Clément Renault
44fec1b6c9
Cache prefixes of a length of 2
2020-01-14 18:17:52 +01:00
Clément Renault
54dacb362d
Use different algorithms for different documents ratios
2020-01-14 17:51:08 +01:00
Clément Renault
6edb460bea
Try with an exponential search
2020-01-14 16:52:24 +01:00
Clément Renault
40dab80dfa
Change the way we filter the documents
2020-01-14 14:18:01 +01:00
Clément Renault
681711fced
Fix query ids to be usize
2020-01-14 13:12:42 +01:00
Clément Renault
21c1473e0c
Introduce the distance data
2020-01-14 11:38:04 +01:00
Clément Renault
8acbdcbbad
wip: Make the new query tree work with the criteria
2020-01-13 14:36:06 +01:00
Clément Renault
da8abebfa2
Introduce the query words mapping along with the query tree
2020-01-13 13:29:47 +01:00
Clément Renault
4f7a7ea0bb
Faster intersection group by
2020-01-09 16:30:03 +01:00
Clément Renault
d6c9ba8f08
Store the postings lists
2020-01-09 15:04:53 +01:00
Clément Renault
ec8916bf54
Change the debug outputs
2020-01-09 12:05:39 +01:00
Clément Renault
81c573ec92
Add the raw document IDs to the postings lists
2020-01-08 15:30:43 +01:00
Clément Renault
9420edadf4
Introduce the Postings type to decorrelate the DocumentIds
2020-01-08 14:48:23 +01:00
Clément Renault
d724a7659e
Introduce a query tree context struct
2020-01-08 13:37:22 +01:00
Clément Renault
887c212b49
Add more logs about the docids construction
2020-01-08 13:22:42 +01:00
Clément Renault
07937ed6d7
Use the prefix caches
2020-01-08 13:14:07 +01:00
Clément Renault
a262c67ec3
limit the search in the FST
2020-01-08 13:06:12 +01:00
Clément Renault
13ca30c4d8
WIP: Made the query tree traversing support prefix search
2020-01-08 12:02:58 +01:00
Clément Renault
fbcec2975d
wip: Impl a basic tree traversing
2020-01-07 18:24:13 +01:00
Clément Renault
6e1f4af833
wip: Create a tree from query but need to show synonyms
2020-01-07 18:24:13 +01:00
Clément Renault
856c5c4214
Fix group offset computing
2019-12-31 14:24:10 +01:00
Clément Renault
670e80c151
Use the cached postings lists in the query system
2019-12-31 13:32:36 +01:00
Clément Renault
eed07c724f
Add more logging for postings lists fetching by word
2019-12-31 13:32:36 +01:00
Clément Renault
99d35fb940
Introduce a first version of a number of candidates reducer
...
It works by ignoring the postings lists associated to documents that the previous words did not returned
2019-12-31 13:32:36 +01:00
Clément Renault
106b886873
Cache the prefix postings lists
2019-12-30 18:01:32 +01:00
Clément Renault
928876b553
Introduce the postings lists caching stores
...
Currently not used
2019-12-30 18:01:27 +01:00
Clément Renault
58836d89aa
Rename the PrefixCache into PrefixDocumentsCache
2019-12-30 15:42:09 +01:00
Clément Renault
1a5a104f13
Display proximity evaluation number of calls
2019-12-30 15:42:09 +01:00
Clément Renault
064cfa4755
Add more debug, where are those 100ms
2019-12-30 15:42:08 +01:00
Clément Renault
ed6172aa94
Add a time measurement of the criterion loop
2019-12-30 15:42:08 +01:00
Clément Renault
8c140f6bcd
Increase the disk usage limit
2019-12-30 15:42:08 +01:00
Clément Renault
1e1f0fcaf5
Introduce a basic cache system for first letters
2019-12-30 15:42:08 +01:00
Clément Renault
d21352a109
Change the time measurement of the FST
2019-12-30 15:42:08 +01:00
Clément Renault
4be11f961b
Use an ugly trick to avoid cloning the FST
2019-12-30 15:42:07 +01:00
Clément Renault
1163f390b3
Restrict FST search to the first letter of the word
2019-12-30 15:42:07 +01:00
Clément Renault
691e2a3c1d
Fix a blocking channel, appearing like a deadlock
2019-12-30 15:28:28 +01:00
Clément Renault
04bb49989f
Add more debug timings
2019-12-20 14:18:48 +01:00
Clément Renault
d12ff15ee3
Set the indexes info in the create_index function
2019-12-19 10:38:56 +01:00
Clément Renault
40c0b14d1c
Reintroduce searchable attributes and reordering
2019-12-13 14:38:25 +01:00
Clément Renault
a4dd033ccf
Rename raw_matches into bare_matches
2019-12-13 14:38:25 +01:00
Clément Renault
48e8778881
Clean up the modules declarations
2019-12-13 14:38:25 +01:00
Clément Renault
4be23efe66
Remove the AttrCount type
...
Could probably be reintroduced later
2019-12-13 14:38:25 +01:00
Clément Renault
7d67750865
Reintroduce exacteness for one word document field
2019-12-13 14:38:25 +01:00
Clément Renault
746e6e170c
Make the test pass again
2019-12-13 14:38:24 +01:00
Clément Renault
d93e35cace
Introduce ContextMut and Context structs
2019-12-13 14:38:24 +01:00
Clément Renault
d75339a271
Prefer summing the attribute
2019-12-13 14:38:24 +01:00
Clément Renault
86ee0cbd6e
Introduce bucket_sort_with_distinct function
2019-12-13 14:38:24 +01:00
Clément Renault
248ccfc0d8
Update the criteria to the new ones
2019-12-13 14:38:24 +01:00
Clément Renault
ea148575cf
Remove the raw_query functions
2019-12-13 14:38:23 +01:00
Clément Renault
efc2be0b7b
Bump the sdset dependency to 0.3.6
2019-12-13 14:38:23 +01:00
Clément Renault
8d71112dcb
Rewrite the phrase query postings lists
...
This simplified the multiword_rewrite_matches function a little bit.
2019-12-13 14:38:23 +01:00
Clément Renault
dd03a6256a
Debug pre filtered number of documents
2019-12-13 14:38:23 +01:00
Clément Renault
9c03bb3428
First probably working phrase query doc filtering
2019-12-13 14:38:23 +01:00
Clément Renault
22b19c0d93
Fix the processed distance algorithm
2019-12-13 14:38:22 +01:00
Clément Renault
0f698d6bd9
Work in progress: Bad Typo detection
...
I have an issue where "speakers" is split into "speaker" and "s",
when I compute the distances for the Typo criterion,
it takes "s" into account and put a distance of zero in the bucket 0
(the "speakers" bucket), therefore it reports any document matching "s"
without typos as best results.
I need to make sure to ignore "s" when its associated part "speaker"
doesn't even exist in the document and is not in the place
it should be ("speaker" followed by "s").
This is hard to think that it will had much computation time to
the Typo criterion like in the previous algorithm where I computed
the real query/words indexes based and removed the invalid ones
before sending the documents to the bucket sort.
2019-12-13 14:38:22 +01:00
Clément Renault
4e91b31b1f
Make the Typo and Words work with synonyms
2019-12-13 14:38:22 +01:00
Clément Renault
f87c67fcad
Improve the QueryEnhancer by doing a single lookup
2019-12-13 14:38:22 +01:00
Clément Renault
902625601a
Work in progress: It seems like we support synonyms, split and concat words
2019-12-13 14:38:22 +01:00
Clément Renault
d17d4dc5ec
Add more debug infos
2019-12-13 14:38:21 +01:00
Clément Renault
ef6a4db182
Before improving fields AttrCount
...
Removing the fields_count fetching reduced by 2 times the serach time, we should look at lazily pulling them form the criterions in needs
ugly-test: Make the fields_count fetching lazy
Just before running the exactness criterion
2019-12-13 14:38:21 +01:00
Clément Renault
11f3d7782d
Introduce the AttrCount type
2019-12-13 14:38:21 +01:00
Clément Renault
951f0bcb10
sqaush-me: Improve benchmarks naming
2019-12-13 14:17:40 +01:00
Clément Renault
d8ba405baf
Add some criterion benchmarks to help mesure improvements
2019-12-13 14:17:40 +01:00
Quentin de Quelen
3a4130f344
Allow to index files with null or boolean
2019-12-12 19:25:05 +01:00
Quentin de Quelen
88b3c05155
Stop words; Do not reindex all documents if there is no documents
2019-12-12 15:31:39 +01:00
Quentin de Quelen
a4f26e8e48
Rewrite the synonym endpoint
2019-12-12 12:47:02 +01:00
Clément Renault
dc1849d291
Bump heed to 0.6.1
2019-12-07 11:49:45 +01:00
Clément Renault
29fd54dcfa
Allow users to send csv files from stdin in examples
2019-12-05 12:23:56 +01:00
Thomas Payet
51636402c2
Add debian package in CI
2019-12-04 18:02:30 +01:00
Clément Renault
4f87465f18
Bump meilisearch crates to v0.8.4
2019-12-03 17:22:45 +01:00
qdequele
773a51e7d0
Rename 'update_type' to 'type' on EnqueuedUpdateResult
2019-11-29 15:09:48 +01:00
qdequele
7923752513
Serialize updates results to camelCase
2019-11-29 15:05:54 +01:00
Clément Renault
30cb60f679
Bump meilisearch crates to v0.8.3
2019-11-29 14:06:17 +01:00
qdequele
3a90233a3d
Add status failed on UpdateStatus
2019-11-28 18:41:11 +01:00
Clément Renault
9a2b4d08e1
Bump meilisearch crates to v0.8.2
2019-11-28 17:15:13 +01:00
Clément Renault
1def56ea11
Change the update loop to be more explicit on index clear
2019-11-27 13:43:28 +01:00
Clément Renault
f6fb31c531
Bump meilisearch crates to v0.8.1
2019-11-27 11:47:27 +01:00
Clément Renault
d08b76a323
Separate the update and main databases
...
We used the heed typed transaction to make it safe (https://github.com/Kerollmops/heed/pull/27 ).
2019-11-27 11:29:06 +01:00
Clément Renault
7cc096e0a2
Rename MeiliDB into MeiliSearch
2019-11-26 11:12:30 +01:00
