-
Notifications
You must be signed in to change notification settings - Fork 10
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Groundwork for some new non-linear algorithms (#196)
* Moving some shared characteristics of non linear algos into a trait * Using par_iter for find_similar_n * Fix clippy lints
- Loading branch information
Showing
7 changed files
with
186 additions
and
297 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -1,181 +1,18 @@ | ||
| #![allow(dead_code)] | ||
| use super::LinearAlgorithm; | ||
| use super::SimilarityVector; | ||
| use ahnlich_types::keyval::StoreKey; | ||
| use std::cmp::Reverse; | ||
| use std::collections::BinaryHeap; | ||
| use std::num::NonZeroUsize; | ||
|
|
||
| pub(crate) struct MinHeap<'a> { | ||
| max_capacity: NonZeroUsize, | ||
| heap: BinaryHeap<Reverse<SimilarityVector<'a>>>, | ||
| pub enum HeapOrder { | ||
| Min, | ||
| Max, | ||
| } | ||
|
|
||
| impl<'a> MinHeap<'a> { | ||
| pub(crate) fn new(capacity: NonZeroUsize) -> Self { | ||
| Self { | ||
| heap: BinaryHeap::new(), | ||
| max_capacity: capacity, | ||
| } | ||
| } | ||
| #[tracing::instrument(skip_all)] | ||
| pub(crate) fn len(&self) -> usize { | ||
| self.heap.len() | ||
| } | ||
| #[tracing::instrument(skip_all)] | ||
| pub(crate) fn push(&mut self, item: SimilarityVector<'a>) { | ||
| self.heap.push(Reverse(item)); | ||
| } | ||
| #[tracing::instrument(skip_all)] | ||
| pub(crate) fn pop(&mut self) -> Option<SimilarityVector<'a>> { | ||
| self.heap.pop().map(|popped_item| popped_item.0) | ||
| } | ||
|
|
||
| #[tracing::instrument(skip_all)] | ||
| pub(crate) fn output(&mut self) -> Vec<(StoreKey, f32)> { | ||
| let mut result: Vec<_> = Vec::with_capacity(self.max_capacity.get()); | ||
|
|
||
| loop { | ||
| match self.pop() { | ||
| Some(value) if result.len() < self.max_capacity.get() => { | ||
| let vector_sim = value.0; | ||
| result.push((vector_sim.0.clone(), vector_sim.1)); | ||
| } | ||
| _ => break, | ||
| } | ||
| } | ||
| result | ||
| } | ||
| } | ||
|
|
||
| pub(crate) struct MaxHeap<'a> { | ||
| max_capacity: NonZeroUsize, | ||
| heap: BinaryHeap<SimilarityVector<'a>>, | ||
| } | ||
|
|
||
| impl<'a> MaxHeap<'a> { | ||
| pub(crate) fn new(capacity: NonZeroUsize) -> Self { | ||
| Self { | ||
| heap: BinaryHeap::new(), | ||
| max_capacity: capacity, | ||
| } | ||
| } | ||
| #[tracing::instrument(skip_all)] | ||
| fn push(&mut self, item: SimilarityVector<'a>) { | ||
| self.heap.push(item); | ||
| } | ||
| #[tracing::instrument(skip_all)] | ||
| pub(crate) fn pop(&mut self) -> Option<SimilarityVector<'a>> { | ||
| self.heap.pop() | ||
| } | ||
| #[tracing::instrument(skip_all)] | ||
| pub(crate) fn len(&self) -> usize { | ||
| self.heap.len() | ||
| } | ||
|
|
||
| #[tracing::instrument(skip_all)] | ||
| fn output(&mut self) -> Vec<(StoreKey, f32)> { | ||
| let mut result: Vec<_> = Vec::with_capacity(self.max_capacity.get()); | ||
|
|
||
| loop { | ||
| match self.heap.pop() { | ||
| Some(value) if result.len() < self.max_capacity.get() => { | ||
| let vector_sim = value.0; | ||
| result.push((vector_sim.0.clone(), vector_sim.1)); | ||
| } | ||
| _ => break, | ||
| } | ||
| } | ||
| result | ||
| } | ||
| } | ||
|
|
||
| pub(crate) enum AlgorithmHeapType<'a> { | ||
| Min(MinHeap<'a>), | ||
| Max(MaxHeap<'a>), | ||
| } | ||
|
|
||
| impl<'a> AlgorithmHeapType<'a> { | ||
| #[tracing::instrument(skip_all)] | ||
| pub(crate) fn push(&mut self, item: SimilarityVector<'a>) { | ||
| match self { | ||
| Self::Max(h) => h.push(item), | ||
| Self::Min(h) => h.push(item), | ||
| } | ||
| } | ||
| #[tracing::instrument(skip_all)] | ||
| pub(crate) fn pop(&mut self) -> Option<SimilarityVector<'a>> { | ||
| match self { | ||
| Self::Max(h) => h.pop(), | ||
| Self::Min(h) => h.pop(), | ||
| } | ||
| } | ||
|
|
||
| #[tracing::instrument(skip_all)] | ||
| pub(crate) fn output(&mut self) -> Vec<(StoreKey, f32)> { | ||
| match self { | ||
| Self::Min(h) => h.output(), | ||
| Self::Max(h) => h.output(), | ||
| } | ||
| } | ||
| } | ||
|
|
||
| impl From<(&LinearAlgorithm, NonZeroUsize)> for AlgorithmHeapType<'_> { | ||
| fn from((value, capacity): (&LinearAlgorithm, NonZeroUsize)) -> Self { | ||
| impl From<&LinearAlgorithm> for HeapOrder { | ||
| fn from(value: &LinearAlgorithm) -> Self { | ||
| match value { | ||
| LinearAlgorithm::EuclideanDistance => AlgorithmHeapType::Min(MinHeap::new(capacity)), | ||
| LinearAlgorithm::EuclideanDistance => HeapOrder::Min, | ||
| LinearAlgorithm::CosineSimilarity | LinearAlgorithm::DotProductSimilarity => { | ||
| AlgorithmHeapType::Max(MaxHeap::new(capacity)) | ||
| HeapOrder::Max | ||
| } | ||
| } | ||
| } | ||
| } | ||
|
|
||
| #[cfg(test)] | ||
| mod tests { | ||
|
|
||
| use super::*; | ||
|
|
||
| #[test] | ||
| fn test_min_heap_ordering_works() { | ||
| let mut heap = MinHeap::new(NonZeroUsize::new(3).unwrap()); | ||
| let mut count = 0.0; | ||
| let first_vector = StoreKey(vec![2.0, 2.0]); | ||
|
|
||
| // If we pop these scores now, they should come back in the reverse order. | ||
| while count < 5.0 { | ||
| let similarity: f32 = 1.0 + count; | ||
|
|
||
| let item: SimilarityVector = (&first_vector, similarity).into(); | ||
|
|
||
| heap.push(item); | ||
|
|
||
| count += 1.0; | ||
| } | ||
|
|
||
| assert_eq!(heap.pop(), Some((&first_vector, 1.0).into())); | ||
| assert_eq!(heap.pop(), Some((&first_vector, 2.0).into())); | ||
| assert_eq!(heap.pop(), Some((&first_vector, 3.0).into())); | ||
| } | ||
|
|
||
| #[test] | ||
| fn test_max_heap_ordering_works() { | ||
| let mut heap = MaxHeap::new(NonZeroUsize::new(3).unwrap()); | ||
| let mut count = 0.0; | ||
| let first_vector = StoreKey(vec![2.0, 2.0]); | ||
|
|
||
| // If we pop these scores now, they should come back the right order(max first). | ||
| while count < 5.0 { | ||
| let similarity: f32 = 1.0 + count; | ||
| let item: SimilarityVector = (&first_vector, similarity).into(); | ||
|
|
||
| heap.push(item); | ||
|
|
||
| count += 1.0; | ||
| } | ||
|
|
||
| assert_eq!(heap.pop(), Some((&first_vector, 5.0).into())); | ||
| assert_eq!(heap.pop(), Some((&first_vector, 4.0).into())); | ||
| assert_eq!(heap.pop(), Some((&first_vector, 3.0).into())); | ||
| } | ||
| } |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.