodes_per_graph = 50
num_graph_classes = 5
num_node_classes = 2
num_graphs_per_class = 100
batch_size = 5
examples_per_classes = [2, 4, 6, 8, 10]

• train set selected by stratified sample
• repeated 10x per examples_per_classes

Observe effect of various matrices for message passing/diffusion.

Compare against existing methods on benchmark data.

Mean classification accuracies for each classifier. For manual search the values p and q as follows: MUTAG and PROTEINS p = 1, q = 0, NCI-1 and NCI-109 p = q = 1, PTC p = q = 0. * indicates PiNet (both models) achieved statistically significant gain.