keras-gcn 0.15.0

Graph convolutional layers

Keras Graph Convolutional Network

Graph convolutional layers.

Install

pip install keras-gcn

Usage

GraphConv

from tensorflow import keras
from keras_gcn import GraphConv


DATA_DIM = 3

data_layer = keras.layers.Input(shape=(None, DATA_DIM))
edge_layer = keras.layers.Input(shape=(None, None))
conv_layer = GraphConv(
    units=32,
    step_num=1,
)([data_layer, edge_layer])

step_num is the maximum distance of two nodes that could be considered as neighbors. If step_num is greater than 1, then the inputs of edges must be 0-1 matrices.

GraphMaxPool & GraphAveragePool

Pooling layers with the step_num argument.