Use NoHashHasher for Candidate HashMaps to improve performance

This commit is contained in:
RunasSudo 2022-08-20 23:20:42 +10:00
parent 61b22b388d
commit 422a198cf5
Signed by: RunasSudo
GPG Key ID: 7234E476BF21C61A
9 changed files with 71 additions and 28 deletions

7
Cargo.lock generated
View File

@ -651,6 +651,12 @@ dependencies = [
"rawpointer",
]
[[package]]
name = "nohash-hasher"
version = "0.2.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "2bf50223579dc7cdcfb3bfcacf7069ff68243f8c363f62ffa99cf000a6b9c451"
[[package]]
name = "normalize-line-endings"
version = "0.3.0"
@ -738,6 +744,7 @@ dependencies = [
"itertools",
"js-sys",
"ndarray",
"nohash-hasher",
"num-bigint",
"num-rational",
"num-traits",

View File

@ -17,6 +17,7 @@ csv = "1.1.6"
derive_builder = "0.10.2"
derive_more = "0.99.14"
git-version = "0.3.4"
nohash-hasher = "0.2.0"
ibig = "0.3.2"
itertools = "0.10.1"
ndarray = "0.15.3"

View File

@ -22,6 +22,7 @@ use crate::ties::{self, TieStrategy};
use itertools::Itertools;
use ndarray::{Array, Dimension, IxDyn};
use nohash_hasher::BuildNoHashHasher;
#[cfg(not(target_arch = "wasm32"))]
use rkyv::{Archive, Deserialize, Serialize};
@ -130,7 +131,7 @@ impl Constraints {
}
/// Check if any elected candidates exceed constrained maximums
pub fn exceeds_maximum<'a, N: Number>(&self, election: &Election<N>, candidates: HashMap<&'a Candidate, CountCard<'a, N>>) -> Option<(&Constraint, &ConstrainedGroup)> {
pub fn exceeds_maximum<'a, N: Number>(&self, election: &Election<N>, candidates: HashMap<&'a Candidate, CountCard<'a, N>, BuildNoHashHasher<Candidate>>) -> Option<(&Constraint, &ConstrainedGroup)> {
for constraint in &self.0 {
for group in &constraint.groups {
let mut num_elected = 0;
@ -360,7 +361,7 @@ impl ConstraintMatrix {
}
/// Update cands/elected in innermost cells based on the provided [CountState::candidates](crate::election::CountState::candidates)
pub fn update_from_state<N: Number>(&mut self, election: &Election<N>, candidates: &HashMap<&Candidate, CountCard<N>>) {
pub fn update_from_state<N: Number>(&mut self, election: &Election<N>, candidates: &HashMap<&Candidate, CountCard<N>, BuildNoHashHasher<Candidate>>) {
let constraints = election.constraints.as_ref().unwrap();
// Reset innermost cells
@ -626,7 +627,7 @@ impl ops::IndexMut<&[usize]> for ConstraintMatrix {
}
/// Return the [Candidate]s referred to in the given [ConstraintMatrixCell] at location `idx`
fn candidates_in_constraint_cell<'a, N: Number>(election: &'a Election<N>, candidates: &HashMap<&Candidate, CountCard<N>>, idx: &[usize]) -> Vec<&'a Candidate> {
fn candidates_in_constraint_cell<'a, N: Number>(election: &'a Election<N>, candidates: &HashMap<&Candidate, CountCard<N>, BuildNoHashHasher<Candidate>>, idx: &[usize]) -> Vec<&'a Candidate> {
let mut result: Vec<&Candidate> = Vec::new();
for (i, candidate) in election.candidates.iter().enumerate() {
let cc = &candidates[candidate];
@ -782,7 +783,7 @@ pub fn init_repeat_count<N: Number>(election: &mut Election<N>) {
/// Initialise the rollback for [ConstraintMode::TwoStage]
pub fn init_repeat_count_rollback<'a, N: Number>(state: &mut CountState<'a, N>, constraint: &'a Constraint, group: &'a ConstrainedGroup) {
let mut rollback_candidates = HashMap::new();
let mut rollback_candidates = HashMap::with_capacity_and_hasher(state.candidates.len(), BuildNoHashHasher::default());
let rollback_exhausted = state.exhausted.clone();
// Copy ballot papers to rollback state

View File

@ -24,6 +24,7 @@ use crate::stv::gregory::TransferTable;
use crate::stv::meek::BallotTree;
use itertools::Itertools;
use nohash_hasher::BuildNoHashHasher;
#[cfg(not(target_arch = "wasm32"))]
use rkyv::{Archive, Deserialize, Serialize};
@ -33,6 +34,7 @@ use crate::numbers::{SerializedNumber, SerializedOptionNumber};
use std::cmp::max;
use std::collections::HashMap;
use std::fmt;
use std::hash::{Hash, Hasher};
/// An election to be counted
#[cfg_attr(not(target_arch = "wasm32"), derive(Archive, Deserialize, Serialize))]
@ -98,15 +100,26 @@ impl<N: Number> Election<N> {
}
/// A candidate in an [Election]
#[derive(Clone, Eq, Hash, PartialEq)]
#[derive(Clone, Eq, PartialEq)]
#[cfg_attr(not(target_arch = "wasm32"), derive(Archive, Deserialize, Serialize))]
pub struct Candidate {
/// Index of the candidate
pub index: usize,
/// Name of the candidate
pub name: String,
/// If this candidate is a dummy candidate (e.g. for --constraint-mode repeat_count)
pub is_dummy: bool,
}
impl Hash for Candidate {
fn hash<H: Hasher>(&self, hasher: &mut H) {
// Custom implementation of hash for use with NoHashHasher, to improve performance
hasher.write_usize(self.index);
}
}
impl nohash_hasher::IsEnabled for Candidate {}
/// The current state of counting an [Election]
#[derive(Clone)]
pub struct CountState<'a, N: Number> {
@ -114,7 +127,7 @@ pub struct CountState<'a, N: Number> {
pub election: &'a Election<N>,
/// [HashMap] of [CountCard]s for each [Candidate] in the election
pub candidates: HashMap<&'a Candidate, CountCard<'a, N>>,
pub candidates: HashMap<&'a Candidate, CountCard<'a, N>, BuildNoHashHasher<Candidate>>,
/// [CountCard] representing the exhausted pile
pub exhausted: CountCard<'a, N>,
/// [CountCard] representing loss by fraction
@ -124,9 +137,9 @@ pub struct CountState<'a, N: Number> {
pub ballot_tree: Option<BallotTree<'a, N>>,
/// Values used to break ties, based on forwards tie-breaking
pub forwards_tiebreak: Option<HashMap<&'a Candidate, usize>>,
pub forwards_tiebreak: Option<HashMap<&'a Candidate, usize, BuildNoHashHasher<Candidate>>>,
/// Values used to break ties, based on backwards tie-breaking
pub backwards_tiebreak: Option<HashMap<&'a Candidate, usize>>,
pub backwards_tiebreak: Option<HashMap<&'a Candidate, usize, BuildNoHashHasher<Candidate>>>,
/// [SHARandom] for random tie-breaking
pub random: Option<SHARandom<'a>>,
@ -161,7 +174,7 @@ impl<'a, N: Number> CountState<'a, N> {
pub fn new(election: &'a Election<N>) -> Self {
let mut state = CountState {
election,
candidates: HashMap::new(),
candidates: HashMap::with_capacity_and_hasher(election.candidates.len(), BuildNoHashHasher::default()),
exhausted: CountCard::new(),
loss_fraction: CountCard::new(),
ballot_tree: None,
@ -590,7 +603,18 @@ pub enum RollbackState<'a, N> {
/// Not rolling back
Normal,
/// Start rolling back next stage
NeedsRollback { candidates: Option<HashMap<&'a Candidate, CountCard<'a, N>>>, exhausted: Option<CountCard<'a, N>>, constraint: &'a Constraint, group: &'a ConstrainedGroup },
NeedsRollback {
candidates: Option<HashMap<&'a Candidate, CountCard<'a, N>, BuildNoHashHasher<Candidate>>>,
exhausted: Option<CountCard<'a, N>>,
constraint: &'a Constraint,
group: &'a ConstrainedGroup
},
/// Rolling back
RollingBack { candidates: Option<HashMap<&'a Candidate, CountCard<'a, N>>>, exhausted: Option<CountCard<'a, N>>, candidate_distributing: Option<&'a Candidate>, constraint: Option<&'a Constraint>, group: Option<&'a ConstrainedGroup> },
RollingBack {
candidates: Option<HashMap<&'a Candidate, CountCard<'a, N>, BuildNoHashHasher<Candidate>>>,
exhausted: Option<CountCard<'a, N>>,
candidate_distributing: Option<&'a Candidate>,
constraint: Option<&'a Constraint>,
group: Option<&'a ConstrainedGroup>
},
}

View File

@ -175,9 +175,10 @@ impl<N: Number, I: Iterator<Item=char>> BLTParser<N, I> {
/// Parse the list of strings at the end of the BLT file
fn strings(&mut self) -> Result<(), ParseError> {
for _ in 0..self.num_candidates {
for index in 0..self.num_candidates {
let name = self.string()?;
self.election.candidates.push(Candidate {
index,
name,
is_dummy: false,
});

View File

@ -36,13 +36,14 @@ pub fn parse_reader<R: Read, N: Number>(reader: R, require_1: bool, require_sequ
let mut candidates = Vec::new();
let mut col_map = HashMap::new(); // Map csp column -> candidates index
for (i, cand_name) in reader.headers()?.into_iter().enumerate() {
for (index, cand_name) in reader.headers()?.into_iter().enumerate() {
if cand_name == "$mult" {
continue;
}
col_map.insert(i, candidates.len());
col_map.insert(index, candidates.len());
candidates.push(Candidate {
index,
name: cand_name.to_string(),
is_dummy: false,
});

View File

@ -15,6 +15,8 @@
* along with this program. If not, see <https://www.gnu.org/licenses/>.
*/
use nohash_hasher::BuildNoHashHasher;
#[cfg(not(target_arch = "wasm32"))]
use prettytable::{Cell, Row, Table};
#[cfg(target_arch = "wasm32")]
@ -52,13 +54,15 @@ pub struct TransferTable<'e, N: Number> {
impl<'e, N: Number> TransferTable<'e, N> {
/// Return a new [TransferTable] for an exclusion
pub fn new_exclusion(hopefuls: Vec<&'e Candidate>) -> Self {
TransferTable {
let num_hopefuls = hopefuls.len();
return TransferTable {
hopefuls,
columns: Vec::new(),
total: TransferTableColumn {
value_fraction: N::new(),
order: 0,
cells: HashMap::new(),
cells: HashMap::with_capacity_and_hasher(num_hopefuls, BuildNoHashHasher::default()),
exhausted: TransferTableCell { ballots: N::new(), votes_in: N::new(), votes_out: N::new() },
total: TransferTableCell { ballots: N::new(), votes_in: N::new(), votes_out: N::new() },
},
@ -71,13 +75,15 @@ impl<'e, N: Number> TransferTable<'e, N> {
/// Return a new [TransferTable] for a surplus distribution
pub fn new_surplus(hopefuls: Vec<&'e Candidate>, surplus: N, surpfrac: Option<N>, surpfrac_numer: Option<N>, surpfrac_denom: Option<N>) -> Self {
TransferTable {
let num_hopefuls = hopefuls.len();
return TransferTable {
hopefuls,
columns: Vec::new(),
total: TransferTableColumn {
value_fraction: N::new(),
order: 0,
cells: HashMap::new(),
cells: HashMap::with_capacity_and_hasher(num_hopefuls, BuildNoHashHasher::default()),
exhausted: TransferTableCell { ballots: N::new(), votes_in: N::new(), votes_out: N::new() },
total: TransferTableCell { ballots: N::new(), votes_in: N::new(), votes_out: N::new() },
},
@ -102,7 +108,7 @@ impl<'e, N: Number> TransferTable<'e, N> {
let mut col = TransferTableColumn {
value_fraction: value_fraction.clone(),
order: order.unwrap_or(0),
cells: HashMap::new(),
cells: HashMap::with_hasher(BuildNoHashHasher::default()),
exhausted: TransferTableCell { ballots: N::new(), votes_in: N::new(), votes_out: N::new() },
total: TransferTableCell { ballots: N::new(), votes_in: N::new(), votes_out: N::new() },
};
@ -124,7 +130,7 @@ impl<'e, N: Number> TransferTable<'e, N> {
let col = TransferTableColumn {
value_fraction: value_fraction.clone(),
order: order.unwrap_or(0),
cells: HashMap::new(),
cells: HashMap::with_hasher(BuildNoHashHasher::default()),
exhausted: TransferTableCell { ballots: ballots.clone(), votes_in: N::new(), votes_out: N::new() },
total: TransferTableCell { ballots: N::new(), votes_in: N::new(), votes_out: N::new() },
};
@ -585,7 +591,7 @@ pub struct TransferTableColumn<'e, N: Number> {
pub order: usize,
/// Cells in this column
pub cells: HashMap<&'e Candidate, TransferTableCell<N>>,
pub cells: HashMap<&'e Candidate, TransferTableCell<N>, BuildNoHashHasher<Candidate>>,
/// Exhausted cell
pub exhausted: TransferTableCell<N>,

View File

@ -21,6 +21,7 @@ use crate::election::{Ballot, Candidate, CandidateState, CountCard, CountState,
use crate::numbers::Number;
use itertools::Itertools;
use nohash_hasher::BuildNoHashHasher;
use std::cmp::max;
use std::collections::HashMap;
@ -39,7 +40,7 @@ struct BallotInTree<'b, N: Number> {
pub struct BallotTree<'t, N: Number> {
num_ballots: N,
ballots: Vec<BallotInTree<'t, N>>,
next_preferences: Option<Box<HashMap<&'t Candidate, BallotTree<'t, N>>>>,
next_preferences: Option<Box<HashMap<&'t Candidate, BallotTree<'t, N>, BuildNoHashHasher<Candidate>>>>,
next_exhausted: Option<Box<BallotTree<'t, N>>>,
}
@ -56,7 +57,7 @@ impl<'t, N: Number> BallotTree<'t, N> {
/// Descend one level of the [BallotTree]
fn descend_tree(&mut self, candidates: &'t [Candidate]) {
let mut next_preferences: HashMap<&Candidate, BallotTree<N>> = HashMap::new();
let mut next_preferences: HashMap<&Candidate, BallotTree<N>, BuildNoHashHasher<Candidate>> = HashMap::with_capacity_and_hasher(candidates.len(), BuildNoHashHasher::default());
let mut next_exhausted = BallotTree::new();
for bit in self.ballots.iter() {
@ -164,7 +165,7 @@ where
/// Distribute preferences recursively
///
/// Called by [distribute_preferences]
fn distribute_recursively<'t, N: Number>(candidates: &mut HashMap<&'t Candidate, CountCard<N>>, exhausted: &mut CountCard<N>, tree: &mut BallotTree<'t, N>, remaining_multiplier: N, election: &'t Election<N>, opts: &STVOptions)
fn distribute_recursively<'t, N: Number>(candidates: &mut HashMap<&'t Candidate, CountCard<N>, BuildNoHashHasher<Candidate>>, exhausted: &mut CountCard<N>, tree: &mut BallotTree<'t, N>, remaining_multiplier: N, election: &'t Election<N>, opts: &STVOptions)
where
for<'r> &'r N: ops::Mul<&'r N, Output=N>,
{

View File

@ -35,6 +35,7 @@ use crate::ties::{self, TieStrategy};
use derive_builder::Builder;
use derive_more::Constructor;
use itertools::Itertools;
use nohash_hasher::BuildNoHashHasher;
#[allow(unused_imports)]
use wasm_bindgen::prelude::wasm_bindgen;
@ -781,7 +782,7 @@ where
/// See [next_preferences]
pub struct NextPreferencesResult<'a, N> {
/// [NextPreferencesEntry] for each [Candidate]
pub candidates: HashMap<&'a Candidate, NextPreferencesEntry<'a, N>>,
pub candidates: HashMap<&'a Candidate, NextPreferencesEntry<'a, N>, BuildNoHashHasher<Candidate>>,
/// [NextPreferencesEntry] for exhausted ballots
pub exhausted: NextPreferencesEntry<'a, N>,
/// Total weight of ballots examined
@ -799,7 +800,7 @@ pub struct NextPreferencesEntry<'a, N> {
/// Count the given votes, grouping according to next available preference
pub fn next_preferences<'a, N: Number>(state: &CountState<'a, N>, votes: Vec<Vote<'a, N>>) -> NextPreferencesResult<'a, N> {
let mut result = NextPreferencesResult {
candidates: HashMap::new(),
candidates: HashMap::with_capacity_and_hasher(state.election.candidates.len(), BuildNoHashHasher::default()),
exhausted: NextPreferencesEntry {
votes: Vec::new(),
num_ballots: N::new(),
@ -1767,7 +1768,7 @@ fn init_tiebreaks<N: Number>(state: &mut CountState<N>, opts: &STVOptions) {
// Update forwards tie-breaking order
if opts.ties.iter().any(|t| t == &TieStrategy::Forwards) {
let mut hm: HashMap<&Candidate, usize> = HashMap::new();
let mut hm: HashMap<&Candidate, usize, BuildNoHashHasher<Candidate>> = HashMap::with_capacity_and_hasher(state.candidates.len(), BuildNoHashHasher::default());
for (i, group) in sorted_candidates.iter().enumerate() {
for (candidate, _) in group.iter() {
hm.insert(candidate, i);
@ -1778,7 +1779,7 @@ fn init_tiebreaks<N: Number>(state: &mut CountState<N>, opts: &STVOptions) {
// Update backwards tie-breaking order
if opts.ties.iter().any(|t| t == &TieStrategy::Backwards) {
let mut hm: HashMap<&Candidate, usize> = HashMap::new();
let mut hm: HashMap<&Candidate, usize, BuildNoHashHasher<Candidate>> = HashMap::with_capacity_and_hasher(state.candidates.len(), BuildNoHashHasher::default());
for (i, group) in sorted_candidates.iter().enumerate() {
for (candidate, _) in group.iter() {
hm.insert(candidate, i);