Struct GraphDataLoader 

pub struct GraphDataLoader { /* private fields */ }

Graph data loader

Implementations

impl GraphDataLoader

pub fn new(batch_size: usize, shuffle: bool) -> Result<Self, Box<dyn Error>>

Create a new graph data loader with validation

Arguments

• batch_size - Size of batches, must be > 0
• shuffle - Whether to shuffle data

Returns

• Ok(Self) - Successfully created data loader
• Err - If batch_size is 0

Example

use torsh_graph::data::GraphDataLoader;
let loader = GraphDataLoader::new(32, true).unwrap();
assert_eq!(loader.batch_size(), 32);

pub fn with_node_features(self, num_node_features: usize) -> Self

Set the number of synthetic node features generated for formats that carry connectivity only (edge lists).

Defaults to DEFAULT_NODE_FEATURES.

pub fn num_node_features(&self) -> usize

Number of synthetic node features used for feature-less formats

pub fn batch_size(&self) -> usize

Get batch size

pub fn shuffle(&self) -> bool

Get shuffle setting

pub fn from_directory(&self, path: &str) -> Result<Vec<GraphData>, TorshError>

Load every supported graph file in path into a GraphData list.

Files are dispatched on their extension to the parsers in crate::datasets:

Extension	Loader
edges, edgelist, txt	[EdgeListLoader]
gml	[GMLLoader]
json	[JSONLoader]

Files with any other extension (and sub-directories) are ignored. Entries are visited in sorted order so the returned dataset is deterministic.

Errors

Returns an error if the directory cannot be read, or if a file with a supported extension fails to parse. A failing file is reported rather than silently skipped, so a mis-typed path can never masquerade as an empty dataset.

Example

use torsh_graph::data::GraphDataLoader;
let loader = GraphDataLoader::new(4, false).unwrap();
let graphs = loader.from_directory("./my_graphs").unwrap();
println!("loaded {} graphs", graphs.len());