diff --git a/sparse_vector/src/main.rs b/sparse_vector/src/main.rs index 948087c..c793fa0 100644 --- a/sparse_vector/src/main.rs +++ b/sparse_vector/src/main.rs @@ -1,3 +1,4 @@ +use std::collections::HashMap; use std::ops::{Add, Mul, Sub}; use std::thread; use std::time::Instant; @@ -12,8 +13,7 @@ static ALLOC: jemallocator::Jemalloc = jemallocator::Jemalloc; /// Only stores more efficiently when at least 50% of all elements are zeros pub struct SparseVec { - values: Vec, - indices: Vec, + data: HashMap } impl SparseVec { @@ -21,9 +21,8 @@ impl SparseVec { pub fn dot(&self, other: &SparseVec) -> f64 { let mut sum = 0.0; - for index in 0..other.indices.len() { - // exponential search for an element in the second vector to have the same index - sum += binary_search(self.indices[index], &other.indices, &other.values) * self.values[index]; + for (k, v) in self.data.iter() { + sum += v * other.data.get(k).unwrap_or(&0.0); } sum @@ -32,49 +31,21 @@ impl SparseVec { pub fn new(elements: usize, non_null: f64) -> Self { let non_zero_elements = (elements as f64 * non_null) as usize; - let mut values = Vec::with_capacity(non_zero_elements); - let mut indices = Vec::with_capacity(non_zero_elements); + let mut data = HashMap::with_capacity(non_zero_elements); let mut rng = rand::thread_rng(); for i in 0..non_zero_elements { - values.push(0.5); - let idx = i as f32 / non_zero_elements as f32 * (elements as f32 - 4.0) + rng.gen_range(0.0..3.0); - indices.push(idx as usize); + data.insert(idx as usize, 0.5); } Self { - values, - indices + data } } } -#[inline] -fn binary_search(target: usize, indices: &[usize], values: &[f64]) -> f64 { - let mut range = 0..indices.len(); - loop { - let mut median = (range.end - range.start) >> 1; - if median == 0 { - break; - } - median += range.start; - - if indices[median] == target { - return values[median]; - } - - if indices[median] > target { - range.end = median; - } else { - range.start = median; - } - } - - 0.0 -} - macro_rules! time { ($name:literal, $block:expr) => {{ let start = Instant::now();