# Suggest feature
The suggest feature suggests similar looking terms based on a provided text by using a suggester. At the moment there the only supported suggester is `fuzzy`. The suggest feature is available since version `0.21.0`.
# Fuzzy suggester
The `fuzzy` suggester suggests terms based on edit distance. The provided suggest text is analyzed before terms are suggested. The suggested terms are provided per analyzed suggest text token. The `fuzzy` suggester doesn't take the query into account that is part of request.
# Suggest API
The suggest request part is defined along side the query part as top field in the json request.
```
curl -s -XPOST 'localhost:9200/_search' -d '{
"query" : {
...
},
"suggest" : {
...
}
}'
```
Several suggestions can be specified per request. Each suggestion is identified with an arbitary name. In the example below two suggestions are requested. The `my-suggest-1` suggestion uses the `body` field and `my-suggest-2` uses the `title` field. The `type` field is a required field and defines what suggester to use for a suggestion.
```
"suggest" : {
"suggestions" : {
"my-suggest-1" : {
"type" : "fuzzy",
"field" : "body",
"text" : "the amsterdma meetpu"
},
"my-suggest-2" : {
"type" : "fuzzy",
"field" : "title",
"text" : "the rottredam meetpu"
}
}
}
```
The below suggest response example includes the suggestions part for `my-suggest-1` and `my-suggest-2`. Each suggestion part contains a terms array, that contains all terms outputted by the analyzed suggest text. Each term object includes the term itself, the original start and end offset in the suggest text and if found an arbitary number of suggestions.
```
{
...
"suggest": {
"my-suggest-1": {
"terms" : [
{
"term" : "amsterdma",
"start_offset": 5,
"end_offset": 14,
"suggestions": [
...
]
}
...
]
},
"my-suggest-2" : {
"terms" : [
...
]
}
}
```
Each suggestions array contains a suggestion object that includes the suggested term, its document frequency and score compared to the suggest text term. The meaning of the score depends on the used suggester. The fuzzy suggester's score is based on the edit distance.
```
"suggestions": [
{
"term": "amsterdam",
"frequency": 77,
"score": 0.8888889
},
...
]
```
# Global suggest text
To avoid repitition of the suggest text, it is possible to define a global text. In the example below the suggest text is a global option and applies to the `my-suggest-1` and `my-suggest-2` suggestions.
```
"suggest" : {
"suggestions" : {
"text" : "the amsterdma meetpu",
"my-suggest-1" : {
"type" : "fuzzy",
"field" : "title"
},
"my-suggest-2" : {
"type" : "fuzzy",
"field" : "body"
}
}
}
```
The suggest text can be specied as global option or as suggestion specific option. The suggest text specified on suggestion level override the suggest text on the global level.
# Other suggest example.
In the below example we request suggestions for the following suggest text: `devloping distibutd saerch engies` on the `title` field with a maximum of 3 suggestions per term inside the suggest text. Note that in this example we use the `count` search type. This isn't required, but a nice optimalization. The suggestions are gather in the `query` phase and in the case that we only care about suggestions (so no hits) we don't need to execute the `fetch` phase.
```
curl -s -XPOST 'localhost:9200/_search?search_type=count' -d '{
"suggest" : {
"suggestions" : {
"my-title-suggestions" : {
"suggester" : "fuzzy",
"field" : "title",
"text" : "devloping distibutd saerch engies",
"size" : 3
}
}
}
}'
```
The above request could yield the response as stated in the code example below. As you can see if we take the first suggested term of each suggest text term we get `developing distributed search engines` as result.
```
{
...
"suggest": {
"my-title-suggestions": {
"terms": [
{
"term": "devloping",
"start_offset": 0,
"end_offset": 9,
"suggestions": [
{
"term": "developing",
"frequency": 77,
"score": 0.8888889
},
{
"term": "deloping",
"frequency": 1,
"score": 0.875
},
{
"term": "deploying",
"frequency": 2,
"score": 0.7777778
}
]
},
{
"term": "distibutd",
"start_offset": 10,
"end_offset": 19,
"suggestions": [
{
"term": "distributed",
"frequency": 217,
"score": 0.7777778
},
{
"term": "disributed",
"frequency": 1,
"score": 0.7777778
},
{
"term": "distribute",
"frequency": 1,
"score": 0.7777778
}
]
},
{
"term": "saerch",
"start_offset": 20,
"end_offset": 26,
"suggestions": [
{
"term": "search",
"frequency": 1038,
"score": 0.8333333
},
{
"term": "smerch",
"frequency": 3,
"score": 0.8333333
},
{
"term": "serch",
"frequency": 2,
"score": 0.8
}
]
},
{
"term": "engies",
"start_offset": 27,
"end_offset": 33,
"suggestions": [
{
"term": "engines",
"frequency": 568,
"score": 0.8333333
},
{
"term": "engles",
"frequency": 3,
"score": 0.8333333
},
{
"term": "eggies",
"frequency": 1,
"score": 0.8333333
}
]
}
]
}
}
...
}
```
# Common suggest options:
* `suggester` - The suggester implementation type. The only supported value is 'fuzzy'. This is a required option.
* `text` - The suggest text. The suggest text is a required option that needs to be set globally or per suggestion.
# Common fuzzy suggest options
* `field` - The field to fetch the candidate suggestions from. This is an required option that either needs to be set globally or per suggestion.
* `analyzer` - The analyzer to analyse the suggest text with. Defaults to the search analyzer of the suggest field.
* `size` - The maximum corrections to be returned per suggest text token.
* `sort` - Defines how suggestions should be sorted per suggest text term. Two possible value:
** `score` - Sort by sore first, then document frequency and then the term itself.
** `frequency` - Sort by document frequency first, then simlarity score and then the term itself.
* `suggest_mode` - The suggest mode controls what suggestions are included or controls for what suggest text terms, suggestions should be suggested. Three possible values can be specified:
** `missing` - Only suggest terms in the suggest text that aren't in the index. This is the default.
** `popular` - Only suggest suggestions that occur in more docs then the original suggest text term.
** `always` - Suggest any matching suggestions based on terms in the suggest text.
# Other fuzzy suggest options:
* `lowercase_terms` - Lower cases the suggest text terms after text analyzation.
* `max_edits` - The maximum edit distance candidate suggestions can have in order to be considered as a suggestion. Can only be a value between 1 and 2. Any other value result in an bad request error being thrown. Defaults to 2.
* `min_prefix` - The number of minimal prefix characters that must match in order be a candidate suggestions. Defaults to 1. Increasing this number improves spellcheck performance. Usually misspellings don't occur in the beginning of terms.
* `min_query_length` - The minimum length a suggest text term must have in order to be included. Defaults to 4.
* `shard_size` - Sets the maximum number of suggestions to be retrieved from each individual shard. During the reduce phase only the top N suggestions are returned based on the `size` option. Defaults to the `size` option. Setting this to a value higher than the `size` can be useful in order to get a more accurate document frequency for spelling corrections at the cost of performance. Due to the fact that terms are partitioned amongst shards, the shard level document frequencies of spelling corrections may not be precise. Increasing this will make these document frequencies more precise.
* `max_inspections` - A factor that is used to multiply with the `shards_size` in order to inspect more candidate spell corrections on the shard level. Can improve accuracy at the cost of performance. Defaults to 5.
* `threshold_frequency` - The minimal threshold in number of documents a suggestion should appear in. This can be specified as an absolute number or as a relative percentage of number of documents. This can improve quality by only suggesting high frequency terms. Defaults to 0f and is not enabled. If a value higher than 1 is specified then the number cannot be fractional. The shard level document frequencies are used for this option.
* `max_query_frequency` - The maximum threshold in number of documents a sugges text token can exist in order to be included. Can be a relative percentage number (e.g 0.4) or an absolute number to represent document frequencies. If an value higher than 1 is specified then fractional can not be specified. Defaults to 0.01f. This can be used to exclude high frequency terms from being spellchecked. High frequency terms are usually spelled correctly on top of this this also improves the spellcheck performance. The shard level document frequencies are used for this option.
Closes#2585
* added configurable MemoryIndexPool that pools MemoryIndex instance across Threads
* Pool can be configured based on the number of pooled instances as well as the maximum number of bytes that is reused across the pooled instances
Closes#2581
* Removed CustmoMemoryIndex in favor of MemoryIndex which as of 4.1 supports adding the same field twice
* Replaced duplicated logic in X[*]FSDirectory for rate limiting with a RateLimitedFSDirectory wrapper
* Remove hacks to find out merge context in rate limiting in favor of IOContext
* replaced Scorer#freq() return type (from float to int)
* Upgraded FVHighlighter to new 'centered' highlighting
* Fixed RobinEngine to use seperate setCommitData
* Default ShardsAllocator is set to BalancedShardsAllocator
* Core ShardsAllocator implementations can be defined via 'cluster.routing.allocation.type'
* Core ShardsAllocator implementations are exposed via short keys 'balanced' (BalancedShardsAllocator) and 'even_shards' (EvenShardsCountAllocator)
* Third party allocators can be loaded via fully-qualified class names.
Closes#2557
* Weights are calculated per index and incorporate index level, global and primary related parameters
* Balance operations are executed based on a win maximation strategy that tries to relocate shards
first that offer the biggest gain towards the weight functions optimum
* The WeightFunction allows settings to prefer index based balance over global balance and vice versa
* Balance operations can be throttled by raising a threshold resulting in less agressive balance operations
* WeightFunction shipps with defaults to achive evenly distributed indexes while maintaining a global balance
Closes#2555
checked jackson, there won't be an overhead in enabling comments. Added, with the caveat that when used with mappings, and calling "get mapping", the comments will not be returned
closes#1394
we need to in order to properly handle bytes, and normalize Integer to Long for example for consistency, the fact that mappers now handle different Objtes help here
Closes: #646
- Introduced HunspellService which holds a repository of hunspell dictionaries
- It is possible to register a dictionary via a plugin or by placing the dictionary files on the file system
Added score support to `has_child` and `has_parent` queries. Both queries support a score_type option. The has_child support the same options as the top_children query and the none option which is the default and yields the current behaviour. The has_parent query support the score type options: score and none. The latter is the default and yields the current behaviour.
If the score_type is set to a value other than none then the has_parent query map the matched parent score into the related children documents. The has_child query then map the matched children documents into the related parent document. The score_type on both queries defines how the children documents scores are mapped in the parent documents. Both queries are executed in two phases. First phase collects the parent uid values of matching documents with an aggregated score per parent uid value. In the second phase either child or parent typed documents are emitted as hit that have the same parent uid value as found during the first phase. The score computed in the first phase will be used as score.
Closes#2502
since we add them internally to the compound mappers, we need to publish the fact, otherwise, for example, the codec won't find the relevant one based on the global mapper service
Fixed error with the top_children query when `DFS_QUERY_*` is used as search_type and wraps a query that gets rewritten (E.g wildcard query).
Closes#2501
If the a routing value isn't id based, the get part of the mlt request couldn't retrieve the document for the second part of the mlt request and a 500 code is returned instead. This fix addresses this issue.
Closes#2489
- Added "regexp" query type (based on Lucene 4 RegexpQuery)
- Added "regexp" filter type
- Fixed a bug in IdFieldMapper where prefixQuery on a single type would be redundantly wrapped in a boolean query
our idea is to apply it on the "filtered/constant" level, and not on compound filters, so we won't apply it multiple times. The solution is conservative a bit now, we can further optimize it in the future, for example, not to wrap it when no caching is done within the filter chain
- don't check no null for liveDocs, since we know they are not null with the check for hasDeletion
- improve iteration over liveDocs vs. innerSet, prefer to iterate over the faster one
- remove the DocSet abstraction, and use Bits where we can by getting it from DocIdSet
- better handling of acceptDocs, though still need to properly apply them when caching is involved
This feature adds the option to configure a `PostingsFormat` and assign it to a field in the mapping. This feature is very expert and in almost all cases Elasticsearch's defaults will suite your needs.
## Configuring a postingsformat per field
There're several default postings formats configured by default which can be used in your mapping:
a* `direct` - A codec that wraps the default postings format during write time, but loads the terms and postinglists into memory directly in memory during read time as raw arrays. This postings format is exceptional memory intensive, but can give a substantial increase in search performance.
* `memory` - A codec that loads and stores terms and postinglists in memory using a FST. Acts like a cached postingslist.
* `bloom_default` - Maintains a bloom filter for the indexed terms, which is stored to disk and builds on top of the `default` postings format. This postings format is useful for low document frequency terms and offers a fail fast for seeks to terms that don't exist.
* `bloom_pulsing` - Similar to the `bloom_default` postings format, but builds on top of the `pulsing` postings format.
* `default` - The default postings format. The default if none is specified.
On all fields it possible to configure a `postings_format` attribute. Example mapping:
```
{
"person" : {
"properties" : {
"second_person_id" : {"type" : "string", "postings_format" : "pulsing"}
}
}
}
```
## Configuring a custom postingsformat
It is possible the instantiate custom postingsformats. This can be specified via the index settings.
```
{
"codec" : {
"postings_format" : {
"my_format" : {
"type" : "pulsing40"
"freq_cut_off" : "5"
}
}
}
}
```
In the above example the `freq_cut_off` is set the 5 (defaults to 1). This tells the pulsing postings format to inline the postinglist of terms with a document frequency lower or equal to 5 in the term dictionary.
Closes#2411
no need to test for boost, we already have specific boost tests, in general, we should get rid of this test, and use more specialized tests if we are missing some
I am not completely sure about this one, but it reduces the number of failing tests from 98 to 31 so I am going to check it in. Please, review and fix it, if there is a better solution.
Because of change in Lucene 4.0, ContextIndexSearcher was bypassed and elasticsearch filters and collectors were ignored.
In lucene 3.6 the stack of Searcher search calls looked like this:
search(Query query, int n)
search(Query query, Filter filter, int n)
search(Weight weight, Filter filter, int nDocs)
search(Weight weight, Filter filter, ScoreDoc after, int nDocs)
search(Weight weight, Filter filter, Collector collector) <-- this is ContextIndexSearcher was injecting combined filter and collector
search(Weight weight, Filter filter, Collector collector)
In Lucene 4.0 the stack looks like this:
search(Query query, int n)
search(Query query, Filter filter, int n) <-- here lucene wraps Query and Filter into Weight
search(Weight weight, ScoreDoc after, int nDocs)
search(List<AtomicReaderContext> leaves, Weight weight, ScoreDoc after, int nDocs)
search(List<AtomicReaderContext> leaves, Weight weight, Collector collector)
...
In other words, when we have Filter, we don't have a Collector yet, but when we have Collector, Filter is already wrapped inside Weight. The only way to fix for the problem that I could think of is by introducing two injection points: one for Filters and another one for Collectors:
search(Query query, int n)
search(Query query, Filter filter, int n) <-- here combined Filters are injected
search(Weight weight, ScoreDoc after, int nDocs)
search(List<AtomicReaderContext> leaves, Weight weight, ScoreDoc after, int nDocs)
search(List<AtomicReaderContext> leaves, Weight weight, Collector collector) <-- here Collectors are injected
Similar problem existed for count(), so I had to override search(Query query, Collector results) as well.
This commit enables setting boost for numeric fields. However, there is still no way to take advantage of boosted numeric fields during searching because all queries against numeric fields are translated into range queries wrapped in ConstantScore. Boost for numeric fields is broken on master as well https://gist.github.com/7ecedea4f6a5219efb89
Shay Banon
Martijn van Groningen
Simon Willnauer
Daniel Muller
synhershko
Igor Motov
uboness
Clinton Gormley
Dave Brosius
Alex Lambert
Chris Male
David Pilato
Nicholas Tung
uboness