datasketches 0.2.1 → 0.2.2
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +4 -4
- data/CHANGELOG.md +4 -0
- data/lib/datasketches/version.rb +1 -1
- data/vendor/datasketches-cpp/CMakeLists.txt +7 -0
- data/vendor/datasketches-cpp/common/include/MurmurHash3.h +11 -7
- data/vendor/datasketches-cpp/common/include/binomial_bounds.hpp +8 -8
- data/vendor/datasketches-cpp/common/include/bounds_binomial_proportions.hpp +12 -15
- data/vendor/datasketches-cpp/common/include/common_defs.hpp +24 -0
- data/vendor/datasketches-cpp/common/include/conditional_forward.hpp +20 -8
- data/vendor/datasketches-cpp/common/include/count_zeros.hpp +2 -2
- data/vendor/datasketches-cpp/common/include/serde.hpp +7 -7
- data/vendor/datasketches-cpp/cpc/include/cpc_compressor.hpp +19 -19
- data/vendor/datasketches-cpp/cpc/include/cpc_compressor_impl.hpp +91 -89
- data/vendor/datasketches-cpp/cpc/include/cpc_sketch.hpp +14 -1
- data/vendor/datasketches-cpp/cpc/include/cpc_sketch_impl.hpp +121 -87
- data/vendor/datasketches-cpp/cpc/include/cpc_union_impl.hpp +14 -14
- data/vendor/datasketches-cpp/cpc/include/cpc_util.hpp +10 -10
- data/vendor/datasketches-cpp/cpc/include/icon_estimator.hpp +4 -4
- data/vendor/datasketches-cpp/cpc/include/u32_table.hpp +8 -8
- data/vendor/datasketches-cpp/cpc/include/u32_table_impl.hpp +14 -14
- data/vendor/datasketches-cpp/cpc/test/compression_test.cpp +10 -10
- data/vendor/datasketches-cpp/cpc/test/cpc_sketch_test.cpp +25 -0
- data/vendor/datasketches-cpp/cpc/test/cpc_union_test.cpp +1 -1
- data/vendor/datasketches-cpp/fi/include/frequent_items_sketch_impl.hpp +65 -80
- data/vendor/datasketches-cpp/fi/include/reverse_purge_hash_map_impl.hpp +10 -10
- data/vendor/datasketches-cpp/fi/test/reverse_purge_hash_map_test.cpp +2 -2
- data/vendor/datasketches-cpp/hll/include/AuxHashMap-internal.hpp +60 -63
- data/vendor/datasketches-cpp/hll/include/AuxHashMap.hpp +19 -19
- data/vendor/datasketches-cpp/hll/include/CompositeInterpolationXTable-internal.hpp +15 -15
- data/vendor/datasketches-cpp/hll/include/CompositeInterpolationXTable.hpp +3 -3
- data/vendor/datasketches-cpp/hll/include/CouponHashSet-internal.hpp +74 -76
- data/vendor/datasketches-cpp/hll/include/CouponHashSet.hpp +6 -6
- data/vendor/datasketches-cpp/hll/include/CouponList-internal.hpp +110 -113
- data/vendor/datasketches-cpp/hll/include/CouponList.hpp +13 -13
- data/vendor/datasketches-cpp/hll/include/CubicInterpolation-internal.hpp +2 -4
- data/vendor/datasketches-cpp/hll/include/HarmonicNumbers-internal.hpp +1 -1
- data/vendor/datasketches-cpp/hll/include/Hll4Array-internal.hpp +80 -76
- data/vendor/datasketches-cpp/hll/include/Hll4Array.hpp +9 -9
- data/vendor/datasketches-cpp/hll/include/Hll6Array-internal.hpp +26 -26
- data/vendor/datasketches-cpp/hll/include/Hll6Array.hpp +6 -6
- data/vendor/datasketches-cpp/hll/include/Hll8Array-internal.hpp +33 -33
- data/vendor/datasketches-cpp/hll/include/Hll8Array.hpp +6 -6
- data/vendor/datasketches-cpp/hll/include/HllArray-internal.hpp +205 -209
- data/vendor/datasketches-cpp/hll/include/HllArray.hpp +36 -36
- data/vendor/datasketches-cpp/hll/include/HllSketch-internal.hpp +28 -28
- data/vendor/datasketches-cpp/hll/include/HllSketchImpl-internal.hpp +22 -22
- data/vendor/datasketches-cpp/hll/include/HllSketchImpl.hpp +13 -13
- data/vendor/datasketches-cpp/hll/include/HllSketchImplFactory.hpp +15 -15
- data/vendor/datasketches-cpp/hll/include/HllUnion-internal.hpp +61 -61
- data/vendor/datasketches-cpp/hll/include/HllUtil.hpp +120 -127
- data/vendor/datasketches-cpp/hll/include/coupon_iterator-internal.hpp +9 -9
- data/vendor/datasketches-cpp/hll/include/coupon_iterator.hpp +5 -5
- data/vendor/datasketches-cpp/hll/include/hll.hpp +21 -21
- data/vendor/datasketches-cpp/hll/test/AuxHashMapTest.cpp +1 -1
- data/vendor/datasketches-cpp/hll/test/CouponHashSetTest.cpp +34 -34
- data/vendor/datasketches-cpp/hll/test/CouponListTest.cpp +25 -25
- data/vendor/datasketches-cpp/hll/test/CrossCountingTest.cpp +2 -2
- data/vendor/datasketches-cpp/hll/test/HllArrayTest.cpp +35 -35
- data/vendor/datasketches-cpp/hll/test/HllSketchTest.cpp +15 -15
- data/vendor/datasketches-cpp/hll/test/HllUnionTest.cpp +10 -14
- data/vendor/datasketches-cpp/hll/test/IsomorphicTest.cpp +3 -3
- data/vendor/datasketches-cpp/hll/test/ToFromByteArrayTest.cpp +4 -4
- data/vendor/datasketches-cpp/kll/include/kll_helper.hpp +5 -4
- data/vendor/datasketches-cpp/kll/include/kll_helper_impl.hpp +6 -6
- data/vendor/datasketches-cpp/kll/include/kll_quantile_calculator.hpp +14 -6
- data/vendor/datasketches-cpp/kll/include/kll_quantile_calculator_impl.hpp +39 -24
- data/vendor/datasketches-cpp/kll/include/kll_sketch.hpp +34 -2
- data/vendor/datasketches-cpp/kll/include/kll_sketch_impl.hpp +72 -62
- data/vendor/datasketches-cpp/kll/include/kolmogorov_smirnov.hpp +67 -0
- data/vendor/datasketches-cpp/kll/include/kolmogorov_smirnov_impl.hpp +78 -0
- data/vendor/datasketches-cpp/kll/test/CMakeLists.txt +1 -0
- data/vendor/datasketches-cpp/kll/test/kll_sketch_test.cpp +68 -45
- data/vendor/datasketches-cpp/kll/test/kolmogorov_smirnov_test.cpp +111 -0
- data/vendor/datasketches-cpp/python/src/hll_wrapper.cpp +4 -4
- data/vendor/datasketches-cpp/python/src/theta_wrapper.cpp +6 -6
- data/vendor/datasketches-cpp/python/src/vo_wrapper.cpp +2 -2
- data/vendor/datasketches-cpp/python/tests/hll_test.py +1 -1
- data/vendor/datasketches-cpp/python/tests/vo_test.py +3 -3
- data/vendor/datasketches-cpp/req/include/req_common.hpp +2 -1
- data/vendor/datasketches-cpp/req/include/req_compactor.hpp +4 -4
- data/vendor/datasketches-cpp/req/include/req_compactor_impl.hpp +26 -39
- data/vendor/datasketches-cpp/req/include/req_sketch.hpp +1 -1
- data/vendor/datasketches-cpp/req/include/req_sketch_impl.hpp +9 -9
- data/vendor/datasketches-cpp/req/test/req_sketch_test.cpp +52 -52
- data/vendor/datasketches-cpp/sampling/include/var_opt_sketch_impl.hpp +47 -56
- data/vendor/datasketches-cpp/sampling/include/var_opt_union_impl.hpp +34 -42
- data/vendor/datasketches-cpp/sampling/test/var_opt_sketch_test.cpp +6 -6
- data/vendor/datasketches-cpp/sampling/test/var_opt_union_test.cpp +13 -13
- data/vendor/datasketches-cpp/setup.py +1 -1
- data/vendor/datasketches-cpp/theta/include/bounds_on_ratios_in_sampled_sets.hpp +1 -1
- data/vendor/datasketches-cpp/theta/include/compact_theta_sketch_parser.hpp +67 -0
- data/vendor/datasketches-cpp/theta/include/compact_theta_sketch_parser_impl.hpp +70 -0
- data/vendor/datasketches-cpp/theta/include/theta_intersection.hpp +9 -4
- data/vendor/datasketches-cpp/theta/include/theta_intersection_impl.hpp +1 -1
- data/vendor/datasketches-cpp/theta/include/theta_jaccard_similarity_base.hpp +18 -14
- data/vendor/datasketches-cpp/theta/include/theta_sketch.hpp +42 -1
- data/vendor/datasketches-cpp/theta/include/theta_sketch_impl.hpp +107 -58
- data/vendor/datasketches-cpp/theta/include/theta_union.hpp +4 -4
- data/vendor/datasketches-cpp/theta/include/theta_union_base_impl.hpp +1 -1
- data/vendor/datasketches-cpp/theta/include/theta_union_impl.hpp +1 -1
- data/vendor/datasketches-cpp/theta/include/theta_update_sketch_base.hpp +2 -0
- data/vendor/datasketches-cpp/theta/include/theta_update_sketch_base_impl.hpp +33 -28
- data/vendor/datasketches-cpp/theta/test/theta_a_not_b_test.cpp +23 -1
- data/vendor/datasketches-cpp/theta/test/theta_intersection_test.cpp +21 -1
- data/vendor/datasketches-cpp/theta/test/theta_jaccard_similarity_test.cpp +58 -2
- data/vendor/datasketches-cpp/theta/test/theta_sketch_test.cpp +37 -1
- data/vendor/datasketches-cpp/theta/test/theta_union_test.cpp +22 -2
- data/vendor/datasketches-cpp/tuple/include/array_of_doubles_sketch_impl.hpp +47 -60
- data/vendor/datasketches-cpp/tuple/include/tuple_sketch_impl.hpp +51 -64
- data/vendor/datasketches-cpp/tuple/test/array_of_doubles_sketch_test.cpp +1 -1
- data/vendor/datasketches-cpp/tuple/test/tuple_a_not_b_test.cpp +17 -17
- data/vendor/datasketches-cpp/tuple/test/tuple_intersection_test.cpp +12 -12
- data/vendor/datasketches-cpp/tuple/test/tuple_jaccard_similarity_test.cpp +5 -5
- data/vendor/datasketches-cpp/tuple/test/tuple_sketch_allocation_test.cpp +1 -1
- data/vendor/datasketches-cpp/tuple/test/tuple_sketch_test.cpp +20 -20
- data/vendor/datasketches-cpp/tuple/test/tuple_union_test.cpp +12 -12
- metadata +8 -3
@@ -192,7 +192,7 @@ public:
|
|
192
192
|
* @param data pointer to the data
|
193
193
|
* @param length of the data in bytes
|
194
194
|
*/
|
195
|
-
void update(const void* value,
|
195
|
+
void update(const void* value, size_t size);
|
196
196
|
|
197
197
|
/**
|
198
198
|
* Returns a human-readable summary of this sketch
|
@@ -235,6 +235,17 @@ public:
|
|
235
235
|
*/
|
236
236
|
static cpc_sketch_alloc<A> deserialize(const void* bytes, size_t size, uint64_t seed = DEFAULT_SEED, const A& allocator = A());
|
237
237
|
|
238
|
+
/**
|
239
|
+
* The actual size of a compressed CPC sketch has a small random variance, but the following
|
240
|
+
* empirically measured size should be large enough for at least 99.9 percent of sketches.
|
241
|
+
*
|
242
|
+
* <p>For small values of <i>n</i> the size can be much smaller.
|
243
|
+
*
|
244
|
+
* @param lg_k the given value of lg_k.
|
245
|
+
* @return the estimated maximum compressed serialized size of a sketch.
|
246
|
+
*/
|
247
|
+
static size_t get_max_serialized_size_bytes(uint8_t lg_k);
|
248
|
+
|
238
249
|
// for internal use
|
239
250
|
uint32_t get_num_coupons() const;
|
240
251
|
|
@@ -303,6 +314,8 @@ private:
|
|
303
314
|
inline void write_hip(std::ostream& os) const;
|
304
315
|
inline size_t copy_hip_to_mem(void* dst) const;
|
305
316
|
|
317
|
+
static void check_lg_k(uint8_t lg_k);
|
318
|
+
|
306
319
|
friend cpc_compressor<A>;
|
307
320
|
friend cpc_union_alloc<A>;
|
308
321
|
};
|
@@ -53,9 +53,7 @@ first_interesting_column(0),
|
|
53
53
|
kxp(1 << lg_k),
|
54
54
|
hip_est_accum(0)
|
55
55
|
{
|
56
|
-
|
57
|
-
throw std::invalid_argument("lg_k must be >= " + std::to_string(CPC_MIN_LG_K) + " and <= " + std::to_string(CPC_MAX_LG_K) + ": " + std::to_string(lg_k));
|
58
|
-
}
|
56
|
+
check_lg_k(lg_k);
|
59
57
|
}
|
60
58
|
|
61
59
|
template<typename A>
|
@@ -176,7 +174,7 @@ void cpc_sketch_alloc<A>::update(float value) {
|
|
176
174
|
|
177
175
|
static inline uint32_t row_col_from_two_hashes(uint64_t hash0, uint64_t hash1, uint8_t lg_k) {
|
178
176
|
if (lg_k > 26) throw std::logic_error("lg_k > 26");
|
179
|
-
const
|
177
|
+
const uint32_t k = 1 << lg_k;
|
180
178
|
uint8_t col = count_leading_zeros_in_u64(hash1); // 0 <= col <= 64
|
181
179
|
if (col > 63) col = 63; // clip so that 0 <= col <= 63
|
182
180
|
const uint32_t row = hash0 & (k - 1);
|
@@ -188,7 +186,7 @@ static inline uint32_t row_col_from_two_hashes(uint64_t hash0, uint64_t hash1, u
|
|
188
186
|
}
|
189
187
|
|
190
188
|
template<typename A>
|
191
|
-
void cpc_sketch_alloc<A>::update(const void* value,
|
189
|
+
void cpc_sketch_alloc<A>::update(const void* value, size_t size) {
|
192
190
|
HashState hashes;
|
193
191
|
MurmurHash3_x64_128(value, size, seed, hashes);
|
194
192
|
row_col_update(row_col_from_two_hashes(hashes.h1, hashes.h2, lg_k));
|
@@ -208,7 +206,7 @@ void cpc_sketch_alloc<A>::row_col_update(uint32_t row_col) {
|
|
208
206
|
|
209
207
|
template<typename A>
|
210
208
|
void cpc_sketch_alloc<A>::update_sparse(uint32_t row_col) {
|
211
|
-
const
|
209
|
+
const uint32_t k = 1 << lg_k;
|
212
210
|
const uint64_t c32pre = static_cast<uint64_t>(num_coupons) << 5;
|
213
211
|
if (c32pre >= 3 * k) throw std::logic_error("c32pre >= 3 * k"); // C < 3K/32, in other words flavor == SPARSE
|
214
212
|
bool is_novel = surprising_value_table.maybe_insert(row_col);
|
@@ -224,7 +222,7 @@ void cpc_sketch_alloc<A>::update_sparse(uint32_t row_col) {
|
|
224
222
|
template<typename A>
|
225
223
|
void cpc_sketch_alloc<A>::update_windowed(uint32_t row_col) {
|
226
224
|
if (window_offset > 56) throw std::logic_error("wrong window offset");
|
227
|
-
const
|
225
|
+
const uint32_t k = 1 << lg_k;
|
228
226
|
const uint64_t c32pre = static_cast<uint64_t>(num_coupons) << 5;
|
229
227
|
if (c32pre < 3 * k) throw std::logic_error("c32pre < 3 * k"); // C < 3K/32, in other words flavor >= HYBRID
|
230
228
|
const uint64_t c8pre = static_cast<uint64_t>(num_coupons) << 3;
|
@@ -266,7 +264,7 @@ void cpc_sketch_alloc<A>::update_windowed(uint32_t row_col) {
|
|
266
264
|
// Call this whenever a new coupon has been collected.
|
267
265
|
template<typename A>
|
268
266
|
void cpc_sketch_alloc<A>::update_hip(uint32_t row_col) {
|
269
|
-
const
|
267
|
+
const uint32_t k = 1 << lg_k;
|
270
268
|
const uint8_t col = row_col & 63;
|
271
269
|
const double one_over_p = static_cast<double>(k) / kxp;
|
272
270
|
hip_est_accum += one_over_p;
|
@@ -276,7 +274,7 @@ void cpc_sketch_alloc<A>::update_hip(uint32_t row_col) {
|
|
276
274
|
// In terms of flavor, this promotes SPARSE to HYBRID
|
277
275
|
template<typename A>
|
278
276
|
void cpc_sketch_alloc<A>::promote_sparse_to_windowed() {
|
279
|
-
const
|
277
|
+
const uint32_t k = 1 << lg_k;
|
280
278
|
const uint64_t c32 = static_cast<uint64_t>(num_coupons) << 5;
|
281
279
|
if (!(c32 == 3 * k || (lg_k == 4 && c32 > 3 * k))) throw std::logic_error("wrong c32");
|
282
280
|
|
@@ -285,16 +283,16 @@ void cpc_sketch_alloc<A>::promote_sparse_to_windowed() {
|
|
285
283
|
u32_table<A> new_table(2, 6 + lg_k, sliding_window.get_allocator());
|
286
284
|
|
287
285
|
const uint32_t* old_slots = surprising_value_table.get_slots();
|
288
|
-
const
|
286
|
+
const uint32_t old_num_slots = 1 << surprising_value_table.get_lg_size();
|
289
287
|
|
290
288
|
if (window_offset != 0) throw std::logic_error("window_offset != 0");
|
291
289
|
|
292
|
-
for (
|
290
|
+
for (uint32_t i = 0; i < old_num_slots; i++) {
|
293
291
|
const uint32_t row_col = old_slots[i];
|
294
292
|
if (row_col != UINT32_MAX) {
|
295
293
|
const uint8_t col = row_col & 63;
|
296
294
|
if (col < 8) {
|
297
|
-
const
|
295
|
+
const uint32_t row = row_col >> 6;
|
298
296
|
sliding_window[row] |= 1 << col;
|
299
297
|
} else {
|
300
298
|
// cannot use u32_table::must_insert(), because it doesn't provide for growth
|
@@ -314,7 +312,7 @@ void cpc_sketch_alloc<A>::move_window() {
|
|
314
312
|
if (new_offset != determine_correct_offset(lg_k, num_coupons)) throw std::logic_error("new_offset is wrong");
|
315
313
|
|
316
314
|
if (sliding_window.size() == 0) throw std::logic_error("no sliding window");
|
317
|
-
const
|
315
|
+
const uint32_t k = 1 << lg_k;
|
318
316
|
|
319
317
|
// Construct the full-sized bit matrix that corresponds to the sketch
|
320
318
|
vector_u64<A> bit_matrix = build_bit_matrix();
|
@@ -328,7 +326,7 @@ void cpc_sketch_alloc<A>::move_window() {
|
|
328
326
|
const uint64_t mask_for_flipping_early_zone = (static_cast<uint64_t>(1) << new_offset) - 1;
|
329
327
|
uint64_t all_surprises_ored = 0;
|
330
328
|
|
331
|
-
for (
|
329
|
+
for (uint32_t i = 0; i < k; i++) {
|
332
330
|
uint64_t pattern = bit_matrix[i];
|
333
331
|
sliding_window[i] = (pattern >> new_offset) & 0xff;
|
334
332
|
pattern &= mask_for_clearing_window;
|
@@ -357,7 +355,7 @@ void cpc_sketch_alloc<A>::move_window() {
|
|
357
355
|
// so that it will reflect changes that were previously outside the mantissa.
|
358
356
|
template<typename A>
|
359
357
|
void cpc_sketch_alloc<A>::refresh_kxp(const uint64_t* bit_matrix) {
|
360
|
-
const
|
358
|
+
const uint32_t k = 1 << lg_k;
|
361
359
|
|
362
360
|
// for improved numerical accuracy, we separately sum the bytes of the U64's
|
363
361
|
double byte_sums[8]; // allocating on the stack
|
@@ -415,44 +413,44 @@ void cpc_sketch_alloc<A>::serialize(std::ostream& os) const {
|
|
415
413
|
const bool has_table = compressed.table_data.size() > 0;
|
416
414
|
const bool has_window = compressed.window_data.size() > 0;
|
417
415
|
const uint8_t preamble_ints = get_preamble_ints(num_coupons, has_hip, has_table, has_window);
|
418
|
-
|
416
|
+
write(os, preamble_ints);
|
419
417
|
const uint8_t serial_version = SERIAL_VERSION;
|
420
|
-
|
418
|
+
write(os, serial_version);
|
421
419
|
const uint8_t family = FAMILY;
|
422
|
-
|
423
|
-
|
424
|
-
|
420
|
+
write(os, family);
|
421
|
+
write(os, lg_k);
|
422
|
+
write(os, first_interesting_column);
|
425
423
|
const uint8_t flags_byte(
|
426
424
|
(1 << flags::IS_COMPRESSED)
|
427
425
|
| (has_hip ? 1 << flags::HAS_HIP : 0)
|
428
426
|
| (has_table ? 1 << flags::HAS_TABLE : 0)
|
429
427
|
| (has_window ? 1 << flags::HAS_WINDOW : 0)
|
430
428
|
);
|
431
|
-
|
429
|
+
write(os, flags_byte);
|
432
430
|
const uint16_t seed_hash(compute_seed_hash(seed));
|
433
|
-
|
431
|
+
write(os, seed_hash);
|
434
432
|
if (!is_empty()) {
|
435
|
-
|
433
|
+
write(os, num_coupons);
|
436
434
|
if (has_table && has_window) {
|
437
435
|
// if there is no window it is the same as number of coupons
|
438
|
-
|
436
|
+
write(os, compressed.table_num_entries);
|
439
437
|
// HIP values can be in two different places in the sequence of fields
|
440
438
|
// this is the first HIP decision point
|
441
439
|
if (has_hip) write_hip(os);
|
442
440
|
}
|
443
441
|
if (has_table) {
|
444
|
-
|
442
|
+
write(os, compressed.table_data_words);
|
445
443
|
}
|
446
444
|
if (has_window) {
|
447
|
-
|
445
|
+
write(os, compressed.window_data_words);
|
448
446
|
}
|
449
447
|
// this is the second HIP decision point
|
450
448
|
if (has_hip && !(has_table && has_window)) write_hip(os);
|
451
449
|
if (has_window) {
|
452
|
-
|
450
|
+
write(os, compressed.window_data.data(), compressed.window_data_words * sizeof(uint32_t));
|
453
451
|
}
|
454
452
|
if (has_table) {
|
455
|
-
|
453
|
+
write(os, compressed.table_data.data(), compressed.table_data_words * sizeof(uint32_t));
|
456
454
|
}
|
457
455
|
}
|
458
456
|
}
|
@@ -471,36 +469,36 @@ vector_u8<A> cpc_sketch_alloc<A>::serialize(unsigned header_size_bytes) const {
|
|
471
469
|
const size_t size = header_size_bytes + (preamble_ints + compressed.table_data_words + compressed.window_data_words) * sizeof(uint32_t);
|
472
470
|
vector_u8<A> bytes(size, 0, sliding_window.get_allocator());
|
473
471
|
uint8_t* ptr = bytes.data() + header_size_bytes;
|
474
|
-
ptr += copy_to_mem(
|
472
|
+
ptr += copy_to_mem(preamble_ints, ptr);
|
475
473
|
const uint8_t serial_version = SERIAL_VERSION;
|
476
|
-
ptr += copy_to_mem(
|
474
|
+
ptr += copy_to_mem(serial_version, ptr);
|
477
475
|
const uint8_t family = FAMILY;
|
478
|
-
ptr += copy_to_mem(
|
479
|
-
ptr += copy_to_mem(
|
480
|
-
ptr += copy_to_mem(
|
476
|
+
ptr += copy_to_mem(family, ptr);
|
477
|
+
ptr += copy_to_mem(lg_k, ptr);
|
478
|
+
ptr += copy_to_mem(first_interesting_column, ptr);
|
481
479
|
const uint8_t flags_byte(
|
482
480
|
(1 << flags::IS_COMPRESSED)
|
483
481
|
| (has_hip ? 1 << flags::HAS_HIP : 0)
|
484
482
|
| (has_table ? 1 << flags::HAS_TABLE : 0)
|
485
483
|
| (has_window ? 1 << flags::HAS_WINDOW : 0)
|
486
484
|
);
|
487
|
-
ptr += copy_to_mem(
|
485
|
+
ptr += copy_to_mem(flags_byte, ptr);
|
488
486
|
const uint16_t seed_hash = compute_seed_hash(seed);
|
489
|
-
ptr += copy_to_mem(
|
487
|
+
ptr += copy_to_mem(seed_hash, ptr);
|
490
488
|
if (!is_empty()) {
|
491
|
-
ptr += copy_to_mem(
|
489
|
+
ptr += copy_to_mem(num_coupons, ptr);
|
492
490
|
if (has_table && has_window) {
|
493
491
|
// if there is no window it is the same as number of coupons
|
494
|
-
ptr += copy_to_mem(
|
492
|
+
ptr += copy_to_mem(compressed.table_num_entries, ptr);
|
495
493
|
// HIP values can be in two different places in the sequence of fields
|
496
494
|
// this is the first HIP decision point
|
497
495
|
if (has_hip) ptr += copy_hip_to_mem(ptr);
|
498
496
|
}
|
499
497
|
if (has_table) {
|
500
|
-
ptr += copy_to_mem(
|
498
|
+
ptr += copy_to_mem(compressed.table_data_words, ptr);
|
501
499
|
}
|
502
500
|
if (has_window) {
|
503
|
-
ptr += copy_to_mem(
|
501
|
+
ptr += copy_to_mem(compressed.window_data_words, ptr);
|
504
502
|
}
|
505
503
|
// this is the second HIP decision point
|
506
504
|
if (has_hip && !(has_table && has_window)) ptr += copy_hip_to_mem(ptr);
|
@@ -517,20 +515,13 @@ vector_u8<A> cpc_sketch_alloc<A>::serialize(unsigned header_size_bytes) const {
|
|
517
515
|
|
518
516
|
template<typename A>
|
519
517
|
cpc_sketch_alloc<A> cpc_sketch_alloc<A>::deserialize(std::istream& is, uint64_t seed, const A& allocator) {
|
520
|
-
|
521
|
-
|
522
|
-
uint8_t
|
523
|
-
|
524
|
-
uint8_t
|
525
|
-
|
526
|
-
|
527
|
-
is.read((char*)&lg_k, sizeof(lg_k));
|
528
|
-
uint8_t first_interesting_column;
|
529
|
-
is.read((char*)&first_interesting_column, sizeof(first_interesting_column));
|
530
|
-
uint8_t flags_byte;
|
531
|
-
is.read((char*)&flags_byte, sizeof(flags_byte));
|
532
|
-
uint16_t seed_hash;
|
533
|
-
is.read((char*)&seed_hash, sizeof(seed_hash));
|
518
|
+
const auto preamble_ints = read<uint8_t>(is);
|
519
|
+
const auto serial_version = read<uint8_t>(is);
|
520
|
+
const auto family_id = read<uint8_t>(is);
|
521
|
+
const auto lg_k = read<uint8_t>(is);
|
522
|
+
const auto first_interesting_column = read<uint8_t>(is);
|
523
|
+
const auto flags_byte = read<uint8_t>(is);
|
524
|
+
const auto seed_hash = read<uint16_t>(is);
|
534
525
|
const bool has_hip = flags_byte & (1 << flags::HAS_HIP);
|
535
526
|
const bool has_table = flags_byte & (1 << flags::HAS_TABLE);
|
536
527
|
const bool has_window = flags_byte & (1 << flags::HAS_WINDOW);
|
@@ -542,31 +533,31 @@ cpc_sketch_alloc<A> cpc_sketch_alloc<A>::deserialize(std::istream& is, uint64_t
|
|
542
533
|
double kxp = 0;
|
543
534
|
double hip_est_accum = 0;
|
544
535
|
if (has_table || has_window) {
|
545
|
-
|
536
|
+
num_coupons = read<uint32_t>(is);
|
546
537
|
if (has_table && has_window) {
|
547
|
-
|
538
|
+
compressed.table_num_entries = read<uint32_t>(is);
|
548
539
|
if (has_hip) {
|
549
|
-
|
550
|
-
|
540
|
+
kxp = read<double>(is);
|
541
|
+
hip_est_accum = read<double>(is);
|
551
542
|
}
|
552
543
|
}
|
553
544
|
if (has_table) {
|
554
|
-
|
545
|
+
compressed.table_data_words = read<uint32_t>(is);
|
555
546
|
}
|
556
547
|
if (has_window) {
|
557
|
-
|
548
|
+
compressed.window_data_words = read<uint32_t>(is);
|
558
549
|
}
|
559
550
|
if (has_hip && !(has_table && has_window)) {
|
560
|
-
|
561
|
-
|
551
|
+
kxp = read<double>(is);
|
552
|
+
hip_est_accum = read<double>(is);
|
562
553
|
}
|
563
554
|
if (has_window) {
|
564
555
|
compressed.window_data.resize(compressed.window_data_words);
|
565
|
-
|
556
|
+
read(is, compressed.window_data.data(), compressed.window_data_words * sizeof(uint32_t));
|
566
557
|
}
|
567
558
|
if (has_table) {
|
568
559
|
compressed.table_data.resize(compressed.table_data_words);
|
569
|
-
|
560
|
+
read(is, compressed.table_data.data(), compressed.table_data_words * sizeof(uint32_t));
|
570
561
|
}
|
571
562
|
if (!has_window) compressed.table_num_entries = num_coupons;
|
572
563
|
}
|
@@ -602,19 +593,19 @@ cpc_sketch_alloc<A> cpc_sketch_alloc<A>::deserialize(const void* bytes, size_t s
|
|
602
593
|
const char* ptr = static_cast<const char*>(bytes);
|
603
594
|
const char* base = static_cast<const char*>(bytes);
|
604
595
|
uint8_t preamble_ints;
|
605
|
-
ptr += copy_from_mem(ptr,
|
596
|
+
ptr += copy_from_mem(ptr, preamble_ints);
|
606
597
|
uint8_t serial_version;
|
607
|
-
ptr += copy_from_mem(ptr,
|
598
|
+
ptr += copy_from_mem(ptr, serial_version);
|
608
599
|
uint8_t family_id;
|
609
|
-
ptr += copy_from_mem(ptr,
|
600
|
+
ptr += copy_from_mem(ptr, family_id);
|
610
601
|
uint8_t lg_k;
|
611
|
-
ptr += copy_from_mem(ptr,
|
602
|
+
ptr += copy_from_mem(ptr, lg_k);
|
612
603
|
uint8_t first_interesting_column;
|
613
|
-
ptr += copy_from_mem(ptr,
|
604
|
+
ptr += copy_from_mem(ptr, first_interesting_column);
|
614
605
|
uint8_t flags_byte;
|
615
|
-
ptr += copy_from_mem(ptr,
|
606
|
+
ptr += copy_from_mem(ptr, flags_byte);
|
616
607
|
uint16_t seed_hash;
|
617
|
-
ptr += copy_from_mem(ptr,
|
608
|
+
ptr += copy_from_mem(ptr, seed_hash);
|
618
609
|
const bool has_hip = flags_byte & (1 << flags::HAS_HIP);
|
619
610
|
const bool has_table = flags_byte & (1 << flags::HAS_TABLE);
|
620
611
|
const bool has_window = flags_byte & (1 << flags::HAS_WINDOW);
|
@@ -628,28 +619,28 @@ cpc_sketch_alloc<A> cpc_sketch_alloc<A>::deserialize(const void* bytes, size_t s
|
|
628
619
|
double hip_est_accum = 0;
|
629
620
|
if (has_table || has_window) {
|
630
621
|
check_memory_size(ptr - base + sizeof(num_coupons), size);
|
631
|
-
ptr += copy_from_mem(ptr,
|
622
|
+
ptr += copy_from_mem(ptr, num_coupons);
|
632
623
|
if (has_table && has_window) {
|
633
624
|
check_memory_size(ptr - base + sizeof(compressed.table_num_entries), size);
|
634
|
-
ptr += copy_from_mem(ptr,
|
625
|
+
ptr += copy_from_mem(ptr, compressed.table_num_entries);
|
635
626
|
if (has_hip) {
|
636
627
|
check_memory_size(ptr - base + sizeof(kxp) + sizeof(hip_est_accum), size);
|
637
|
-
ptr += copy_from_mem(ptr,
|
638
|
-
ptr += copy_from_mem(ptr,
|
628
|
+
ptr += copy_from_mem(ptr, kxp);
|
629
|
+
ptr += copy_from_mem(ptr, hip_est_accum);
|
639
630
|
}
|
640
631
|
}
|
641
632
|
if (has_table) {
|
642
633
|
check_memory_size(ptr - base + sizeof(compressed.table_data_words), size);
|
643
|
-
ptr += copy_from_mem(ptr,
|
634
|
+
ptr += copy_from_mem(ptr, compressed.table_data_words);
|
644
635
|
}
|
645
636
|
if (has_window) {
|
646
637
|
check_memory_size(ptr - base + sizeof(compressed.window_data_words), size);
|
647
|
-
ptr += copy_from_mem(ptr,
|
638
|
+
ptr += copy_from_mem(ptr, compressed.window_data_words);
|
648
639
|
}
|
649
640
|
if (has_hip && !(has_table && has_window)) {
|
650
641
|
check_memory_size(ptr - base + sizeof(kxp) + sizeof(hip_est_accum), size);
|
651
|
-
ptr += copy_from_mem(ptr,
|
652
|
-
ptr += copy_from_mem(ptr,
|
642
|
+
ptr += copy_from_mem(ptr, kxp);
|
643
|
+
ptr += copy_from_mem(ptr, hip_est_accum);
|
653
644
|
}
|
654
645
|
if (has_window) {
|
655
646
|
compressed.window_data.resize(compressed.window_data_words);
|
@@ -688,6 +679,49 @@ cpc_sketch_alloc<A> cpc_sketch_alloc<A>::deserialize(const void* bytes, size_t s
|
|
688
679
|
std::move(uncompressed.window), has_hip, kxp, hip_est_accum, seed);
|
689
680
|
}
|
690
681
|
|
682
|
+
/*
|
683
|
+
* These empirical values for the 99.9th percentile of size in bytes were measured using 100,000
|
684
|
+
* trials. The value for each trial is the maximum of 5*16=80 measurements that were equally
|
685
|
+
* spaced over values of the quantity C/K between 3.0 and 8.0. This table does not include the
|
686
|
+
* worst-case space for the preamble, which is added by the function.
|
687
|
+
*/
|
688
|
+
static const uint8_t CPC_EMPIRICAL_SIZE_MAX_LGK = 19;
|
689
|
+
static const size_t CPC_EMPIRICAL_MAX_SIZE_BYTES[] = {
|
690
|
+
24, // lg_k = 4
|
691
|
+
36, // lg_k = 5
|
692
|
+
56, // lg_k = 6
|
693
|
+
100, // lg_k = 7
|
694
|
+
180, // lg_k = 8
|
695
|
+
344, // lg_k = 9
|
696
|
+
660, // lg_k = 10
|
697
|
+
1292, // lg_k = 11
|
698
|
+
2540, // lg_k = 12
|
699
|
+
5020, // lg_k = 13
|
700
|
+
9968, // lg_k = 14
|
701
|
+
19836, // lg_k = 15
|
702
|
+
39532, // lg_k = 16
|
703
|
+
78880, // lg_k = 17
|
704
|
+
157516, // lg_k = 18
|
705
|
+
314656 // lg_k = 19
|
706
|
+
};
|
707
|
+
static const double CPC_EMPIRICAL_MAX_SIZE_FACTOR = 0.6; // 0.6 = 4.8 / 8.0
|
708
|
+
static const size_t CPC_MAX_PREAMBLE_SIZE_BYTES = 40;
|
709
|
+
|
710
|
+
template<typename A>
|
711
|
+
size_t cpc_sketch_alloc<A>::get_max_serialized_size_bytes(uint8_t lg_k) {
|
712
|
+
check_lg_k(lg_k);
|
713
|
+
if (lg_k <= CPC_EMPIRICAL_SIZE_MAX_LGK) return CPC_EMPIRICAL_MAX_SIZE_BYTES[lg_k - CPC_MIN_LG_K] + CPC_MAX_PREAMBLE_SIZE_BYTES;
|
714
|
+
const uint32_t k = 1 << lg_k;
|
715
|
+
return (int) (CPC_EMPIRICAL_MAX_SIZE_FACTOR * k) + CPC_MAX_PREAMBLE_SIZE_BYTES;
|
716
|
+
}
|
717
|
+
|
718
|
+
template<typename A>
|
719
|
+
void cpc_sketch_alloc<A>::check_lg_k(uint8_t lg_k) {
|
720
|
+
if (lg_k < CPC_MIN_LG_K || lg_k > CPC_MAX_LG_K) {
|
721
|
+
throw std::invalid_argument("lg_k must be >= " + std::to_string(CPC_MIN_LG_K) + " and <= " + std::to_string(CPC_MAX_LG_K) + ": " + std::to_string(lg_k));
|
722
|
+
}
|
723
|
+
}
|
724
|
+
|
691
725
|
template<typename A>
|
692
726
|
uint32_t cpc_sketch_alloc<A>::get_num_coupons() const {
|
693
727
|
return num_coupons;
|
@@ -696,7 +730,7 @@ uint32_t cpc_sketch_alloc<A>::get_num_coupons() const {
|
|
696
730
|
template<typename A>
|
697
731
|
bool cpc_sketch_alloc<A>::validate() const {
|
698
732
|
vector_u64<A> bit_matrix = build_bit_matrix();
|
699
|
-
const uint64_t num_bits_set = count_bits_set_in_matrix(bit_matrix.data(),
|
733
|
+
const uint64_t num_bits_set = count_bits_set_in_matrix(bit_matrix.data(), 1ULL << lg_k);
|
700
734
|
return num_bits_set == num_coupons;
|
701
735
|
}
|
702
736
|
|
@@ -744,7 +778,7 @@ typename cpc_sketch_alloc<A>::flavor cpc_sketch_alloc<A>::determine_flavor() con
|
|
744
778
|
|
745
779
|
template<typename A>
|
746
780
|
typename cpc_sketch_alloc<A>::flavor cpc_sketch_alloc<A>::determine_flavor(uint8_t lg_k, uint64_t c) {
|
747
|
-
const
|
781
|
+
const uint32_t k = 1 << lg_k;
|
748
782
|
const uint64_t c2 = c << 1;
|
749
783
|
const uint64_t c8 = c << 3;
|
750
784
|
const uint64_t c32 = c << 5;
|
@@ -757,15 +791,15 @@ typename cpc_sketch_alloc<A>::flavor cpc_sketch_alloc<A>::determine_flavor(uint8
|
|
757
791
|
|
758
792
|
template<typename A>
|
759
793
|
uint8_t cpc_sketch_alloc<A>::determine_correct_offset(uint8_t lg_k, uint64_t c) {
|
760
|
-
const
|
794
|
+
const uint32_t k = 1 << lg_k;
|
761
795
|
const int64_t tmp = static_cast<int64_t>(c << 3) - static_cast<int64_t>(19 * k); // 8C - 19K
|
762
796
|
if (tmp < 0) return 0;
|
763
|
-
return tmp >> (lg_k + 3); // tmp / 8K
|
797
|
+
return static_cast<uint8_t>(tmp >> (lg_k + 3)); // tmp / 8K
|
764
798
|
}
|
765
799
|
|
766
800
|
template<typename A>
|
767
801
|
vector_u64<A> cpc_sketch_alloc<A>::build_bit_matrix() const {
|
768
|
-
const
|
802
|
+
const uint32_t k = 1 << lg_k;
|
769
803
|
if (window_offset > 56) throw std::logic_error("offset > 56");
|
770
804
|
|
771
805
|
// Fill the matrix with default rows in which the "early zone" is filled with ones.
|
@@ -782,12 +816,12 @@ vector_u64<A> cpc_sketch_alloc<A>::build_bit_matrix() const {
|
|
782
816
|
}
|
783
817
|
|
784
818
|
const uint32_t* slots = surprising_value_table.get_slots();
|
785
|
-
const
|
819
|
+
const uint32_t num_slots = 1 << surprising_value_table.get_lg_size();
|
786
820
|
for (size_t i = 0; i < num_slots; i++) {
|
787
821
|
const uint32_t row_col = slots[i];
|
788
822
|
if (row_col != UINT32_MAX) {
|
789
823
|
const uint8_t col = row_col & 63;
|
790
|
-
const
|
824
|
+
const uint32_t row = row_col >> 6;
|
791
825
|
// Flip the specified matrix bit from its default value.
|
792
826
|
// In the "early" zone the bit changes from 1 to 0.
|
793
827
|
// In the "late" zone the bit changes from 0 to 1.
|
@@ -799,8 +833,8 @@ vector_u64<A> cpc_sketch_alloc<A>::build_bit_matrix() const {
|
|
799
833
|
|
800
834
|
template<typename A>
|
801
835
|
void cpc_sketch_alloc<A>::write_hip(std::ostream& os) const {
|
802
|
-
|
803
|
-
|
836
|
+
write(os, kxp);
|
837
|
+
write(os, hip_est_accum);
|
804
838
|
}
|
805
839
|
|
806
840
|
template<typename A>
|