The thread: Measured, not assumed — page 5
The columns the candidates share
Three thousand tables against one query, and most of them begin the same way. Stored as a trie, the 2,424-word vocabulary has 7,710 distinct prefixes holding 17,239 letters, and a search that computes one column per prefix reads 61,449 cells against 156,714 — before it applies any bound at all. Apply the bound at a prefix instead of at a word and it reads 16,958, beating a list search that was told the answer in advance.
The floorsA floor with two variables in it
Under round-robin a Space-Saving summary's floor is 0.0203·n^1.018 over a hundred-and-twenty-eight-fold range of shard size, worst residual 2.7%. Under hashing the same measurement has no exponent at all — the local slope runs from n^5.17 to n^1.19 — and a least-squares line through it reports n^1.73 at a 441% residual.
What is taught wronglyThe filter that proposes everything
Seed-and-extend saves two thousand times the work at zero errors and costs more than doing nothing at four. Between them the selectivity falls through the floor, and where it falls is set by two numbers that can be computed before the filter is run — one of which does not contain the length of the text at all.
One pass, and no roomThe tuples a merge does not give back
A merge of thirty-two quantile summaries keeps seven times the tuples of one summary over the same values, and sixty-four keeps ten and a half. Fitted across the sweep the count goes as the shard number to the power 0.56, which answers what it converges to — it does not.
The index that replaces the textThe candidates a filter cannot avoid
A phrase index answers a search by intersecting two ranges of boundaries, and it does the intersection by walking the smaller one. On a collection of thirty-two copies that is 4,355 phrase examinations to produce 32 occurrences — 136 examinations each, and rising.
The data that is not a numberA parse that will not follow a long chain
A greedy self-referential parse bounds the copy depth by nothing at all — thirty-two copies of a text give a position costing twenty-two phrase follows. Restricting every phrase to sources no deeper than D holds it at D, and the whole question is what that costs.
What the libraries doThe number that would choose a cap
A depth histogram is one linear pass — 3.16 operations a character over thirty-two thousand of them — and it says the whole text sits at a mean depth of 3.98 with a worst of ten. Nobody prints it, and every choice of cap in this collection was made without it.
CountingThe sort whose count has no distribution
Batcher's network makes nineteen comparisons on every one of the 40,320 orderings of eight elements — sorted, reversed, adversarial or random — because it decides which pairs to compare before it sees any of them. That is two above merge sort's worst case and four below heapsort's best. At 65,536 elements the same refusal to adapt costs 4.07 times merge sort's worst case, and it buys three things no adaptive sort has, one of which is a proof of correctness that takes 65,536 inputs instead of twenty trillion.
When the algorithm is a tableThe lattice that decides the ties
Rounding a fitted substitution matrix to whole bits puts 63 of 200 alignments on a tie where the exact fit puts 15. Rounding to half bits — a finer grain, and the obvious repair — puts 79. What tracks the ties is not how fine the lattice is but how many of the six fitted costs it keeps apart: whole and half bits both leave three, an eighth of a bit leaves all six, and matches the exact fit exactly.
When the algorithm is a tableA trie fitted to its queries
A nearest-word search over a trie of 2,424 words gives a column to about 30% of the prefixes each query, and it was proposed that a stream of queries would pick out a part of the trie worth keeping and a part worth collapsing. A hundred queries have visited 93% of it; two thousand, 99.5%. A stream does not choose a part of the trie. It chooses a distance: misspelt vocabulary words all lie within two edits, and a search that starts at the median past answer does a fifth of the work of one that discovers its bound. Misspellings of words the vocabulary lacks break that, unless the search tries the learned bound once and then gives it up.
What is taught wronglyA distance that is not a distance
Under unit cost the edit distance obeys the triangle inequality and this site asserts that it does. Under a stated substitution matrix it need not, and on 42,840 enumerated triples it fails 2,832 times — taking with it every structure that prunes by distance, at a measured 13.22% of the bounds they rely on.
StructuresThe count that outlives its arrivals
A Misra-Gries counter holding six thousand is not a record of six thousand arrivals. It is a number that has been added to and taken from, and nothing in the structure says when any of it happened — so when the key stops arriving the counter stays, and goes on reporting a key with nothing in the window as the heaviest thing in it.
One pass, and no roomThe window that is even in the wrong currency
A window of four thousand arrivals in eight blocks retires a block every 5.1 seconds on a steady stream and anywhere between 0.57 and 14.8 seconds on a stream whose rate moves. The structure cannot tell, because it is counting arrivals, and the alert written against it is in seconds.
The data that is not a numberThe errors the rest of the pattern needs
Read the pattern left to right in an index of the reversed text and count the points where the interval empties. That count is a lower bound on the errors any alignment of the prefix must contain, it costs 72 rank operations, and it removes 70% of a search tree.
What a bound isThe cost that is the size of the answer
Ten range-minimum queries answer the listing at every point of a sweep where the occurrences run from 30 to 790. They cost 256 to 288 node visits — and the scan they replace costs 30 to 790, so the output-sensitive method loses until about thirty occurrences per document.
What the libraries doWhat the generated collection was right about
Five strands of conclusions, drawn on collections made by one line with one dial, checked against a corpus nobody made. Most hold. One headline was a property of the generator's alphabet, and one crossing that was guessed at turns out to be met — but only with the structure the strand on range minima built.
The floorsWork that falls as the answer grows
Output-sensitive usually means the cost rises with the answer instead of with the input. A descent over a document array costs five operations per document at an answer of seven and two at an answer of thirty-two, because the paths to many leaves share their tops.
The floorsProportional to the answer, not the alphabet
At a fixed alphabet of thirty-two, a loop costs three hundred and twenty ranks whether one symbol is present or all of them. The descent costs ten and sixty-two. The experiment has to move the answer without moving the alphabet, and the obvious sweep moves both.
CountingThe count of the part that was read
Handing back the smallest ten of 65,536 keys in order costs 965,656 comparisons by sorting them and 65,670 by a knockout tournament, against a floor of 65,526. Read to the last element, the same tournament makes exactly merge sort's 965,656 — it is merge sort, charged one element at a time. A sort's count has no term for how much of its answer anyone reads, and the two floors that do have one cannot simply be added.
When it does not fitRuns twice as long as memory
Feed 262,144 random records through a heap that holds 4,096 and the sorted runs that come out average 1.94 memories — the snowplow's famous factor of two. At a fan-in of 63 that saves a merge pass at 262,144 records, and at two of fourteen sizes in all. Feed the same heap a sorted file with one record in a thousand out of place and it writes two runs instead of sixty-four. And it spends 19 comparisons a record doing so, on every input, where sorting the chunks spends five on sorted data. The factor of two is the least of what the method does.
When the algorithm is a tableThe grain a corpus chooses for itself
A fitted substitution matrix can choose its own rounding lattice: the coarsest grain that keeps all six of its costs apart. On eight resamples of the same 400-pair corpus that grain comes out anywhere from an eighth of a bit to a forty-seventh, because it is set by whichever two costs happen to land closest, and those two differ by less than a tenth of their own standard error. Asked instead which costs it can actually tell apart, the corpus pools six into the generator's two and chooses an alignment the generator agrees with for every one of 200 test pairs on every resample.
When the algorithm is a tableA second try before any feature
A nearest-word search that learned one starting bound for a whole stream was proposed a bound for each query instead, read off features the search can compute before it walks. The proposed feature separates nothing: 58% of misspelt vocabulary words and 53% of words the vocabulary lacks begin with a real three-letter prefix. Length and unseen letter-triples do predict distance, and still buy about 1%, because the best start is two for almost every class. What closes half the gap to an oracle is a second cheap walk: one, then two, then discover.
StructuresThe counter that takes the smallest slot
Space-Saving keeps two numbers per key and they bracket the truth from both sides. On the twenty heaviest keys of a stream its mean error is a tenth of one arrival, against a hundred and ten for Misra-Gries at the same bits — and on the keys ranked past a hundred the ordering reverses.
One pass, and no roomThe period that is not a promise
Greenwald–Khanna's ε appears twice — once as the rank tolerance the structure promises, and once as ⌊1/2ε⌋, the number of updates between compressions. Unhook the second from the first and sweep it across a thousand-fold range. The tuples held move by 10%, the worst rank error by 21%, the peak by ten times and the housekeeping by seventy.
What is taught wronglyTwo structures that are one
Space-Saving never underestimates and Misra-Gries never overestimates, and they are taught as rival structures with opposite failure modes. Subtract one number from every Space-Saving counter and what is left is the Misra-Gries table, key for key and count for count, at all twenty thousand prefixes of a stream and at every table size tried.
The data that is not a numberThe table the links already knew
The exact good-suffix rule costs 757,058 character comparisons to build from its definition at 128 patterns, and 3,824 from the trie's failure links. Same table, checked at every node — a factor of 198, and the definition was never the algorithm.
The index that replaces the textA list of documents is not a list of occurrences
A pattern occurring 790 times in eight documents has an answer of size eight. Reading every occurrence to find out costs 790 array reads; the question a collection has that a text does not is the one its index does not answer.
What a bound isWhere a crossing moved to
The prediction was that a succinct range minimum would move the document listing's crossing "to a handful". It moves it from 32 occurrences per document to 11 — a factor of three, not an order of magnitude — because a constant-time query is ten lookups rather than one.
The floorsFrom the right, two of sixteen
Run the same sweep in the opposite direction and every occurrence it produces lands behind its own cursor. It reports two of sixteen, every one of them genuinely there, and nothing about the answer says fourteen are missing.
When it does not fitThe keys that arrive late
Insert 131,072 keys into a B+-tree in random order and its leaves end up 70.5% full; in ascending order, 51.6%; in descending order, 50.0%. The rule databases use to fix ascending inserts — split a full leaf at its right-hand end — fills them completely, and it does nothing for descending keys. Let one key in a hundred arrive late in an otherwise ascending stream and the rule's leaves fall from 100% to 53.4% full. How much of an index is empty is decided by the order its keys arrived in, and a trickle of disorder undoes the fix.