use fnntw::Tree; use ndarray_npy::write_npy; use rayon::prelude::*; use std::error::Error; // const DIMS: usize = 2; // const NDATA: usize = 10; // const NQUERY: usize = 1000; const DIMS: usize = 3; const NDATA: usize = 100_000; const NQUERY: usize = 1_000_000; const DATA_FILE: &'static str = "data.npy"; const QUERY_FILE: &'static str = "query.npy"; const RESULT_FILE: &'static str = "results.npy"; const INDICES_FILE: &'static str = "indices.npy"; type T = f64; fn main() -> Result<(), Box> { // Gather data and query points let data = get_data(); let queries = get_queries(); save_data_queries(&data, &queries)?; // Build tree let leafsize = 4; let tree = Tree::new_parallel(&data, leafsize, 3).unwrap(); println!("Built tree"); // Query tree, in parallel let (sqdists, indices): (Vec, Vec) = queries .par_iter() // .iter() .map_with(&tree, |t, q| { let result = t.query_nearest(q).unwrap(); (result.0, result.1) }) // .map(|q| tree.query_nearest(q)) .unzip(); println!("Queried"); save_results(sqdists, indices) } fn get_data() -> Vec<[T; DIMS]> { gen_points::() } fn get_queries() -> Vec<[T; DIMS]> { gen_points::() } fn gen_points() -> Vec<[T; DIMS]> { (0..N).map(|_| [(); DIMS].map(|_| rand::random())).collect() } fn save_results(sqdists: Vec, indices: Vec) -> Result<(), Box> { write_npy(RESULT_FILE, &ndarray::Array1::from_vec(sqdists))?; write_npy(INDICES_FILE, &ndarray::Array1::from_vec(indices))?; Ok(()) } fn save_data_queries( data: &Vec<[T; DIMS]>, queries: &Vec<[T; DIMS]>, ) -> Result<(), Box> { let flat_data: Vec = data.iter().cloned().flatten().collect(); let flat_query: Vec = queries.iter().cloned().flatten().collect(); write_npy( DATA_FILE, &ndarray::Array2::from_shape_vec((NDATA, DIMS), flat_data)?, )?; write_npy( QUERY_FILE, &ndarray::Array2::from_shape_vec((NQUERY, DIMS), flat_query)?, )?; Ok(()) }