Commit Graph

97 Commits

Author SHA1 Message Date
Viktor Lofgren
4fcd4a8197 (index) Refactor to reduce the level of indirection 2024-05-19 12:40:33 +02:00
Viktor Lofgren
19163fa883 (array) Clean up the Array library
IntArray gets the YAGNI axe.   The array library had two implementations, one for longs which was used, and one for ints, which only ever saw bit rot.   Removing the latter, as all it ever did was clutter up the codebase and add technical debt.  If we need int arrays, we fork LongArray again (or add int capabilities to it)

Also cleaning up the interfaces, removing layers of redundant abstractions and adding javadocs.

Finally adding sz=2 specializations to the quick- and insertion sort algorithms.  It seems the JIT isn't optimizing these particularly well, this is an attempt to help it out a bit.
2024-05-18 13:23:06 +02:00
Viktor
2d49071e96
Merge branch 'master' into run-outside-docker 2024-04-25 18:53:26 +02:00
Viktor Lofgren
e4b34b6ee6 (index) Correctly detect the presence of an all-virtual path through the query 2024-04-25 14:01:46 +02:00
Viktor Lofgren
32fe864a33 (build) Java 22 and its consequences has been a disaster for Marginalia Search
Roll back to JDK 21 for now, and make Java version configurable in the root build.gradle

The project has run into no less than three distinct show-stopping bugs in JDK22, across multiple vendors, and gradle still doesn't fully support it, meaning you need multiple JDK versions installed.
2024-04-24 14:44:39 +02:00
Viktor Lofgren
f46733a47a (ranking) TermCoherenceFactory should be run for size=2 queries 2024-04-24 14:44:39 +02:00
Viktor Lofgren
5165cf6d15 (ranking) Set regularMask correctly 2024-04-24 14:44:39 +02:00
Viktor Lofgren
4489b21528 (ranking) Cleanup 2024-04-24 14:44:39 +02:00
Viktor Lofgren
f623b37577 (ranking) Suppress NaN:s in ranking output 2024-04-24 14:44:39 +02:00
Viktor Lofgren
f4a2fea451 (ranking, bugfix) Use bm25NgramWeight and not full weight for bM25N 2024-04-24 14:44:39 +02:00
Viktor Lofgren
a748fc5448 (index, bugfix) Pass url quality to query service 2024-04-24 14:44:39 +02:00
Viktor Lofgren
0dcca0cb83 (index) Fix TCF bug where the ngram terms would be considered instead of the regular ones due to a logical derp 2024-04-24 14:44:39 +02:00
Viktor Lofgren
b80a83339b (qs) Additional info in query debug UI 2024-04-24 14:44:39 +02:00
Viktor Lofgren
eb74d08f2a (qs) Additional info in query debug UI 2024-04-24 14:44:39 +02:00
Viktor Lofgren
462aa9af26 (query) Update ranking parameters with new variables for bm25 ngrams and tcf mutual jaccard
The change also makes it so that as long as the values are defaults, they don't need to be sent over the wire and decoded.
2024-04-24 14:44:39 +02:00
Viktor Lofgren
44b33798f3 (index) Clean up jaccard index term code and down-tune the parameter's importance a bit 2024-04-24 14:44:39 +02:00
Viktor Lofgren
2f0b648fad (index) Add jaccard index term to boost results based on term overlap 2024-04-24 14:44:39 +02:00
Viktor Lofgren
de0e56f027 (index) Remove position overlap check, coherences will do the work instead 2024-04-24 14:44:39 +02:00
Viktor Lofgren
973ced7b13 (index) Omit absent terms from coherence checks 2024-04-24 14:44:39 +02:00
Viktor Lofgren
cb4b824a85 (index) Split ngram and regular keyword bm25 calculation and add ngram score as a bonus 2024-04-24 14:44:39 +02:00
Viktor Lofgren
e0224085b4 (index) Improve recall for small queries
Partially reverse the previous commit and add a query head for the priority index when there are few query interpretations.
2024-04-24 14:44:39 +02:00
Viktor Lofgren
44c1e1d6d9 (index) Remove dead code
Since the performance fix in 3359f72239 had a huge positive impact without reducing result quality, it's possible to remove the QueryBranchWalker and associated code.
2024-04-24 14:44:39 +02:00
Viktor Lofgren
c620e9c026 (index) Experimental performance regression fix 2024-04-24 14:44:39 +02:00
Viktor Lofgren
1bb88968c5 (test) Fix broken test 2024-04-24 14:44:39 +02:00
Viktor Lofgren
df75e8f4aa (index) Explicitly free LongQueryBuffers 2024-04-24 14:44:39 +02:00
Viktor Lofgren
adf846bfd2 (index) Fix term coherence evaluation
The code was incorrectly using the documentId instead of the combined id, resulting in almost all result sets being incorrectly seen as zero.
2024-04-24 14:44:39 +02:00
Viktor Lofgren
1748fcc5ac (valuation) Impose stronger constraints on locality of terms
Clean up logic a bit
2024-04-24 14:44:39 +02:00
Viktor Lofgren
08416393e0 (valuation) Impose stronger constraints on locality of terms 2024-04-24 14:44:39 +02:00
Viktor Lofgren
155be1078d (index) Fix priority search terms
This functionality fell into disrepair some while ago.  It's supposed to allow non-mandatory search terms that boost the ranking if they are present in the document.
2024-04-24 14:44:39 +02:00
Viktor Lofgren
6efc0f21fe (index) Clean up data model
The change set cleans up the data model for the term-level data.  This used to contain a bunch of fields with document-level metadata.  This data-duplication means a larger memory footprint and worse memory locality.

The ranking code is also modified to not accept SearchResultKeywordScores, but rather CompiledQueryLong and CqDataInts containing only the term metadata and the frequency information needed for ranking.  This is again an effort to improve memory locality.
2024-04-24 14:44:39 +02:00
Viktor Lofgren
55f627ed4c (index) Clean up the code 2024-04-24 14:44:38 +02:00
Viktor Lofgren
4fb86ac692 (search) Fix outdated assumptions about the results
We no longer break the query into "sets" of search terms and need to adapt the code to not use this assumption.

For the API service, we'll simulate the old behavior to keep the API stable.

For the search service, we'll introduce a new way of calculating positions through tree aggregation.
2024-04-24 14:44:38 +02:00
Viktor Lofgren
6cba6aef3b (minor) Remove dead code 2024-04-24 14:44:38 +02:00
Viktor Lofgren
7e216db463 (index) Add origin trace information for index readers
This used to be supported by the system but got lost in refactoring at some point.
2024-04-24 14:44:38 +02:00
Viktor Lofgren
e3316a3672 (index) Clean up new index query code 2024-04-24 14:44:38 +02:00
Viktor Lofgren
a3a6d6292b (qs, index) New query model integrated with index service.
Seems to work, tests are green and initial testing finds no errors.  Still a bit untested, committing WIP as-is because it would suck to lose weeks of work due to a drive failure or something.
2024-04-24 14:44:38 +02:00
Viktor Lofgren
4668b1ddcb (build) Java 22 and its consequences has been a disaster for Marginalia Search
Roll back to JDK 21 for now, and make Java version configurable in the root build.gradle

The project has run into no less than three distinct show-stopping bugs in JDK22, across multiple vendors, and gradle still doesn't fully support it, meaning you need multiple JDK versions installed.
2024-04-24 13:54:04 +02:00
Viktor Lofgren
8769704462 (ranking) TermCoherenceFactory should be run for size=2 queries 2024-04-21 12:29:25 +02:00
Viktor Lofgren
ed250f57f2 (ranking) Set regularMask correctly 2024-04-19 14:31:57 +02:00
Viktor Lofgren
e92c25f7e0 (ranking) Cleanup 2024-04-19 14:13:12 +02:00
Viktor Lofgren
3ab563f314 (ranking) Suppress NaN:s in ranking output 2024-04-19 13:58:28 +02:00
Viktor Lofgren
426338cb45 (ranking, bugfix) Use bm25NgramWeight and not full weight for bM25N 2024-04-19 12:41:48 +02:00
Viktor Lofgren
5fa2375898 (index, bugfix) Pass url quality to query service 2024-04-19 12:41:26 +02:00
Viktor Lofgren
41782a0ab5 (index) Fix TCF bug where the ngram terms would be considered instead of the regular ones due to a logical derp 2024-04-19 12:19:26 +02:00
Viktor Lofgren
9b06433b82 (qs) Additional info in query debug UI 2024-04-19 12:18:53 +02:00
Viktor Lofgren
def607d840 (qs) Additional info in query debug UI 2024-04-19 11:46:27 +02:00
Viktor Lofgren
7641a02f31 (query) Update ranking parameters with new variables for bm25 ngrams and tcf mutual jaccard
The change also makes it so that as long as the values are defaults, they don't need to be sent over the wire and decoded.
2024-04-18 10:36:15 +02:00
Viktor Lofgren
d64bd227cf (index) Clean up jaccard index term code and down-tune the parameter's importance a bit 2024-04-17 17:40:16 +02:00
Viktor Lofgren
c5ab0a9054 (index) Add jaccard index term to boost results based on term overlap 2024-04-17 16:50:26 +02:00
Viktor Lofgren
dac948973d (index) Remove position overlap check, coherences will do the work instead 2024-04-17 14:20:01 +02:00