dataset_graph_mnist: MNIST Graph Dataset

In rdinnager/rspektral: What the Package Does (One Line, Title Case)

Description

Loads the MNIST dataset and a K-NN graph to perform graph signal classification, as described by Defferrard et al. (2016). The K-NN graph is statically determined from a regular grid of pixels using the 2d coordinates. The node features of each graph are the MNIST digits vectorized and rescaled to 0, 1. Two nodes are connected if they are neighbours according to the K-NN graph. Labels are the MNIST class associated to each sample.

Usage

dataset_graph_mnist(k = 8L, noise_level = 0)

Arguments

k	int, number of neighbours for each node
noise_level	fraction of edges to flip (from 0 to 1 and vice versa)
return_type	Data format to return data in. One of either "list", or "tidygraph"

Value

• X_train, y_train: training node features and labels; - X_val, y_val: validation node features and labels; - X_test, y_test: test node features and labels; - A: adjacency matrix of the grid;

rdinnager/rspektral documentation built on June 12, 2021, 1:26 a.m.