This reverts commit 12fb509d84.
We revert this commit because it's causing the bug #150.
The initial algorithm we implemented for the stop_words was:
1. remove the stop_words from the dataset
2. keep the stop_words in the query to see if we can generate new words by
integrating typos or if the word was a prefix
=> This was causing the bug since, in the case of “The hobbit”, we were
**always** looking for something starting with “t he” or “th e”
instead of ignoring the word completely.
For now we are going to fix the bug by completely ignoring the
stop_words in the query.
This could cause another problem were someone mistyped a normal word and
ended up typing a stop_word.
For example imagine someone searching for the music “Won't he do it”.
If that person misplace one space and write “Won' the do it” then we
will loose a part of the request.
One fix would be to update our query tree to something like that:
---------------------
OR
OR
TOLERANT hobbit # the first option is to ignore the stop_word
AND
CONSECUTIVE # the second option is to do as we are doing
EXACT t # currently
EXACT he
TOLERANT hobbit
---------------------
This would increase drastically the size of our query tree on request
with a lot of stop_words. For example think of “The Lord Of The Rings”.
For now whatsoever we decided we were going to ignore this problem and consider
that it doesn't reduce too much the relevancy of the search to do that
while it improves the performances.
fixes after review
bump the version of the tokenizer
implement a first version of the stop_words
The front must provide a BTreeSet containing the stop words
The stop_words are set at None if an empty Set is provided
add the stop-words in the http-ui interface
Use maplit in the test
and remove all the useless drop(rtxn) at the end of all tests
Integrate the stop_words in the querytree
remove the stop_words from the querytree except if it was a prefix or a typo
more fixes after review
104: Update all the response format (issue #64) r=MarinPostma a=irevoire
closes#64
Co-authored-by: Irevoire <tamo@meilisearch.com>
Co-authored-by: tamo <tamo@meilisearch.com>
113: snapshots r=MarinPostma a=MarinPostma
This pr adds support for snapshoting.
The snapshoting process for an index requires that no other update is processing at the same time. A mutex lock has been added to prevent a snapshot from occuring at the same time as an update, while still premitting updates to be pushed.
The list of the indexes to snapshot is first retrieved from the `UuidResolver` which also performs its snapshot.
This list is passed to the update store, which attempts to acquire a lock on the update store while it snaphots itself and it's associated index store.
This means that a snapshot can only be completed once all indexes have finished their ongoing update.
This pr also adds refactoring of the code to allow unit testing and mocking, and unit test the snapshot creation.
Co-authored-by: mpostma <postma.marin@protonmail.com>
Co-authored-by: tamo <irevoire@protonmail.ch>
Co-authored-by: marin <postma.marin@protonmail.com>
Co-authored-by: Marin Postma <postma.marin@protonmail.com>
The front must provide a BTreeSet containing the stop words
The stop_words are set at None if an empty Set is provided
add the stop-words in the http-ui interface
Use maplit in the test
and remove all the useless drop(rtxn) at the end of all tests
