tisonkun commented on code in PR #1: URL: https://github.com/apache/datasketches-rust/pull/1#discussion_r2616631864
########## src/hll/sketch.rs: ########## @@ -0,0 +1,422 @@ +// Licensed to the Apache Software Foundation (ASF) under one +// or more contributor license agreements. See the NOTICE file +// distributed with this work for additional information +// regarding copyright ownership. The ASF licenses this file +// to you under the Apache License, Version 2.0 (the +// "License"); you may not use this file except in compliance +// with the License. You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, +// software distributed under the License is distributed on an +// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +// KIND, either express or implied. See the License for the +// specific language governing permissions and limitations +// under the License. + +//! HyperLogLog sketch implementation +//! +//! This module provides the main [`HllSketch`] struct, which is the primary interface +//! for creating and using HLL sketches for cardinality estimation. +//! +//! # Adaptive Mode System +//! +//! The sketch automatically transitions between three internal modes based on cardinality: +//! +//! - **List mode**: Stores individual coupons in a compact list for small cardinalities. +//! Used when fewer than ~32 unique values have been seen. +//! +//! - **Set mode**: Uses a hash set with open addressing for medium cardinalities. +//! Provides better performance than list mode while still being space-efficient. +//! The set grows dynamically until it reaches K/8 entries. +//! +//! - **HLL mode**: Uses the full HLL array (Array4, Array6, or Array8) for large cardinalities. +//! Provides constant memory usage and accurate estimates for billions of unique values. +//! +//! Mode transitions are automatic and transparent to the user. Each promotion preserves +//! all previously observed values and maintains estimation accuracy. +//! +//! # Serialization +//! +//! Sketches can be serialized and deserialized while preserving all state, including: +//! - Current mode and HLL type +//! - All observed values (coupons or register values) +//! - HIP accumulator state for accurate estimation +//! - Out-of-order flag for merged/deserialized sketches +//! +//! The serialization format is compatible with Apache DataSketches implementations +//! in Java and C++, enabling cross-platform sketch exchange. + +use std::hash::Hash; +use std::io; + +use crate::hll::array4::Array4; +use crate::hll::array6::Array6; +use crate::hll::array8::Array8; +use crate::hll::container::Container; +use crate::hll::hash_set::HashSet; +use crate::hll::list::List; +use crate::hll::serialization::*; +use crate::hll::{HllType, RESIZE_DENOM, RESIZE_NUMER, coupon}; + +/// Current sketch mode +#[derive(Debug, Clone, Copy, PartialEq, Eq)] +enum CurMode { + List = 0, + Set = 1, + Hll = 2, +} + +#[derive(Debug, Clone)] +pub struct HllSketch { + lg_config_k: u8, + mode: Mode, +} + +impl PartialEq for HllSketch { + fn eq(&self, other: &Self) -> bool { + if self.lg_config_k != other.lg_config_k { + return false; + } + + match (&self.mode, &other.mode) { + ( + Mode::List { + list: l1, + hll_type: t1, + }, + Mode::List { + list: l2, + hll_type: t2, + }, + ) => l1 == l2 && t1 == t2, + ( + Mode::Set { + set: s1, + hll_type: t1, + }, + Mode::Set { + set: s2, + hll_type: t2, + }, + ) => s1 == s2 && t1 == t2, + (Mode::Array4(a1), Mode::Array4(a2)) => a1 == a2, + (Mode::Array6(a1), Mode::Array6(a2)) => a1 == a2, + (Mode::Array8(a1), Mode::Array8(a2)) => a1 == a2, + _ => false, // Different modes are not equal + } + } +} + +#[derive(Debug, Clone)] +enum Mode { + List { list: List, hll_type: HllType }, + Set { set: HashSet, hll_type: HllType }, + Array4(Array4), + Array6(Array6), + Array8(Array8), +} + +impl HllSketch { + /// Create a new HLL sketch + /// + /// # Arguments + /// * `lg_config_k` - Log2 of the number of buckets (K). Must be in [4, 21]. + /// - lg_k=4: 16 buckets, ~26% relative error + /// - lg_k=12: 4096 buckets, ~1.6% relative error (common choice) + /// - lg_k=21: 2M buckets, ~0.4% relative error + /// * `hll_type` - Target HLL array type (Hll4, Hll6, or Hll8) + /// + /// # Panics + /// Panics if lg_config_k is not in [4, 21] + pub fn new(lg_config_k: u8, hll_type: HllType) -> Self { + assert!( + lg_config_k > 4 && lg_config_k < 21, + "lg_config_k must be in [4, 21]" + ); + + let list = List::default(); + + Self { + lg_config_k, + mode: Mode::List { list, hll_type }, + } + } + + pub fn lg_config_k(&self) -> u8 { + self.lg_config_k + } + + /// Update the sketch with a value + /// + /// This accepts any type that implements `Hash`. The value is hashed + /// and converted to a coupon, which is then inserted into the sketch. + /// The sketch will automatically promote from List → Set → HLL as needed. + pub fn update<T: Hash>(&mut self, value: T) { + let coupon = coupon(value); + self.update_with_coupon(coupon); + } + + /// Update the sketch with a raw coupon value + /// + /// This is useful when you've already computed the coupon externally, + /// or when deserializing and replaying coupons. + fn update_with_coupon(&mut self, coupon: u32) { + // Perform the update + match &mut self.mode { + Mode::List { list, hll_type } => { + list.update(coupon); + let should_promote = list.container().is_full(); + if should_promote { + self.mode = if self.lg_config_k < 8 { + promote_container_to_array(list.container(), *hll_type, self.lg_config_k) + } else { + promote_container_to_set(list.container(), *hll_type) + } + } + } + Mode::Set { set, hll_type } => { + set.update(coupon); + let should_promote = RESIZE_DENOM as usize * set.container().len() + > RESIZE_NUMER as usize * set.container().capacity(); + if should_promote { + self.mode = if set.container().lg_size() == self.lg_config_k as usize - 3 { + promote_container_to_array(set.container(), *hll_type, self.lg_config_k) + } else { + grow_set(set, *hll_type) + } + } + } + Mode::Array4(arr) => arr.update(coupon), + Mode::Array6(arr) => arr.update(coupon), + Mode::Array8(arr) => arr.update(coupon), + } + } + + /// Get the current cardinality estimate + pub fn estimate(&self) -> f64 { + match &self.mode { + Mode::List { list, .. } => list.container().estimate(), + Mode::Set { set, .. } => set.container().estimate(), + Mode::Array4(arr) => arr.estimate(), + Mode::Array6(arr) => arr.estimate(), + Mode::Array8(arr) => arr.estimate(), + } + } + + /// Get upper bound for cardinality estimate + /// + /// Returns the upper confidence bound for the cardinality estimate based on + /// the number of standard deviations requested. + pub fn upper_bound(&self, num_std_dev: u8) -> f64 { + match &self.mode { + Mode::List { list, .. } => list.container().upper_bound(num_std_dev), + Mode::Set { set, .. } => set.container().upper_bound(num_std_dev), + Mode::Array4(arr) => arr.upper_bound(num_std_dev), + Mode::Array6(arr) => arr.upper_bound(num_std_dev), + Mode::Array8(arr) => arr.upper_bound(num_std_dev), + } + } + + /// Get lower bound for cardinality estimate + /// + /// Returns the lower confidence bound for the cardinality estimate based on + /// the number of standard deviations requested. + pub fn lower_bound(&self, num_std_dev: u8) -> f64 { + match &self.mode { + Mode::List { list, .. } => list.container().lower_bound(num_std_dev), + Mode::Set { set, .. } => set.container().lower_bound(num_std_dev), + Mode::Array4(arr) => arr.lower_bound(num_std_dev), + Mode::Array6(arr) => arr.lower_bound(num_std_dev), + Mode::Array8(arr) => arr.lower_bound(num_std_dev), + } + } + + pub fn deserialize(bytes: &[u8]) -> io::Result<HllSketch> { Review Comment: For all `serialize`/`deserialize` methods here, since we don't do IO anyway, I suggest we define DataSketches serde error. We may further optionally integrate with the [`serde`](https://serde.rs/impl-serialize.html) crate but ser/de sketches should work without `serde` dep. (Its usage requires a serializer/deserializer that involves other concepts beyond sketches from/to bytes) -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
