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
// Copyright (C) 2024 Christian Mauduit <ufoot@ufoot.org>
use std::fmt;
/// Parameters used to build bloom filters.
///
/// This is a utility to set up various bloom filter parameters.
/// There are typically 4 parameters:
///
/// - n (`nb_items`): number of items in the filter
/// - p (`fp_rate`): probability of false positives, fraction between 0 and 1
/// - m (`bit_len`): number of bits in the filter
/// - k (`nb_hash`): number of hash functions
///
/// All of them are linked, if one changes the size of the filter, obviously,
/// it changes the number of items it can hold.
///
/// A few things to keep in mind:
/// - for a pure Bloom filter, the only technical parameters that matter
/// are `nb_hash` and `bit_len`. They fully define the filter.
/// - from a functional point of view, most of the time you are interested
/// in a given `nb_items` and `fp_rate`. From there the 2 others are derived.
///
/// You may partially fill this struct, anything which contains `zero` will
/// be inferred either by deducing it from other defined params, or by
/// replacing it with a default value. This is done by calling `.adjust()`.
///
/// Also, there is a `predict` field which can be used to turn the
/// filter into a predicatable filter. This is convenient for testing,
/// and may be of interest in edge cases but most of the time a real random
/// version is prefered. It avoids bias and also can protect against
/// some DDOS attacks based on hash collision.
///
/// A few helpers are defined as well, for common use-cases.
///
/// # A bit of theory
///
/// This [interactive Bloom filter params calulator](https://hur.st/bloomfilter/?n=300000&p=0.01&m=&k=)
/// proved very useful while developping this.
/// Also it recaps the usual formulas, with:
///
/// * n -> number of items in the filter
/// * p -> probability of false positives, fraction between 0 and 1
/// * m -> number of bits in the filter
/// * k -> number of hash functions
///
/// We have:
///
/// * `n = ceil(m / (-k / log(1 - exp(log(p) / k))))`
/// * `p = pow(1 - exp(-k / (m / n)), k)`
/// * `m = ceil((n * log(p)) / log(1 / pow(2, log(2))))`
/// * `k = round((m / n) * log(2))`
///
/// A command line equivalent would be this practical
/// [bloom-filter-calculator.rb gist](https://gist.github.com/brandt/8f9ab3ceae37562a2841)
///
/// Note that there are corner cases, for example, the formula that gives
/// `bit_len` (m) from `nb_items` (n) and `fp_rate` (p), corresponds to
/// the optimal case, when `nb_hash` (k) has been chosen optimally. If not,
/// it has to be revisited and adapted to the real value of `nb_hash` (k).
/// In practice, unless you impose it, what this package does is enforce
/// a `nb_hash` (k) of 2, which is generally optimal if you consider CPU time
/// and not memory usage.
///
/// # Links
///
/// * [Bloom Filter on Wikipedia](https://en.wikipedia.org/wiki/Bloom_filter)
/// * [All About Bloom Filters](https://freecontent.manning.com/all-about-bloom-filters/)
/// * [Buffered Bloom Filters on Solid State Storage](https://www.vldb.org/archives/workshop/2010/proceedings/files/vldb_2010_workshop/ADMS_2010/adms10-canim.pdf)
/// * [Age-Partitioned Bloom Filters](https://arxiv.org/pdf/2001.03147.pdf)
/// * [countBF: A General-purpose High Accuracy and Space Efficient Counting Bloom Filter](https://arxiv.org/pdf/2106.04364.pdf)
/// * [Stable Learned Bloom Filters for Data Streams](https://www.vldb.org/pvldb/vol13/p2355-liu.pdf)
/// * [Building a Better Bloom Filter](https://www.eecs.harvard.edu/~michaelm/postscripts/tr-02-05.pdf)
///
/// # Examples
///
/// Getting a filter for a given number of items, everything else default:
///
/// ```
/// use ofilter::Params;
///
/// let params = Params::with_nb_items(10_000);
///
/// assert_eq!("{ nb_hash: 2, bit_len: 189825, nb_items: 10000, fp_rate: 0.010000, predict: false }", format!("{}", ¶ms));
/// ```
///
/// Getting a filter for a given number of items, false positive rate, and enforcing number of hash:
///
/// ```
/// use ofilter::Params;
///
/// let params = Params{
/// nb_hash: 3,
/// bit_len: 0,
/// nb_items: 100_000,
/// fp_rate: 0.1,
/// predict: false,
/// }.adjust();
///
/// assert_eq!("{ nb_hash: 3, bit_len: 480833, nb_items: 100000, fp_rate: 0.100000, predict: false }", format!("{}", ¶ms));
/// ```
#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
#[derive(Debug, Clone)]
pub struct Params {
/// Number of hash functions used by the filter.
///
/// Also referred to as `k` is most Bloom filter papers.
pub nb_hash: usize,
/// Length of the bit vector used by the filter.
///
/// Also referred to as `m` is most Bloom filter papers.
pub bit_len: usize,
/// Number of items the Bloom filter is designed for.
///
/// Also referred to as `n` is most Bloom filter papers.
pub nb_items: usize,
/// False positive rate of the Bloom filter.
///
/// Also referred to as `p` is most Bloom filter papers.
pub fp_rate: f64,
/// If set to true, Bloom filter is predictable.
///
/// Use this for testing, when you want something that
/// is 100% predictable and avoid flaky behavior.
/// In production, it would be safer to rely on the
/// random, statistical default behavior.
///
/// One reason is security, among other examples, using
/// a predictable hash for a cache may expose you to
/// some sort of DDOS attack.
///
/// If in doubt, leave it to false.
pub predict: bool,
}
/// Compare two parameters sets.
///
/// The false positive rate is rounded to 6 digits.
///
/// # Examples
///
/// ```
/// use ofilter::Params;
///
/// let p1 = Params::with_nb_items(100);
/// let mut p2 = Params::with_nb_items(100);
/// p2.fp_rate += 1e-8;
/// assert_eq!(p1, p2);
/// ```
impl std::cmp::PartialEq for Params {
fn eq(&self, other: &Self) -> bool {
if self.nb_hash != other.nb_hash {
return false;
}
if self.bit_len != other.bit_len {
return false;
}
if self.nb_items != other.nb_items {
return false;
}
((self.fp_rate * 1e6).round() as isize) == ((other.fp_rate * 1e6).round() as isize)
}
}
impl std::cmp::Eq for Params {}
/// Pretty-print parameters
///
/// # Examples
///
/// ```
/// use ofilter::Params;
///
/// let p = Params::with_nb_items(100);
/// assert_eq!("{ nb_hash: 2, bit_len: 1899, nb_items: 100, fp_rate: 0.009992, predict: false }", format!("{}", p));
/// ```
impl fmt::Display for Params {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
write!(
f,
"{{ nb_hash: {}, bit_len: {}, nb_items: {}, fp_rate: {:0.6}, predict: {} }}",
self.nb_hash, self.bit_len, self.nb_items, self.fp_rate, self.predict
)
}
}
/// Default number of hash functions.
///
/// We use 2 because most of the time, the game changer is CPU time,
/// and limiting the number of hashes makes things quite optimal from
/// that point of view. It is only a default, feel free to increase
/// when relevant, but this proved a quite sensible setting.
pub const DEFAULT_NB_HASH: usize = 2;
/// Default size of the bit vector, in bits.
///
/// This value looks pretty random and corresponds to the required
/// buffer len to power a 1000 items filter with 1% accuracy.
/// Generally speaking, do not impose the buffer len, it is safe
/// to derive it from other more functional aspects such as number
/// of logical items or false positive rate.
pub const DEFAULT_BIT_LEN: usize = 18_983;
/// Default number of items the filter should contain.
///
/// Think of this as a maximum capacity, above this, the filter
/// stops honoring the false positive rate and ultimately,
/// every item will test positive.
pub const DEFAULT_NB_ITEMS: usize = 1_000;
/// Default false positive rate.
///
/// Bloom filters, by design, have false positives. This is the
/// default rate, feel free to fine-tune it. But beware that
/// lowering this too much can yield very big filters with
/// a huge bit buffer. It's a tradeoff.
pub const DEFAULT_FALSE_P: f64 = 0.01;
const MIN_FALSE_P: f64 = 0.000_000_001;
const MAX_FALSE_P: f64 = 0.999_999_999;
impl Params {
/// Parameters to store a given number of items.
///
/// All other values are using defaults, or are derive from this.
/// This is the go-to function to create a reasonable filter that
/// holds N items.
///
/// # Examples:
///
/// ```
/// use ofilter::Params;
///
/// let p = Params::with_nb_items(10_000);
/// assert_eq!("{ nb_hash: 2, bit_len: 189825, nb_items: 10000, fp_rate: 0.010000, predict: false }", format!("{}", p));
/// ```
pub fn with_nb_items(nb_items: usize) -> Params {
Params {
nb_hash: DEFAULT_NB_HASH,
bit_len: 0,
nb_items,
fp_rate: DEFAULT_FALSE_P,
predict: false,
}
.adjust()
}
/// Parameters to store a given number of items, enforcing false positive rate.
///
/// By defining `nb_items` and `fp_rate`, the filter is almost completely defined.
/// The other missing parameter is the number of hash functions. There is an
/// optimal number for this, but this function will enforce a default of 2,
/// which most of the time yields excellent results, and avoids calculating
/// too many hashes.
///
/// If you want to use the optimal number of hashes, create the Params struct
/// with `nb_items`, `fp_rate` and `nb_hash` defined, then call `.adjust()`.
///
/// # Examples:
///
/// ```
/// use ofilter::Params;
///
/// let p = Params::with_nb_items_and_fp_rate(10_000, 0.001);
/// assert_eq!("{ nb_hash: 2, bit_len: 622402, nb_items: 10000, fp_rate: 0.001000, predict: false }", format!("{}", p));
/// ```
pub fn with_nb_items_and_fp_rate(nb_items: usize, fp_rate: f64) -> Params {
Params {
nb_hash: DEFAULT_NB_HASH,
bit_len: 0,
nb_items,
fp_rate,
predict: false,
}
.adjust()
}
/// Parameters to guess the number of items from bit buffer size.
///
/// If the only thing you care about, this will give you the best
/// result for that amount of memory.
///
/// # Examples:
///
/// ```
/// use ofilter::Params;
///
/// let p = Params::with_bit_len(8_192);
/// assert_eq!("{ nb_hash: 2, bit_len: 8192, nb_items: 432, fp_rate: 0.010019, predict: false }", format!("{}", p));
/// ```
pub fn with_bit_len(bit_len: usize) -> Params {
Params {
nb_hash: DEFAULT_NB_HASH,
bit_len,
nb_items: 0,
fp_rate: DEFAULT_FALSE_P,
predict: false,
}
.adjust()
}
/// Parameters to store a given number of items, enforcing bit buffer size
///
/// If the only thing you care about, this will give you the best
/// result for that amount of memory.
///
/// # Examples:
///
/// ```
/// use ofilter::Params;
///
/// let p = Params::with_bit_len_and_nb_items(8_192, 1_000);
/// assert_eq!("{ nb_hash: 2, bit_len: 8192, nb_items: 1000, fp_rate: 0.046925, predict: false }", format!("{}", p));
/// ```
pub fn with_bit_len_and_nb_items(bit_len: usize, nb_items: usize) -> Params {
Params {
nb_hash: DEFAULT_NB_HASH,
bit_len,
nb_items,
fp_rate: 0.0,
predict: false,
}
.adjust()
}
/// Estimate the number of items one can store.
///
/// This applies the formula `n = ceil(m / (-k / log(1 - exp(log(p) / k))))`.
///
/// # Examples:
///
/// ```
/// use ofilter::Params;
///
/// assert_eq!(1994, Params::estimate_nb_items(5, 10_000, 0.1));
/// ```
pub fn estimate_nb_items(nb_hash: usize, bit_len: usize, fp_rate: f64) -> usize {
// n = ceil(m / (-k / log(1 - exp(log(p) / k))))
std::cmp::max(
(bit_len as f64
/ (-(nb_hash as f64) / (1.0 - (fp_rate.ln() / nb_hash as f64).exp()).ln()))
.ceil() as usize,
1,
)
}
/// Estimate the false positive rate.
///
/// This applies the formula `p = powf(1 - exp(-k / (m / n)), k)`
///
/// # Examples:
///
/// ```
/// use ofilter::Params;
///
/// assert_eq!("0.009431", format!("{:0.6}", Params::estimate_fp_rate(5, 10_000, 1_000)));
/// ```
pub fn estimate_fp_rate(nb_hash: usize, bit_len: usize, nb_items: usize) -> f64 {
// p = powf(1 - exp(-k / (m / n)), k)
(1.0 - (-(nb_hash as f64) / (bit_len as f64 / nb_items as f64)).exp()).powf(nb_hash as f64)
}
/// Calculate the optimal number of bits.
///
/// This applies the formula `m = ceil((n * log(p)) / log(1 / pow(2, log(2))))`
///
/// Note that this is the *optimal* number of bits, but for the case
/// where the number of hash has also been chosen optimally. You'll notice
/// the formula does not depend on `fp_rate` (p).
///
/// See `guess_bit_len_for_fp_rate` if you want to also take `fp_rate` in account.
///
/// # Examples:
///
/// ```
/// use ofilter::Params;
///
/// assert_eq!(62353, Params::optimal_bit_len(10_000, 0.05));
/// ```
pub fn optimal_bit_len(nb_items: usize, fp_rate: f64) -> usize {
// m = ceil((n * log(p)) / log(1 / pow(2, log(2))));
std::cmp::max(
((nb_items as f64 * fp_rate.ln()) / (1.0 / 2f64.powf(2f64.ln())).ln()).ceil() as usize,
1,
)
}
/// Calculate the optimal number of hash.
///
/// This applies the formula `k = round((m / n) * log(2))`
///
/// Note that this is the *optimal* number of hash, but it does not
/// account for false positive rate, which you may want to impose.
/// You'll notice the formula does not depend on `fp_rate` (p).
///
/// See `guess_bit_len_for_fp_rate` if you want to also take `fp_rate` in account.
///
/// # Examples:
///
/// ```
/// use ofilter::Params;
///
/// assert_eq!(13, Params::optimal_nb_hash(10_000, 500));
/// ```
pub fn optimal_nb_hash(bit_len: usize, nb_items: usize) -> usize {
// k = round((m / n) * log(2));
std::cmp::max(
((bit_len as f64 / nb_items as f64) * 2f64.ln()).floor() as usize,
1,
)
}
/// Guess the number of bits for a given false positive rate.
///
/// There is a formula `m = ceil((n * log(p)) / log(1 / pow(2, log(2))))` to get this
/// result but it gives it for the optimal case. Sometimes you can
/// prefer to, typically, lower the number of bits in the bits buffer to
/// get something smaller in memory, and just give up a bit on the optimal
/// `fp_rate` (p).
///
/// This function does a bisect search to find the correct value.
///
/// # Examples:
///
/// ```
/// use ofilter::Params;
///
/// assert_eq!(480833, Params::guess_bit_len_for_fp_rate(3, 100_000, 0.1));
/// ```
pub fn guess_bit_len_for_fp_rate(nb_hash: usize, nb_items: usize, fp_rate: f64) -> usize {
// Doing a bisect to find the point where fp_rate is exactly what we want,
// knowing that fp_rate decreases as nb_items increases
let mut bit_len_hi = usize::MAX / 2 - 1; // to avoid out-of-bounds when summing
let mut bit_len_lo = std::cmp::min(Self::optimal_bit_len(nb_items, fp_rate), bit_len_hi);
let fp_rate = Self::adjust_fp_rate(fp_rate);
while bit_len_hi > bit_len_lo + 1 {
let fp_rate_lo = Self::estimate_fp_rate(nb_hash, bit_len_lo, nb_items);
if fp_rate_lo < fp_rate {
return bit_len_lo;
}
let bit_len_mi = (bit_len_lo + bit_len_hi) / 2;
let fp_rate_mi = Self::estimate_fp_rate(nb_hash, bit_len_mi, nb_items);
if fp_rate_mi <= fp_rate {
bit_len_hi = bit_len_mi;
} else {
bit_len_lo = bit_len_mi;
}
}
bit_len_hi
}
/// Guess the number of hash for a given false positive rate.
///
/// There is a formula `k = round((m / n) * log(2))` to get this
/// result but it gives it for the optimal case. Sometimes you can
/// prefer to, typically, lower the number of hash functions to
/// get something faster, and just give up a bit on the optimal
/// `fp_rate` (p).
///
/// This function starts with the optimal `nb_hash` as a seed,
/// and then tries to lower the it as long as the false positive rate
/// gets closer to the target `fp_rate`.
///
/// # Examples:
///
/// ```
/// use ofilter::Params;
///
/// assert_eq!(6, Params::guess_nb_hash_for_fp_rate(100_000, 10_000, 0.0005));
/// ```
pub fn guess_nb_hash_for_fp_rate(bit_len: usize, nb_items: usize, fp_rate: f64) -> usize {
let mut best_diff = Self::estimate_fp_rate(DEFAULT_NB_HASH, bit_len, nb_items);
let mut best_nb_hash = DEFAULT_NB_HASH;
let mut tmp_nb_hash = Self::optimal_nb_hash(bit_len, nb_items);
while tmp_nb_hash > 0 {
let diff = (Self::estimate_fp_rate(tmp_nb_hash, bit_len, nb_items) - fp_rate).abs();
if diff < best_diff {
best_diff = diff;
best_nb_hash = tmp_nb_hash;
}
tmp_nb_hash -= 1;
}
best_nb_hash
}
fn adjust_fp_rate(fp_rate: f64) -> f64 {
if fp_rate == 0.0 {
0.0
} else if fp_rate <= MIN_FALSE_P {
MIN_FALSE_P
} else if fp_rate >= MAX_FALSE_P {
MAX_FALSE_P
} else {
fp_rate
}
}
/// Adjust params so that they are consistent.
///
/// The 4 parameters in a Bloom filter are closely linked.
/// More precisely, with:
///
/// * n (`nb_items`) -> number of items in the filter
/// * p (`fp_rate`) -> probability of false positives, fraction between 0 and 1
/// * m (`bit_len`) -> number of bits in the filter
/// * k (`nb_hash`) -> number of hash functions
///
/// We have:
///
/// * `n = ceil(m / (-k / log(1 - exp(log(p) / k))))`
/// * `p = pow(1 - exp(-k / (m / n)), k)`
/// * `m = ceil((n * log(p)) / log(1 / pow(2, log(2))))`
/// * `k = round((m / n) * log(2))`
///
/// This function works as follows:
/// - if any parameter is non-zero, tries to respect it
/// - if any parameter is zero, and can be inferred from others, calculate it
/// - for all other parameters, use defaults
///
/// Once everything has a value, it will ultimately re-calculate the `fp_rate`
/// so that it matches the other 3, which are integers, so may not be changed
/// linearily.
///
/// In practice, use it to feed the `new_with_params` constructors for filters.
///
/// # Examples:
///
/// ```
/// use ofilter::Params;
///
/// let params = Params{
/// nb_hash: 2,
/// bit_len: 0,
/// nb_items: 1_000,
/// fp_rate: 0.1,
/// predict: false,
/// };
///
/// assert_eq!("{ nb_hash: 2, bit_len: 0, nb_items: 1000, fp_rate: 0.100000, predict: false }",
/// format!("{}", params)
/// );
/// assert_eq!("{ nb_hash: 2, bit_len: 5262, nb_items: 1000, fp_rate: 0.099980, predict: false }",
/// format!("{}", params.adjust())
/// );
/// ```
pub fn adjust(self) -> Self {
let fp_rate = Self::adjust_fp_rate(self.fp_rate);
// bit_len + nb_items -> nb_hash (optimal, guess)
// nb_hash + bit_len + fp_rate -> nb_items (exact, no choice)
// nb_items + fp_rate -> bit_len (optimal, guess)
// nb_hash + bit_len + nb_items -> fp_rate (exact, no choice)
// Everything is defined, just decide what we're going to
// expose in terms of "this filter should be capable of that".
if self.nb_hash > 0 && self.bit_len > 0 {
let fp_rate = if fp_rate == 0.0 {
DEFAULT_FALSE_P
} else {
fp_rate
};
let nb_items = match self.nb_items {
0 => Self::estimate_nb_items(self.nb_hash, self.bit_len, fp_rate),
nb_items => nb_items,
};
// Adjust, always, the fp_rate, as it's a float altering it a bit
// is never so surprising, it's understandable that is has to
// be rounded to match the other values.
return Params {
nb_hash: self.nb_hash,
bit_len: self.bit_len,
nb_items,
fp_rate: Self::estimate_fp_rate(self.nb_hash, self.bit_len, nb_items),
predict: self.predict,
};
// The only output on which we do not call adjust()
};
// Functionnally, the filter is fully defined by
// number of items and false positives, derive tech details.
if self.nb_items > 0 && fp_rate > 0.0 {
if self.nb_hash == 0 && self.bit_len > 0 {
return Params {
// Too many constraints, what we do here is find the
// right number of hashes to satisfy best the constraints
// on false positives. Sometimes there is litterally no
// way to do it, sometimes lowering the number of hash
// may lower the CPU for a result which is good enough.
nb_hash: Self::guess_nb_hash_for_fp_rate(self.bit_len, self.nb_items, fp_rate),
bit_len: self.bit_len,
nb_items: self.nb_items,
fp_rate,
predict: self.predict,
}
.adjust();
};
if self.nb_hash > 0 && self.bit_len == 0 {
return Params {
nb_hash: self.nb_hash,
bit_len: Self::guess_bit_len_for_fp_rate(self.nb_hash, self.nb_items, fp_rate),
nb_items: self.nb_items,
fp_rate,
predict: self.predict,
}
.adjust();
}
}
// The filter is partially defined, start by defining the
// number of hash, because it's performance impacting, and
// by default a small number of hash makes it faster.
if self.nb_hash == 0 {
let nb_hash = if self.bit_len > 0 && self.nb_items > 0 {
Self::optimal_nb_hash(self.bit_len, self.nb_items)
} else {
DEFAULT_NB_HASH
};
return Params {
nb_hash,
bit_len: self.bit_len,
nb_items: self.nb_items,
fp_rate,
predict: self.predict,
}
.adjust();
};
// If no choice could be made so far, define default
// functional requirements and see if we can do better.
if fp_rate == 0.0 {
return Params {
nb_hash: self.nb_hash,
bit_len: self.bit_len,
nb_items: self.nb_items,
fp_rate: DEFAULT_FALSE_P,
predict: self.predict,
}
.adjust();
};
if self.nb_items == 0 {
return Params {
nb_hash: self.nb_hash,
bit_len: self.bit_len,
nb_items: DEFAULT_NB_ITEMS,
fp_rate,
predict: self.predict,
}
.adjust();
};
// This should be unreachable, but keeping it as a catch all.
if self.bit_len == 0 {
return Params {
nb_hash: self.nb_hash,
bit_len: DEFAULT_BIT_LEN,
nb_items: self.nb_items,
fp_rate,
predict: self.predict,
}
.adjust();
};
// No clue what to do, pure defaults...
Self::default()
}
}
impl std::default::Default for Params {
fn default() -> Self {
Params {
nb_hash: DEFAULT_NB_HASH,
bit_len: DEFAULT_BIT_LEN,
nb_items: DEFAULT_NB_ITEMS,
fp_rate: DEFAULT_FALSE_P,
predict: false,
}
.adjust()
}
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn test_default() {
assert_eq!(Params::default(), Params::default().adjust());
}
#[test]
fn test_nb_items() {
assert_eq!(17, Params::estimate_nb_items(2, 1_000, 0.001));
}
#[test]
fn test_fp_rate() {
assert_eq!(
32_859,
(Params::estimate_fp_rate(2, 1_000, 100) * 1e6).round() as usize
);
}
#[test]
fn test_optimal_bit_len() {
assert_eq!(1438, Params::optimal_bit_len(100, 0.001));
}
#[test]
fn test_optimal_nb_hash() {
assert_eq!(69, Params::optimal_nb_hash(10_000, 100));
}
#[test]
fn test_params_default() {
assert_eq!(
"{ nb_hash: 2, bit_len: 18983, nb_items: 1000, fp_rate: 0.009999, predict: false }",
format!("{}", Params::default())
);
}
#[test]
fn test_params_adjust() {
let params = Params {
nb_hash: 0,
bit_len: 0,
nb_items: 1_000_000,
fp_rate: 0.001,
predict: false,
};
assert_eq!(
"{ nb_hash: 2, bit_len: 62240198, nb_items: 1000000, fp_rate: 0.001000, predict: false }",
format!("{}", params.adjust())
);
let params = Params {
nb_hash: 4,
bit_len: 0,
nb_items: 1_000_000,
fp_rate: 0.1,
predict: false,
};
assert_eq!(
"{ nb_hash: 4, bit_len: 4840764, nb_items: 1000000, fp_rate: 0.100000, predict: false }",
format!("{}", params.adjust())
);
let params = Params {
nb_hash: 0,
bit_len: 10_000_000,
nb_items: 1_000_000,
fp_rate: 0.0001,
predict: false,
};
assert_eq!(
"{ nb_hash: 6, bit_len: 10000000, nb_items: 1000000, fp_rate: 0.008436, predict: false }",
format!("{}", params.adjust())
);
let params = Params {
nb_hash: 0,
bit_len: 10_000_000,
nb_items: 1_000_000,
fp_rate: 0.1,
predict: false,
};
assert_eq!(
"{ nb_hash: 1, bit_len: 10000000, nb_items: 1000000, fp_rate: 0.095163, predict: false }",
format!("{}", params.adjust())
);
let params = Params {
nb_hash: 2,
bit_len: 0,
nb_items: 1_000_000,
fp_rate: 0.00001,
predict: false,
};
assert_eq!(
"{ nb_hash: 2, bit_len: 631455005, nb_items: 1000000, fp_rate: 0.000010, predict: false }",
format!("{}", params.adjust())
);
let params = Params {
nb_hash: 2,
bit_len: 0,
nb_items: 1_000_000,
fp_rate: 0.1,
predict: false,
};
assert_eq!(
"{ nb_hash: 2, bit_len: 5261353, nb_items: 1000000, fp_rate: 0.100000, predict: false }",
format!("{}", params.adjust())
);
}
}