Precision/Recall measures for a query using aggs

QueryBenchmarker

//================================
// Here a script for gathering the precision/recall stats for a query (see http://en.wikipedia.org/wiki/Precision_and_recall)
// A candidate classifier query's effectiveness is determined by counting hits on pre-classified content
// If we compute the F-measure we can potentially use it as the fitness function for a genetic algo that mutates our query
// (introducing phrases, minShouldMatch clauses etc) to move us towards our target goal of balancing precision/recall in our classifier.
//=================================

// Our candidate query for classifying documents in a category
var candidateQuery={ "terms": {"body":    ["vs", "shr", "cts", "net", "revs", "note", "loss", "mths", "shrs", "avg", "profit"]}};
// Our filter criteria for identifying documents in our target category
var categoryTest= { "term" : { "topics" :  "earn" } };
// The name of our category field (used for summarising false positives)
var categoryField="topics";

var queryJson={
	"query" :candidateQuery,
	"aggs" : {
        "globals":{
            "global":{},
            "aggs":{
                "requiredHits":{ 
                    // The count of docs in this bucket minus that of "truePositive" bucket gives our false negative figure
                    "filter" : categoryTest,  
                },
                "requiredMisses":{ 
                    // The count of docs in this bucket minus that of "falsePositive" bucket gives our true negative figure
                    "filter":{ "bool" : {"mustNot" : categoryTest } }
                }
            }
        },        
	"truePositive" : { 
             "filter" : categoryTest,            
             "aggs":{
                 //For documents with >1 classification this will summarise any  "also-connected" categories
                 "relatedCategories":{
                     "terms" : {"field" : categoryField} 
                 }                 
            }
        },
        "falsePositive":{
            "filter":{
                "bool" : { "mustNot" : categoryTest }
            },
             "aggs":{
                 //This will list all of the category fields we got a false positive on - identifies categories that are "close"
                 "fpCollateral":{"terms" : {"field" : categoryField} }                 
                 }
        }
    }                
};
return queryJson;