asap_sketchlib 0.3.0

A high-performance sketching library for approximate stream processing
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
//! Count Sketch implementation.
//!
//! A frequency-estimation sketch that uses random sign projections to estimate
//! item counts with bounded error in the L2 norm.
//!
//! Reference:
//! - Charikar, Chen & Farach-Colton, "Finding Frequent Items in Data Streams,"
//!   ICALP 2002. <https://www.cs.rutgers.edu/~farach/pubs/FrequentStream.pdf>

use crate::common::structure_utils::AdmittedRows;
use crate::{
    DataInput, DefaultMatrixI32, DefaultMatrixI64, DefaultMatrixI128, DefaultXxHasher, FastPath,
    FastPathHasher, FixedMatrix, MatrixFastHash, MatrixStorage, NitroTarget, QuickMatrixI64,
    QuickMatrixI128, RegularPath, SketchHasher, Vector2D, hash64_seeded, nitro_delta_saturated_i32,
};
use serde::{Deserialize, Serialize};
use std::marker::PhantomData;
use std::ops::Neg;

mod wire;
pub(crate) use wire::{CsWireCounter, CsWireMode};

const DEFAULT_ROW_NUM: usize = 3;
const DEFAULT_COL_NUM: usize = 4096;
const LOWER_32_MASK: u64 = (1u64 << 32) - 1;

/// A frequency-estimation sketch using random sign projections (Count Sketch).
#[derive(Clone, Debug, Serialize, Deserialize)]
#[serde(bound(serialize = "S: Serialize", deserialize = "S: Deserialize<'de>"))]
pub struct Count<
    S: MatrixStorage = Vector2D<i32>,
    Mode = RegularPath,
    H: SketchHasher = DefaultXxHasher,
> {
    counts: S,
    row: usize,
    col: usize,
    #[serde(skip)]
    _mode: PhantomData<Mode>,
    #[serde(skip)]
    _hasher: PhantomData<H>,
}

/// Counter trait required by Count Sketch backends.
pub trait CountSketchCounter: Copy + std::ops::AddAssign + Neg<Output = Self> + From<i32> {
    /// Converts the counter into `f64`.
    fn to_f64(self) -> f64;
}

// Implements CountSketchCounter for i32.
impl CountSketchCounter for i32 {
    fn to_f64(self) -> f64 {
        self as f64
    }
}

// Implements CountSketchCounter for i64.
impl CountSketchCounter for i64 {
    fn to_f64(self) -> f64 {
        self as f64
    }
}

// Implements CountSketchCounter for i128.
impl CountSketchCounter for i128 {
    fn to_f64(self) -> f64 {
        self as f64
    }
}

// Default Count sketch for Vector2D<i32> (RegularPath).
impl Default for Count<Vector2D<i32>, RegularPath> {
    fn default() -> Self {
        Self::with_dimensions(DEFAULT_ROW_NUM, DEFAULT_COL_NUM)
    }
}

// Default Count sketch for Vector2D<i32> (FastPath).
impl Default for Count<Vector2D<i32>, FastPath> {
    fn default() -> Self {
        Self::with_dimensions(DEFAULT_ROW_NUM, DEFAULT_COL_NUM)
    }
}

// Default Count sketch for Vector2D<i64> (RegularPath).
impl Default for Count<Vector2D<i64>, RegularPath> {
    fn default() -> Self {
        Self::with_dimensions(DEFAULT_ROW_NUM, DEFAULT_COL_NUM)
    }
}

// Default Count sketch for Vector2D<i64> (FastPath).
impl Default for Count<Vector2D<i64>, FastPath> {
    fn default() -> Self {
        Self::with_dimensions(DEFAULT_ROW_NUM, DEFAULT_COL_NUM)
    }
}

// Default Count sketch for Vector2D<i128> (RegularPath).
impl Default for Count<Vector2D<i128>, RegularPath> {
    fn default() -> Self {
        Self::with_dimensions(DEFAULT_ROW_NUM, DEFAULT_COL_NUM)
    }
}

// Default Count sketch for Vector2D<i128> (FastPath).
impl Default for Count<Vector2D<i128>, FastPath> {
    fn default() -> Self {
        Self::with_dimensions(DEFAULT_ROW_NUM, DEFAULT_COL_NUM)
    }
}

// Default Count sketch for FixedMatrix (RegularPath).
impl Default for Count<FixedMatrix, RegularPath> {
    fn default() -> Self {
        Count::from_storage(FixedMatrix::default())
    }
}

// Default Count sketch for FixedMatrix (FastPath).
impl Default for Count<FixedMatrix, FastPath> {
    fn default() -> Self {
        Count::from_storage(FixedMatrix::default())
    }
}

// Default Count sketch for DefaultMatrixI32 (RegularPath).
impl Default for Count<DefaultMatrixI32, RegularPath> {
    fn default() -> Self {
        Count::from_storage(DefaultMatrixI32::default())
    }
}

// Default Count sketch for DefaultMatrixI32 (FastPath).
impl Default for Count<DefaultMatrixI32, FastPath> {
    fn default() -> Self {
        Count::from_storage(DefaultMatrixI32::default())
    }
}

// Default Count sketch for DefaultMatrixI64 (RegularPath).
impl Default for Count<DefaultMatrixI64, RegularPath> {
    fn default() -> Self {
        Count::from_storage(DefaultMatrixI64::default())
    }
}

// Default Count sketch for DefaultMatrixI64 (FastPath).
impl Default for Count<DefaultMatrixI64, FastPath> {
    fn default() -> Self {
        Count::from_storage(DefaultMatrixI64::default())
    }
}

// Default Count sketch for DefaultMatrixI128 (RegularPath).
impl Default for Count<DefaultMatrixI128, RegularPath> {
    fn default() -> Self {
        Count::from_storage(DefaultMatrixI128::default())
    }
}

// Default Count sketch for DefaultMatrixI128 (FastPath).
impl Default for Count<DefaultMatrixI128, FastPath> {
    fn default() -> Self {
        Count::from_storage(DefaultMatrixI128::default())
    }
}

// Default Count sketch for QuickMatrixI64 (RegularPath).
impl Default for Count<QuickMatrixI64, RegularPath> {
    fn default() -> Self {
        Count::from_storage(QuickMatrixI64::default())
    }
}

// Default Count sketch for QuickMatrixI64 (FastPath).
impl Default for Count<QuickMatrixI64, FastPath> {
    fn default() -> Self {
        Count::from_storage(QuickMatrixI64::default())
    }
}

// Default Count sketch for QuickMatrixI128 (RegularPath).
impl Default for Count<QuickMatrixI128, RegularPath> {
    fn default() -> Self {
        Count::from_storage(QuickMatrixI128::default())
    }
}

// Default Count sketch for QuickMatrixI128 (FastPath).
impl Default for Count<QuickMatrixI128, FastPath> {
    fn default() -> Self {
        Count::from_storage(QuickMatrixI128::default())
    }
}

// Count constructors for Vector2D-backed storage.
impl<T, M, H: SketchHasher> Count<Vector2D<T>, M, H>
where
    T: CountSketchCounter,
{
    /// Creates a sketch with the requested number of rows and columns.
    pub fn with_dimensions(rows: usize, cols: usize) -> Self {
        let mut sk = Count {
            counts: Vector2D::init(rows, cols),
            row: rows,
            col: cols,
            _mode: PhantomData,
            _hasher: PhantomData,
        };
        sk.counts.fill(T::from(0));
        sk
    }
}

// Core Count API for any storage/counter.
impl<S, C, Mode, H: SketchHasher> Count<S, Mode, H>
where
    S: MatrixStorage<Counter = C>,
    C: CountSketchCounter,
{
    /// Wraps an existing matrix storage as a Count Sketch.
    pub fn from_storage(counts: S) -> Self {
        let row = counts.rows();
        let col = counts.cols();
        Self {
            counts,
            row,
            col,
            _mode: PhantomData,
            _hasher: PhantomData,
        }
    }

    /// Number of rows in the sketch.
    pub fn rows(&self) -> usize {
        self.counts.rows()
    }

    /// Number of columns in the sketch.
    pub fn cols(&self) -> usize {
        self.counts.cols()
    }

    /// Merges another sketch while asserting compatible dimensions.
    pub fn merge(&mut self, other: &Self) {
        let self_rows = self.counts.rows();
        let self_cols = self.counts.cols();
        assert_eq!(
            (self_rows, self_cols),
            (other.counts.rows(), other.counts.cols()),
            "dimension mismatch while merging CountMin sketches"
        );

        for i in 0..self_rows {
            for j in 0..self_cols {
                self.counts.update_one_counter(
                    i,
                    j,
                    |a, b| *a += b,
                    other.counts.query_one_counter(i, j),
                );
            }
        }
    }

    /// Exposes the backing matrix for inspection/testing.
    pub fn as_storage(&self) -> &S {
        &self.counts
    }

    /// Mutable access used internally for testing scenarios.
    pub fn as_storage_mut(&mut self) -> &mut S {
        &mut self.counts
    }
}

// Regular-path Count operations.
impl<S, C, H: SketchHasher> Count<S, RegularPath, H>
where
    S: MatrixStorage<Counter = C>,
    C: CountSketchCounter,
{
    /// Inserts an observation with standard Count Sketch updating algorithm.
    pub fn insert(&mut self, value: &DataInput) {
        let rows = self.counts.rows();
        let cols = self.counts.cols();
        for r in 0..rows {
            let hashed = H::hash64_seeded(r, value);
            let col = ((hashed & LOWER_32_MASK) as usize) % cols;
            let bit = ((hashed >> 63) & 1) as i32;
            let sign_bit = if bit == 1 { 1 } else { -1 };
            let delta = if sign_bit > 0 {
                C::from(1)
            } else {
                -C::from(1)
            };
            self.counts
                .update_one_counter(r, col, |a, b| *a += b, delta);
        }
    }

    /// Inserts an observation with the given count (weight).
    pub fn insert_many(&mut self, value: &DataInput, many: C) {
        let rows = self.counts.rows();
        let cols = self.counts.cols();
        for r in 0..rows {
            let hashed = H::hash64_seeded(r, value);
            let col = ((hashed & LOWER_32_MASK) as usize) % cols;
            let bit = ((hashed >> 63) & 1) as i32;
            let sign_bit = if bit == 1 { 1 } else { -1 };
            let delta = if sign_bit > 0 { many } else { -many };
            self.counts
                .update_one_counter(r, col, |a, b| *a += b, delta);
        }
    }

    /// Returns the frequency estimate for the provided value.
    pub fn estimate(&self, value: &DataInput) -> f64 {
        let rows = self.counts.rows();
        let cols = self.counts.cols();
        let mut estimates = Vec::with_capacity(rows);
        for r in 0..rows {
            let hashed = H::hash64_seeded(r, value);
            let col = ((hashed & LOWER_32_MASK) as usize) % cols;
            let bit = ((hashed >> 63) & 1) as i32;
            let sign_bit = if bit == 1 { 1 } else { -1 };
            let counter = self.counts.query_one_counter(r, col);
            if sign_bit > 0 {
                estimates.push(counter.to_f64());
            } else {
                estimates.push(-counter.to_f64());
            }
        }
        if estimates.is_empty() {
            return 0.0;
        }
        estimates.sort_unstable_by(|a, b| a.partial_cmp(b).unwrap());
        let mid = estimates.len() / 2;
        if estimates.len() % 2 == 1 {
            estimates[mid]
        } else {
            (estimates[mid - 1] + estimates[mid]) / 2.0
        }
    }
}

// Fast-path Count operations using precomputed hashes.
impl<S, H: SketchHasher> Count<S, FastPath, H>
where
    S: MatrixStorage + crate::FastPathHasher<H>,
    S::Counter: CountSketchCounter,
{
    /// Inserts an observation using the combined hash optimization.
    #[inline(always)]
    pub fn insert(&mut self, value: &DataInput) {
        let hashed_val = <S as FastPathHasher<H>>::hash_for_matrix(&self.counts, value);
        self.counts.fast_insert(
            |counter, value, row| {
                let sign = hashed_val.sign_for_row(row);
                let delta = if sign > 0 { *value } else { -*value };
                *counter += delta;
            },
            S::Counter::from(1),
            &hashed_val,
        );
    }

    /// Inserts an observation with the given count using the combined hash optimization.
    #[inline(always)]
    pub fn insert_many(&mut self, value: &DataInput, many: S::Counter) {
        let hashed_val = <S as FastPathHasher<H>>::hash_for_matrix(&self.counts, value);
        self.counts.fast_insert(
            |counter, value, row| {
                let sign = hashed_val.sign_for_row(row);
                let delta = if sign > 0 { *value } else { -*value };
                *counter += delta;
            },
            many,
            &hashed_val,
        );
    }

    /// Returns the frequency estimate for the provided value.
    #[inline(always)]
    pub fn estimate(&self, value: &DataInput) -> f64 {
        let hashed_val = <S as FastPathHasher<H>>::hash_for_matrix(&self.counts, value);
        self.counts
            .fast_query_median(&hashed_val, |val, row, hash| {
                let sign = hash.sign_for_row(row);
                if sign > 0 {
                    (*val).to_f64()
                } else {
                    -(*val).to_f64()
                }
            })
    }

    /// Inserts an observation using a pre-computed hash value.
    /// Hash value can be reused with other sketches.
    #[inline(always)]
    pub fn fast_insert_with_hash_value(&mut self, hashed_val: &H::HashType) {
        self.counts.fast_insert(
            |counter, value, row| {
                let sign = hashed_val.sign_for_row(row);
                let delta = if sign > 0 { *value } else { -*value };
                *counter += delta;
            },
            S::Counter::from(1),
            hashed_val,
        );
    }

    /// Inserts an observation with the given count using a pre-computed hash value.
    #[inline(always)]
    pub fn fast_insert_many_with_hash_value(&mut self, hashed_val: &H::HashType, many: S::Counter) {
        self.counts.fast_insert(
            |counter, value, row| {
                let sign = hashed_val.sign_for_row(row);
                let delta = if sign > 0 { *value } else { -*value };
                *counter += delta;
            },
            many,
            hashed_val,
        );
    }

    /// Returns the frequency estimate using a pre-computed hash value.
    #[inline(always)]
    pub fn fast_estimate_with_hash(&self, hashed_val: &H::HashType) -> f64 {
        self.counts.fast_query_median(hashed_val, |val, row, hash| {
            let sign = hash.sign_for_row(row);
            if sign > 0 {
                (*val).to_f64()
            } else {
                -(*val).to_f64()
            }
        })
    }
}

// Debug helpers for i32 Vector2D Count.
impl<M, H: SketchHasher> Count<Vector2D<i32>, M, H> {
    /// Human-friendly helper used by the serializer demo binaries.
    pub fn debug(&self) {
        for row in 0..self.counts.rows() {
            println!("row {row}: {:?}", self.counts.row_slice(row));
        }
    }
}

// Nitro sampling helpers for fast-path Count.
impl<H: SketchHasher> Count<Vector2D<i32>, FastPath, H> {
    /// Enables Nitro sampling with the provided rate.
    pub fn enable_nitro(&mut self, sampling_rate: f64) {
        self.counts.enable_nitro(sampling_rate);
    }

    /// Enables Nitro sampling with a reproducible, seed-selected schedule.
    ///
    /// See [`crate::Nitro::init_nitro_seeded`]: the unseeded path starts every
    /// sketch at the same point in the skip table, so two sketches at the same
    /// rate admit the same subset and are not independent trials.
    pub fn enable_nitro_with_seed(&mut self, sampling_rate: f64, seed: u64) {
        self.counts.enable_nitro_with_seed(sampling_rate, seed);
    }

    /// Inserts an observation through Nitro's per-row sampling schedule.
    ///
    /// Cells and signs are derived exactly as a plain `insert` derives them —
    /// `FastPathHasher::hash_for_matrix`, then `col_for_row` and
    /// `sign_for_row` — so `estimate` reads back what this wrote. The hash is
    /// computed once per observation, and the admitted rows are collected into
    /// an inline buffer, so the hot path does not allocate.
    #[inline(always)]
    pub fn fast_insert_nitro(&mut self, value: &DataInput) {
        let rows = self.counts.rows();
        let mut admitted = AdmittedRows::new();
        self.counts.nitro_mut().admit_rows(rows, &mut admitted);
        if admitted.is_empty() {
            return;
        }
        let cols = self.counts.cols();
        let hashed = <Vector2D<i32> as FastPathHasher<H>>::hash_for_matrix(&self.counts, value);
        for (row, weight) in admitted {
            let col = MatrixFastHash::col_for_row(&hashed, row, cols);
            let signed = if hashed.sign_for_row(row) > 0 {
                nitro_delta_saturated_i32(weight)
            } else {
                -nitro_delta_saturated_i32(weight)
            };
            self.counts
                .update_one_counter(row, col, |a: &mut i32, b: i32| *a += b, signed);
        }
    }
}

// NitroTarget integration for fast-path Count.
impl<H: SketchHasher> NitroTarget for Count<Vector2D<i32>, FastPath, H> {
    #[inline(always)]
    fn rows(&self) -> usize {
        self.counts.rows()
    }

    #[inline(always)]
    fn update_row(&mut self, row: usize, hashed: u128, delta: u64) {
        let bit = (hashed >> (127 - row)) & 1;
        let sign = (bit << 1) as i32 - 1;
        self.counts.update_by_row(
            row,
            hashed,
            |a, b| *a += b,
            sign * nitro_delta_saturated_i32(delta),
        );
    }

    #[inline(always)]
    fn update_sample(&mut self, value: &DataInput, delta: u64) {
        // Hash with the SAME derivation the estimator uses so both share one
        // hash domain (raw hash128_seeded bit-slicing diverges in Packed64 mode).
        let hashed = <Vector2D<i32> as FastPathHasher<H>>::hash_for_matrix(&self.counts, value);
        let cols = self.counts.cols();
        for row in 0..self.counts.rows() {
            let col = <H::HashType as MatrixFastHash>::col_for_row(&hashed, row, cols);
            let sign = <H::HashType as MatrixFastHash>::sign_for_row(&hashed, row);
            self.counts.update_one_counter(
                row,
                col,
                |a, b| *a += b,
                sign * nitro_delta_saturated_i32(delta),
            );
        }
    }
}

use crate::input_to_owned;
use crate::octo_delta::{COUNT_PROMASK, CountDelta, KeyedCountDelta, MAX_PROMASK};

impl<S: MatrixStorage<Counter = i32>, H: SketchHasher> Count<S, RegularPath, H> {
    /// Inserts a value, emitting a delta at every promotion of the default
    /// threshold `COUNT_PROMASK`.
    #[inline(always)]
    pub fn insert_emit_delta(&mut self, value: &DataInput, emit: &mut impl FnMut(CountDelta)) {
        self.insert_emit_delta_with_threshold(value, COUNT_PROMASK, emit);
    }

    /// Inserts a value, emitting a delta and clearing the counter each time
    /// `|counter|` reaches `threshold`. Count sketch counters are signed, so
    /// the magnitude is what the threshold is applied to (§4.4).
    #[inline(always)]
    pub fn insert_emit_delta_with_threshold(
        &mut self,
        value: &DataInput,
        threshold: u32,
        emit: &mut impl FnMut(CountDelta),
    ) {
        let threshold = threshold.clamp(1, MAX_PROMASK);
        let rows = self.counts.rows();
        let cols = self.counts.cols();
        for r in 0..rows {
            let hashed = hash64_seeded(r, value);
            let col = ((hashed & LOWER_32_MASK) as usize) % cols;
            let sign: i32 = if ((hashed >> 63) & 1) == 1 { 1 } else { -1 };
            self.counts.increment_by_row(r, col, sign);
            let current = self.counts.query_one_counter(r, col);
            if current.unsigned_abs() >= threshold {
                emit(CountDelta {
                    row: r as u32,
                    col: col as u32,
                    value: current,
                });
                self.counts.update_one_counter(r, col, |c, _| *c = 0, ());
            }
        }
    }

    /// As `insert_emit_delta_with_threshold`, but every delta carries the flow
    /// key so an aggregator can maintain a heavy-hitter heap.
    pub fn insert_emit_keyed_delta_with_threshold(
        &mut self,
        value: &DataInput,
        threshold: u32,
        emit: &mut impl FnMut(KeyedCountDelta),
    ) {
        self.insert_emit_delta_with_threshold(value, threshold, &mut |delta| {
            emit(KeyedCountDelta {
                key: input_to_owned(value),
                delta,
            })
        });
    }
}

impl<S, H: SketchHasher> Count<S, FastPath, H>
where
    S: MatrixStorage<Counter = i32> + FastPathHasher<H>,
{
    /// Inserts a value using the fast path, emitting a delta at every
    /// promotion of the default threshold `COUNT_PROMASK`.
    #[inline(always)]
    pub fn insert_emit_delta(&mut self, value: &DataInput, emit: &mut impl FnMut(CountDelta)) {
        self.insert_emit_delta_with_threshold(value, COUNT_PROMASK, emit);
    }

    /// Fast-path counterpart of the regular-path threshold API.
    #[inline(always)]
    pub fn insert_emit_delta_with_threshold(
        &mut self,
        value: &DataInput,
        threshold: u32,
        emit: &mut impl FnMut(CountDelta),
    ) {
        let threshold = threshold.clamp(1, MAX_PROMASK);
        let hashed_val = <S as FastPathHasher<H>>::hash_for_matrix(&self.counts, value);
        let rows = self.counts.rows();
        let cols = self.counts.cols();
        for r in 0..rows {
            let col = hashed_val.col_for_row(r, cols);
            let sign = hashed_val.sign_for_row(r);
            self.counts.increment_by_row(r, col, sign);
            let current = self.counts.query_one_counter(r, col);
            if current.unsigned_abs() >= threshold {
                emit(CountDelta {
                    row: r as u32,
                    col: col as u32,
                    value: current,
                });
                self.counts.update_one_counter(r, col, |c, _| *c = 0, ());
            }
        }
    }

    /// Fast-path counterpart of `insert_emit_keyed_delta_with_threshold`.
    pub fn insert_emit_keyed_delta_with_threshold(
        &mut self,
        value: &DataInput,
        threshold: u32,
        emit: &mut impl FnMut(KeyedCountDelta),
    ) {
        self.insert_emit_delta_with_threshold(value, threshold, &mut |delta| {
            emit(KeyedCountDelta {
                key: input_to_owned(value),
                delta,
            })
        });
    }
}

impl<S: MatrixStorage, Mode, H: SketchHasher> Count<S, Mode, H>
where
    S::Counter: Copy + std::ops::AddAssign + From<i32>,
{
    /// Applies a previously emitted delta to this sketch.
    pub fn apply_delta(&mut self, delta: CountDelta) {
        self.counts.increment_by_row(
            delta.row as usize,
            delta.col as usize,
            S::Counter::from(delta.value),
        );
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::test_utils::{all_counter_zero_i32, counter_index, sample_zipf_u64};
    use crate::{DataInput, hash64_seeded};
    use std::collections::HashMap;

    #[test]
    fn count_child_insert_emits_at_threshold() {
        let mut child = Count::<Vector2D<i32>, RegularPath>::with_dimensions(3, 64);
        let key = DataInput::U64(99);
        let mut deltas: Vec<CountDelta> = Vec::new();

        for _ in 0..200 {
            child.insert_emit_delta(&key, &mut |d| deltas.push(d));
        }
        assert!(
            deltas.len() >= 3,
            "expected at least one promoted delta per row"
        );
    }

    fn counter_sign(row: usize, key: &DataInput) -> i32 {
        let hash = hash64_seeded(row, key);
        if (hash >> 63) & 1 == 1 { 1 } else { -1 }
    }

    fn run_zipf_stream(
        rows: usize,
        cols: usize,
        domain: usize,
        exponent: f64,
        samples: usize,
        seed: u64,
    ) -> (Count, HashMap<u64, i32>) {
        let mut truth = HashMap::<u64, i32>::new();
        let mut sketch = Count::<Vector2D<i32>, RegularPath>::with_dimensions(rows, cols);

        for value in sample_zipf_u64(domain, exponent, samples, seed) {
            let key = DataInput::U64(value);
            sketch.insert(&key);
            *truth.entry(value).or_insert(0) += 1;
        }

        (sketch, truth)
    }

    #[test]
    fn default_initializes_expected_dimensions() {
        let cs = Count::<Vector2D<i32>, RegularPath>::default();
        assert_eq!(cs.rows(), 3);
        assert_eq!(cs.cols(), 4096);
        all_counter_zero_i32(cs.as_storage());
    }

    #[test]
    fn with_dimensions_uses_custom_sizes() {
        let cs = Count::<Vector2D<i32>, RegularPath>::with_dimensions(3, 17);
        assert_eq!(cs.rows(), 3);
        assert_eq!(cs.cols(), 17);

        let storage = cs.as_storage();
        for row in 0..cs.rows() {
            assert!(
                storage.row_slice(row).iter().all(|&value| value == 0),
                "expected row {} to be zero-initialized, got {:?}",
                row,
                storage.row_slice(row)
            );
        }
    }

    #[test]
    fn insert_updates_signed_counters_per_row() {
        let mut sketch = Count::<Vector2D<i32>, RegularPath>::with_dimensions(3, 64);
        let key = DataInput::Str("alpha");

        sketch.insert(&key);

        for row in 0..sketch.rows() {
            let idx = counter_index(row, &key, sketch.cols());
            let expected = counter_sign(row, &key);
            assert_eq!(
                sketch.counts.query_one_counter(row, idx),
                expected,
                "row {row} counter mismatch"
            );
        }
    }

    #[test]
    fn fast_insert_produces_consistent_estimates() {
        let mut fast = Count::<Vector2D<i32>, FastPath>::with_dimensions(4, 128);

        let keys = vec![
            DataInput::Str("alpha"),
            DataInput::Str("beta"),
            DataInput::Str("gamma"),
            DataInput::Str("delta"),
            DataInput::Str("epsilon"),
        ];

        for key in &keys {
            fast.insert(key);
        }

        for key in &keys {
            let estimate = fast.estimate(key);
            assert!(
                (estimate - 1.0).abs() < f64::EPSILON,
                "fast estimate for key {key:?} should be 1.0, got {estimate}"
            );
        }
    }

    #[test]
    fn insert_produces_consistent_estimates() {
        let mut sketch = Count::<Vector2D<i32>, RegularPath>::with_dimensions(3, 64);

        let keys = vec![
            DataInput::Str("alpha"),
            DataInput::Str("beta"),
            DataInput::Str("gamma"),
            DataInput::Str("delta"),
            DataInput::Str("epsilon"),
        ];

        for key in &keys {
            sketch.insert(key);
        }

        for key in &keys {
            let estimate = sketch.estimate(key);
            assert!(
                (estimate - 1.0).abs() < f64::EPSILON,
                "estimate for key {key:?} should be 1.0, got {estimate}"
            );
        }
    }

    #[test]
    fn estimate_recovers_frequency_for_repeated_key() {
        let mut sketch = Count::<Vector2D<i32>, RegularPath>::with_dimensions(3, 64);
        let key = DataInput::Str("theta");

        let repeats = 37;
        for _ in 0..repeats {
            sketch.insert(&key);
        }

        let estimate = sketch.estimate(&key);
        assert!(
            (estimate - repeats as f64).abs() < f64::EPSILON,
            "expected estimate {repeats}, got {estimate}"
        );
    }

    #[test]
    fn fast_path_recovers_repeated_insertions() {
        let mut sketch = Count::<Vector2D<i32>, FastPath>::with_dimensions(4, 256);
        let keys = vec![
            DataInput::Str("alpha"),
            DataInput::Str("beta"),
            DataInput::Str("gamma"),
            DataInput::Str("delta"),
            DataInput::Str("epsilon"),
        ];

        for _ in 0..5 {
            for key in &keys {
                sketch.insert(key);
            }
        }

        for key in &keys {
            let estimate = sketch.estimate(key);
            assert!(
                (estimate - 5.0).abs() < f64::EPSILON,
                "fast estimate for key {key:?} should be 5.0, got {estimate}"
            );
        }
    }

    #[test]
    fn merge_adds_counters_element_wise() {
        let mut left = Count::<Vector2D<i32>, RegularPath>::with_dimensions(2, 32);
        let mut right = Count::<Vector2D<i32>, RegularPath>::with_dimensions(2, 32);
        let key = DataInput::Str("delta");

        left.insert(&key);
        right.insert(&key);
        right.insert(&key);

        let left_indices: Vec<_> = (0..left.rows())
            .map(|row| counter_index(row, &key, left.cols()))
            .collect();

        left.merge(&right);

        for (row, idx) in left_indices.into_iter().enumerate() {
            let expected = counter_sign(row, &key) * 3;
            assert_eq!(left.as_storage().query_one_counter(row, idx), expected);
        }
    }

    #[test]
    #[should_panic(expected = "dimension mismatch while merging CountMin sketches")]
    fn merge_requires_matching_dimensions() {
        let mut left = Count::<Vector2D<i32>, RegularPath>::with_dimensions(2, 32);
        let right = Count::<Vector2D<i32>, RegularPath>::with_dimensions(3, 32);
        left.merge(&right);
    }

    #[test]
    fn zipf_stream_stays_within_twenty_percent_for_most_keys() {
        let (sketch, truth) = run_zipf_stream(5, 8192, 8192, 1.1, 200_000, 0x5eed_c0de);
        let mut within_tolerance = 0usize;
        for (&value, &count) in &truth {
            let estimate = sketch.estimate(&DataInput::U64(value));
            let rel_error = ((estimate - count as f64).abs()) / (count as f64);
            if rel_error < 0.20 {
                within_tolerance += 1;
            }
        }

        let total = truth.len();
        let accuracy = within_tolerance as f64 / total as f64;
        assert!(
            accuracy >= 0.70,
            "Only {:.2}% of keys within tolerance ({} of {}); expected at least 70%",
            accuracy * 100.0,
            within_tolerance,
            total
        );
    }

    #[test]
    fn cs_regular_path_correctness() {
        let mut sk = Count::<Vector2D<i32>, RegularPath>::default();
        // Insert values 0..9 once using the regular path.
        for i in 0..10 {
            sk.insert(&DataInput::I32(i));
        }

        // Build the expected counter array by mirroring the regular-path hashing logic.
        let storage = sk.as_storage();
        let rows = storage.rows();
        let cols = storage.cols();
        let mut expected_once = vec![0_i32; rows * cols];
        for i in 0..10 {
            let value = DataInput::I32(i);
            for r in 0..rows {
                let hashed = hash64_seeded(r, &value);
                let col = ((hashed & LOWER_32_MASK) as usize) % cols;
                let bit = ((hashed >> 63) & 1) as i32;
                let sign_bit = -(1 - 2 * bit);
                let idx = r * cols + col;
                expected_once[idx] += sign_bit;
            }
        }
        // All counters should match the expected single-pass values.
        assert_eq!(storage.as_slice(), expected_once.as_slice());

        // Insert the same values again; counters should double.
        for i in 0..10 {
            sk.insert(&DataInput::I32(i));
        }
        let expected_twice: Vec<i32> = expected_once.iter().map(|v| v * 2).collect();
        assert_eq!(sk.as_storage().as_slice(), expected_twice.as_slice());

        // Estimates for inserted keys should be exactly 2.
        for i in 0..10 {
            let estimate = sk.estimate(&DataInput::I32(i));
            assert!(
                (estimate - 2.0).abs() < f64::EPSILON,
                "estimate for {i} should be 2.0, but get {estimate}"
            );
        }
    }

    #[test]
    fn cs_fast_path_correctness() {
        let mut sk = Count::<Vector2D<i32>, FastPath>::default();
        // Insert values 0..9 once using the fast path.
        for i in 0..10 {
            sk.insert(&DataInput::I32(i));
        }

        // Build the expected counter array by mirroring the fast-path hashing logic.
        let storage = sk.as_storage();
        let rows = storage.rows();
        let cols = storage.cols();
        let mask_bits = storage.get_mask_bits();
        let mask = (1u128 << mask_bits) - 1;
        let mut expected_once = vec![0_i32; rows * cols];

        for i in 0..10 {
            let value = DataInput::I32(i);
            let hash = <Vector2D<i32> as FastPathHasher<DefaultXxHasher>>::hash_for_matrix(
                storage, &value,
            );
            for row in 0..rows {
                let hashed = hash.row_hash(row, mask_bits, mask);
                let col = (hashed % cols as u128) as usize;
                let idx = row * cols + col;
                expected_once[idx] += hash.sign_for_row(row);
            }
        }

        assert_eq!(storage.as_slice(), expected_once.as_slice());
    }

    #[test]
    fn countsketch_error_stays_rank_independent_across_frequency_deciles() {
        const ROWS: usize = 5;
        const COLS: usize = 4096;
        const DOMAIN: usize = 8192;
        const SAMPLES: usize = 200_000;
        const STREAM_SEED: u64 = 1007;
        const DECILES: usize = 10;

        let mut truth = HashMap::<u64, i64>::new();
        let mut cs = Count::<Vector2D<i64>, RegularPath>::with_dimensions(ROWS, COLS);
        for value in sample_zipf_u64(DOMAIN, 1.1, SAMPLES, STREAM_SEED) {
            cs.insert(&DataInput::U64(value));
            *truth.entry(value).or_insert(0) += 1;
        }

        let mut pairs: Vec<(u64, i64)> = truth.into_iter().collect();
        pairs.sort_by_key(|(_, c)| *c);
        let per = pairs.len() / DECILES;
        let means: Vec<f64> = (0..DECILES)
            .map(|d| {
                let slice = &pairs[d * per..(d + 1) * per];
                slice
                    .iter()
                    .map(|(k, c)| (cs.estimate(&DataInput::U64(*k)) - *c as f64).abs())
                    .sum::<f64>()
                    / slice.len() as f64
            })
            .collect();
        let lo = means.iter().cloned().fold(f64::INFINITY, f64::min);
        let hi = means.iter().cloned().fold(0.0f64, f64::max);
        assert!(
            hi <= lo * 3.0,
            "Count Sketch mean |error| per frequency decile spans {lo:.1}..{hi:.1} ({:.2}x), \
             beyond the documented 3x empirical band — the error has started tracking key \
             frequency, which Count Sketch's L2 guarantee says it must not. \
             stream_seed={STREAM_SEED} rows={ROWS} cols={COLS} deciles={means:?}",
            hi / lo
        );
    }
}