CODEWITHSUNDEEP
elasticsearch
Setsuna
Setsuna

Reputation: 2141

How to facet on one field then another in elasticsearch?

I have an ES document with the following fields "buydatefield", "itemboughtfield" among others.

How do I make an ES query such that I can get a facet on date, then item bought?

{
"query":{"match_all":{}
},
"facets": {
  "buydateFacet": {
    "terms": {
      "field": "buydatefield",
      "all_terms": true
    }
  },
  "itemboughtFacet": {
    "terms": {
      "field": "itemboughtfield",
      "all_terms": true
    }

  }
}

}

The above returns two separate facets for buyFacet and itemboughtFacet. What I want is to get "subfacets" where for each date, there is a nested count of all the "items bought" for that date. Is this possible? If so, how?

I would like some output that is for example:

terms: [{
  term: "Bannana",
  // total: 11 bannanas
  buydates:{
     // 5/31/2013 bought 5 bannana
     // 6/2/2013 bought 6 bannana
  }
},
{
  term: "Apple",
  // total: 3 apples
  buydates:{
     // 5/30/2013 bought 2 apple
     // 6/1/2013 bought 1 apple
  }
},

]

ALso, is it possible to specify a date range for facet?

Upvotes: 2

Views: 1940

Answers (2)

Tikitu
Tikitu

Reputation: 699

If buydatefield is indexed as a date field you could use the FacetedDateHistogram from elasticfacets (disclosure: author is my former lead). It gives you a "two-level" facet: top-level is equivalent to the builtin date histogram, but within each bucket you get to put any other facet you like, which operates only on the values in that bucket (here you would term-facet on item bought).

This won't give you quite what you specify in your example, but instead:

"buydatefacet": {
  "_type": "faceted_date_histogram",
  "entries": [
    { 
      "time": 1356994800000, // buy date 25/01/2013
      "facet": {
        "_type": "terms",
        "terms": [
           { "term": "apple", "count": 3 },
           { "term": "banana", "count": 1 }
        ],
        "missing": 0,
        "total": 4,
        "other": 0
      }
    },
    { ... more days here ... }
  ]
 }

When ES 1.0 hits there will be some kind of more general built-in support for nesting facets in this way, not limited to date histogram at top-level (they're renaming the notion to "Aggregations").

Upvotes: 1

Transact Charlie
Transact Charlie

Reputation: 2203

On the first question -- the sub faceting:

in Solr 4 it's called Facet Pivoting -- last I checked it didn't work in a clustered configuration though.

I believe that it's part of the Lucene4 specification which ES just moved on to for 0.9.

It's an often requested feature: example: http://elasticsearch-users.115913.n3.nabble.com/Pivot-facets-td2981519.html

However pivot faceting tends to be pretty slow.

For your use case you could also add a field that is the two terms concatenated together with a character (pipe |) between and then facet on that field -- then in your front end parse the hirerarchy and display to users.

However what you are doing is increasing the number of unique entires substancially and this will hurt performance.

Upvotes: 1

Related Questions