3 unstable releases

0.2.2 Feb 24, 2022
0.1.1 Feb 23, 2022
0.1.0 Feb 22, 2022

#784 in Machine learning

23 downloads per month

MIT license

130KB
2.5K SLoC

gcn_agg

Rust

a gcn simulator write in rust with ramulator inside


lib.rs:

the crate gcn_agg is a graph convolutional neural network accelerator simulator. there are 4 parts in the crate:

  • accelerator: the accelerator is a graph convolutional neural network accelerator.
  • graph: the data structure to represent the graph.
  • node_features: the data structure to represent the node features.
  • statics: the result statics to record the result.

Examples

    use chrono::Local;
    use gcn_agg::{
    accelerator::System, gcn_result::GcnAggResult, graph::Graph, node_features::NodeFeatures,
    settings::Settings,
    };
    use itertools::Itertools;

    fn test_system() -> Result<(), Box<dyn std::error::Error>> {
        std::fs::create_dir_all("output")?;

        simple_logger::init_with_level(log::Level::Info).unwrap_or(());
        let current_time: String = Local::now().format("%Y-%m-%d-%H-%M-%S%.6f").to_string();

        let start_time = std::time::Instant::now();
        let mut results = GcnAggResult::default();

        let settings = Settings::new(vec!["configs/default.toml".into()]).unwrap();
        results.settings = Some(settings.clone());
        // create the folder for output
        std::fs::create_dir_all("output")?;

        let graph_name = &settings.graph_path;
        let features_name = &settings.features_paths;

        let graph = Graph::new(graph_name.as_str())?;

        let node_features: Vec<_> = features_name
            .iter()
            .map(|x| NodeFeatures::new(x.as_str()))
            .try_collect()?;

        let mem_stat_path = format!("output/{}_mem_stat.txt", current_time);
        let mut system = System::new(
            &graph,
            &node_features,
            settings.accelerator_settings,
            &mem_stat_path,
        );

        // run the system
        let mut stat = system.run()?;

        // record the simulation time
        let simulation_time = start_time.elapsed().as_secs();
        // record the result
        let seconds = simulation_time % 60;
        let minutes = (simulation_time / 60) % 60;
        let hours = (simulation_time / 60) / 60;
        let time_str = format!("{}:{}:{}", hours, minutes, seconds);
        stat.simulation_time = time_str;

        results.stats = Some(stat);
        let output_path = format!("output/{}.json", current_time);

        println!("{}", serde_json::to_string_pretty(&results)?);
        // write json of results to output_path
        std::fs::write(output_path, serde_json::to_string_pretty(&results)?)?;
        Ok(())
    }
    match test_system() {
        Ok(_) => println!("test_system success"),
        Err(e) => println!("test_system failed: {}", e),
   }

Dependencies

~5.5–7.5MB
~137K SLoC