SERVER-105543: BACKPORT-25609: [v8.2] Strengthen rankFusion scoreDetails testing (#39319)
GitOrigin-RevId: 3a9d05edd8548be468cee1fadf033c9ab8c30d10
This commit is contained in:
Adithi Raghavan
MongoDB Bot
parent
8a3d441ee9
commit
812035b596
@ -13,6 +13,7 @@ import {
|
||||
getMovieSearchIndexSpec,
|
||||
getMovieVectorSearchIndexSpec
|
||||
} from "jstests/with_mongot/e2e_lib/data/movies.js";
|
||||
import {getRentalData, getRentalSearchIndexSpec} from "jstests/with_mongot/e2e_lib/data/rentals.js";
|
||||
|
||||
const collName = "search_rank_fusion";
|
||||
const coll = db.getCollection(collName);
|
||||
@ -51,28 +52,106 @@ const searchStage = {
|
||||
$search: searchStageSpec
|
||||
};
|
||||
|
||||
const searchStageSpecNoDetails = {
|
||||
index: getMovieSearchIndexSpec().name,
|
||||
text: {query: "ape", path: ["fullplot", "title"]},
|
||||
scoreDetails: false
|
||||
};
|
||||
|
||||
const searchStageNoDetails = {
|
||||
$search: searchStageSpecNoDetails
|
||||
};
|
||||
|
||||
const calculateReciprocalRankFusionScore = (weight, rank) => {
|
||||
return (weight * (1 / (60 + rank)));
|
||||
};
|
||||
|
||||
const scoreDetailsDescription =
|
||||
"value output by reciprocal rank fusion algorithm, computed as sum of (weight * (1 / (60 " +
|
||||
"+ rank))) across input pipelines from which this document is output, from:";
|
||||
|
||||
// Test search/vectorSearch where only search has scoreDetails.
|
||||
let testQuery = [
|
||||
{
|
||||
$rankFusion: {
|
||||
input: {pipelines: {vector: [vectorStage], search: [searchStage, {$limit: limit}]}},
|
||||
combination: {weights: {search: 2}},
|
||||
scoreDetails: true,
|
||||
},
|
||||
},
|
||||
{$project: {score: {$meta: "score"}, details: {$meta: "scoreDetails"}}}
|
||||
];
|
||||
let results = coll.aggregate(testQuery).toArray();
|
||||
|
||||
function fieldPresent(field, containingObj) {
|
||||
return containingObj.hasOwnProperty(field);
|
||||
/**
|
||||
* All input pipelines should contain the following fields when $rankFusion's scoreDetails is
|
||||
* enabled: inputPipelineName, rank, and weight. Only inputPipelineName's and weight's values are
|
||||
* constant across the results.
|
||||
*/
|
||||
function checkDefaultPipelineScoreDetails(assertFieldPresent, subDetails, pipelineName, weight) {
|
||||
assertFieldPresent("inputPipelineName", subDetails);
|
||||
assert.eq(subDetails["inputPipelineName"], pipelineName);
|
||||
assertFieldPresent("rank", subDetails);
|
||||
assertFieldPresent("weight", subDetails);
|
||||
assert.eq(subDetails["weight"], weight);
|
||||
}
|
||||
|
||||
for (const foundDoc of results) {
|
||||
/**
|
||||
* Checks the scoreDetails (inputPipelineName, rank, weight) for a search input pipeline. If a
|
||||
* document was ouput from the input pipeline (value field is present in scoreDetails), then check
|
||||
* that the value and details fields are present. If the search input pipeline has scoreDetails
|
||||
* enabled, check the description field is accurate and that the pipeline's scoreDetails aren't
|
||||
* empty. Returns the RRF score for this input pipeline.
|
||||
*/
|
||||
function checkSearchScoreDetails(
|
||||
assertFieldPresent, subDetails, pipelineName, weight, isScoreDetails) {
|
||||
assertFieldPresent("inputPipelineName", subDetails);
|
||||
assert.eq(subDetails["inputPipelineName"], pipelineName);
|
||||
assertFieldPresent("rank", subDetails);
|
||||
// If there isn't a value, we didn't get this back from search at all.
|
||||
let searchScore = 0;
|
||||
if (subDetails.hasOwnProperty("value")) {
|
||||
assertFieldPresent("weight", subDetails);
|
||||
assert.eq(subDetails["weight"], weight);
|
||||
assertFieldPresent("value", subDetails); // Output of rank calculation.
|
||||
assertFieldPresent("details", subDetails);
|
||||
if (isScoreDetails) {
|
||||
assertFieldPresent("description", subDetails);
|
||||
assert.eq(subDetails["description"], "sum of:");
|
||||
// Note we won't check the shape of the search scoreDetails beyond here.
|
||||
assert.neq(subDetails["details"], []);
|
||||
} else {
|
||||
assert.eq(subDetails["details"], []);
|
||||
}
|
||||
searchScore = calculateReciprocalRankFusionScore(subDetails["weight"], subDetails["rank"]);
|
||||
} else {
|
||||
assert.eq(subDetails["rank"], "NA");
|
||||
}
|
||||
return searchScore;
|
||||
}
|
||||
|
||||
/**
|
||||
* Checks the scoreDetails (inputPipelineName, rank, weight, details) for a vectorSearch input
|
||||
* pipeline. Note that vectorSearch input pipeline do not have scoreDetails so the details field
|
||||
* should always be an empty array. Returns the RRF score for this input pipeline.
|
||||
*/
|
||||
function checkVectorScoreDetails(assertFieldPresent, subDetails, pipelineName, weight) {
|
||||
checkDefaultPipelineScoreDetails(assertFieldPresent, subDetails, pipelineName, weight);
|
||||
assertFieldPresent("value", subDetails); // Original 'score' AKA vectorSearchScore.
|
||||
assertFieldPresent("details", subDetails);
|
||||
assert.eq(subDetails["details"], []);
|
||||
const vectorSearchScore =
|
||||
calculateReciprocalRankFusionScore(subDetails["weight"], subDetails["rank"]);
|
||||
return vectorSearchScore;
|
||||
}
|
||||
|
||||
/**
|
||||
* Checks the scoreDetails (inputPipelineName, rank, weight, details) for a geoNear input
|
||||
* pipeline. Note that geoNear input pipeline do not have scoreDetails so the details field
|
||||
* should always be an empty array. Returns the RRF score for this input pipeline.
|
||||
*/
|
||||
function checkGeoNearScoreDetails(assertFieldPresent, subDetails, pipelineName, weight) {
|
||||
checkDefaultPipelineScoreDetails(assertFieldPresent, subDetails, pipelineName, weight);
|
||||
assertFieldPresent("details", subDetails);
|
||||
assert.eq(subDetails["details"], []);
|
||||
const geoNearScore =
|
||||
calculateReciprocalRankFusionScore(subDetails["weight"], subDetails["rank"]);
|
||||
return geoNearScore;
|
||||
}
|
||||
|
||||
/**
|
||||
* For each document or result, check the follwoing fields in the outer scoreDetails: score,
|
||||
* details, value, description, and that the subDetails array contains two entries, 1 for each input
|
||||
* pipeline.
|
||||
*/
|
||||
function checkOuterScoreDetails(foundDoc, numInputPipelines) {
|
||||
// Assert that the score metadata has been set.
|
||||
assert(fieldPresent("score", foundDoc), foundDoc);
|
||||
const score = foundDoc["score"];
|
||||
@ -93,229 +172,391 @@ for (const foundDoc of results) {
|
||||
// Description of rank fusion. Wrapper on both search / vector.
|
||||
assertFieldPresent("details", details);
|
||||
const subDetails = details["details"];
|
||||
assert.eq(subDetails.length, 2);
|
||||
assert.eq(subDetails.length, numInputPipelines);
|
||||
|
||||
const searchDetails = subDetails[0];
|
||||
assertFieldPresent("inputPipelineName", searchDetails);
|
||||
assert.eq(searchDetails["inputPipelineName"], "search");
|
||||
assertFieldPresent("rank", searchDetails);
|
||||
// If there isn't a value, we didn't get this back from search at all.
|
||||
if (searchDetails.hasOwnProperty("value")) {
|
||||
assertFieldPresent("weight", searchDetails);
|
||||
assert.eq(searchDetails["weight"], 2);
|
||||
assertFieldPresent("value", searchDetails); // Output of rank calculation.
|
||||
assertFieldPresent("details",
|
||||
searchDetails); // Not checking description contents, just that its
|
||||
// present and not our placeholder value.
|
||||
assert.neq(searchDetails["details"], []);
|
||||
// Note we won't check the shape of the search scoreDetails beyond here.
|
||||
} else {
|
||||
assert.eq(searchDetails["rank"], "NA");
|
||||
assert.eq(searchDetails.hasOwnProperty("weight"), false);
|
||||
assert.eq(searchDetails.hasOwnProperty("details"), false);
|
||||
}
|
||||
|
||||
const vectorDetails = subDetails[1];
|
||||
assertFieldPresent("inputPipelineName", vectorDetails);
|
||||
assert.eq(vectorDetails["inputPipelineName"], "vector");
|
||||
assertFieldPresent("details", vectorDetails);
|
||||
assert.eq(vectorDetails["details"], []);
|
||||
assertFieldPresent("rank", vectorDetails);
|
||||
assertFieldPresent("weight", vectorDetails);
|
||||
assert.eq(vectorDetails["weight"], 1);
|
||||
return [assertFieldPresent, subDetails, score];
|
||||
}
|
||||
|
||||
// Test vectorSearch/vectorSearch where neither has score details.
|
||||
testQuery = [
|
||||
{
|
||||
$rankFusion: {
|
||||
input: {pipelines: {vector: [vectorStage], secondVector: [vectorStage]}},
|
||||
combination: {weights: {vector: 0.5, secondVector: 2.8}},
|
||||
scoreDetails: true,
|
||||
},
|
||||
},
|
||||
{$project: {score: {$meta: "score"}, details: {$meta: "scoreDetails"}}}
|
||||
];
|
||||
results = coll.aggregate(testQuery).toArray();
|
||||
|
||||
for (const foundDoc of results) {
|
||||
// Assert that the score metadata has been set.
|
||||
assert(fieldPresent("score", foundDoc), foundDoc);
|
||||
const score = foundDoc["score"];
|
||||
assert(fieldPresent("details", foundDoc), foundDoc);
|
||||
const details = foundDoc["details"];
|
||||
assert(fieldPresent("value", details), details);
|
||||
// The output of the rank calculation.
|
||||
// We don't care about the actual score, just assert that its been calculated.
|
||||
assert.gt(details["value"], 0);
|
||||
// Assert that the score metadata is the same value as what scoreDetails set.
|
||||
assert.eq(details["value"], score);
|
||||
assert(fieldPresent("description", details), details);
|
||||
assert.eq(details["description"], scoreDetailsDescription);
|
||||
|
||||
function assertFieldPresent(field, obj) {
|
||||
assert(fieldPresent(field, obj),
|
||||
`Looked for ${field} in ${tojson(obj)}. Full details: ${tojson(details)}`);
|
||||
}
|
||||
// Description of rank fusion. Wrapper on both secondVector / vector.
|
||||
assertFieldPresent("details", details);
|
||||
const subDetails = details["details"];
|
||||
assert.eq(subDetails.length, 2);
|
||||
|
||||
const secondVectorDetails = subDetails[0];
|
||||
assertFieldPresent("inputPipelineName", secondVectorDetails);
|
||||
assert.eq(secondVectorDetails["inputPipelineName"], "secondVector");
|
||||
assertFieldPresent("rank", secondVectorDetails);
|
||||
assertFieldPresent("weight", secondVectorDetails);
|
||||
assert.eq(secondVectorDetails["weight"], 2.8);
|
||||
assertFieldPresent("value", secondVectorDetails); // Original 'score' AKA vectorSearchScore.
|
||||
assertFieldPresent("details",
|
||||
secondVectorDetails); // Not checking description contents, just that its
|
||||
// present and not our placeholder value.
|
||||
assert.eq(secondVectorDetails["details"], []);
|
||||
|
||||
const vectorDetails = subDetails[1];
|
||||
assertFieldPresent("inputPipelineName", vectorDetails);
|
||||
assert.eq(vectorDetails["inputPipelineName"], "vector");
|
||||
assertFieldPresent("details", vectorDetails);
|
||||
assert.eq(vectorDetails["details"], []);
|
||||
assertFieldPresent("value", vectorDetails); // Original 'score' AKA vectorSearchScore.
|
||||
assertFieldPresent("rank", vectorDetails);
|
||||
assertFieldPresent("weight", vectorDetails);
|
||||
assert.eq(vectorDetails["weight"], 0.5);
|
||||
function fieldPresent(field, containingObj) {
|
||||
return containingObj.hasOwnProperty(field);
|
||||
}
|
||||
|
||||
// Test search/vectorSearch where search scoreDetails is off but $rankFusion's scoreDetails is on.
|
||||
const searchStageSpecNoDetails = {
|
||||
index: getMovieSearchIndexSpec().name,
|
||||
text: {query: "ape", path: ["fullplot", "title"]},
|
||||
scoreDetails: false
|
||||
};
|
||||
|
||||
const searchStageNoDetails = {
|
||||
$search: searchStageSpecNoDetails
|
||||
};
|
||||
|
||||
testQuery = [
|
||||
{
|
||||
$rankFusion: {
|
||||
input: {
|
||||
pipelines: {vector: [vectorStage], search: [searchStageNoDetails, {$limit: limit}]}
|
||||
/**
|
||||
* Test search/vectorSearch where only search has scoreDetails.
|
||||
* "score" : 0.04918032786885246,
|
||||
* "details" : {
|
||||
"value" : 0.04918032786885246,
|
||||
"description" : "value output by reciprocal rank fusion algorithm...",
|
||||
"details" : [
|
||||
{
|
||||
"inputPipelineName" : "search",
|
||||
"rank" : 1,
|
||||
"weight" : 2,
|
||||
"value" : 1.5521023273468018,
|
||||
"description" : "sum of:",
|
||||
"details" : [
|
||||
// $search scoreDetails go here
|
||||
]
|
||||
},
|
||||
scoreDetails: true,
|
||||
},
|
||||
},
|
||||
{$project: {score: {$meta: "score"}, details: {$meta: "scoreDetails"}}}
|
||||
];
|
||||
|
||||
results = coll.aggregate(testQuery).toArray();
|
||||
for (const foundDoc of results) {
|
||||
// Assert that the score metadata has been set.
|
||||
assert(fieldPresent("score", foundDoc), foundDoc);
|
||||
const score = foundDoc["score"];
|
||||
assert(fieldPresent("details", foundDoc), foundDoc);
|
||||
const details = foundDoc["details"];
|
||||
assert(fieldPresent("value", details), details);
|
||||
// We don't care about the actual score, just assert that its been calculated.
|
||||
assert.gt(details["value"], 0);
|
||||
// Assert that the score metadata is the same value as what scoreDetails set.
|
||||
assert.eq(details["value"], score);
|
||||
assert(fieldPresent("description", details), details);
|
||||
assert.eq(details["description"], scoreDetailsDescription);
|
||||
|
||||
function assertFieldPresent(field, obj) {
|
||||
assert(fieldPresent(field, obj),
|
||||
`Looked for ${field} in ${tojson(obj)}. Full details: ${tojson(details)}`);
|
||||
{
|
||||
"inputPipelineName" : "vector",
|
||||
"rank" : 1,
|
||||
"weight" : 1,
|
||||
"value" : 1,
|
||||
"details" : [ ]
|
||||
}
|
||||
]
|
||||
}
|
||||
// Description of rank fusion. Wrapper on both search / vector.
|
||||
assertFieldPresent("details", details);
|
||||
const subDetails = details["details"];
|
||||
assert.eq(subDetails.length, 2);
|
||||
|
||||
const searchDetails = subDetails[0];
|
||||
assertFieldPresent("inputPipelineName", searchDetails);
|
||||
assert.eq(searchDetails["inputPipelineName"], "search");
|
||||
assertFieldPresent("rank", searchDetails);
|
||||
// If there isn't a value, we didn't get this back from search at all.
|
||||
if (searchDetails.hasOwnProperty("value")) {
|
||||
assertFieldPresent("weight", searchDetails);
|
||||
assert.eq(searchDetails["weight"], 1);
|
||||
assertFieldPresent("value", searchDetails); // Output of rank calculation.
|
||||
assertFieldPresent("details", searchDetails);
|
||||
assert.eq(searchDetails["details"], []);
|
||||
// Note we won't check the shape of the search scoreDetails beyond here.
|
||||
} else {
|
||||
assert.eq(searchDetails["rank"], "NA");
|
||||
assert.eq(searchDetails.hasOwnProperty("weight"), false);
|
||||
assert.eq(searchDetails.hasOwnProperty("details"), false);
|
||||
}
|
||||
|
||||
const vectorDetails = subDetails[1];
|
||||
assertFieldPresent("inputPipelineName", vectorDetails);
|
||||
assert.eq(vectorDetails["inputPipelineName"], "vector");
|
||||
assertFieldPresent("details", vectorDetails);
|
||||
assert.eq(vectorDetails["details"], []);
|
||||
assertFieldPresent("rank", vectorDetails);
|
||||
assertFieldPresent("weight", vectorDetails);
|
||||
assert.eq(vectorDetails["weight"], 1);
|
||||
}
|
||||
|
||||
// Test search/vectorSearch where search scoreDetails is off and $rankFusion's scoreDetails is off.
|
||||
testQuery = [
|
||||
{
|
||||
$rankFusion: {
|
||||
input: {
|
||||
pipelines: {vector: [vectorStage], search: [searchStageNoDetails, {$limit: limit}]}
|
||||
*/
|
||||
(function testSearchScoreDetailsWithRankFusionScoreDetailsTwoInputPipelines() {
|
||||
const testQuery = [
|
||||
{
|
||||
$rankFusion: {
|
||||
input: {pipelines: {vector: [vectorStage], search: [searchStage, {$limit: limit}]}},
|
||||
combination: {weights: {search: 2}},
|
||||
scoreDetails: true,
|
||||
},
|
||||
scoreDetails: false,
|
||||
},
|
||||
},
|
||||
{$project: {score: {$meta: "score"}}}
|
||||
];
|
||||
{$project: {score: {$meta: "score"}, details: {$meta: "scoreDetails"}}}
|
||||
];
|
||||
|
||||
results = coll.aggregate(testQuery).toArray();
|
||||
for (const foundDoc of results) {
|
||||
// Assert that the score metadata has been set.
|
||||
assert(fieldPresent("score", foundDoc), foundDoc);
|
||||
const score = foundDoc["score"];
|
||||
assert.gte(score, 0);
|
||||
}
|
||||
const results = coll.aggregate(testQuery).toArray();
|
||||
|
||||
for (const foundDoc of results) {
|
||||
const [assertFieldPresent, subDetails, score] = checkOuterScoreDetails(foundDoc, 2);
|
||||
|
||||
const searchDetails = subDetails[0];
|
||||
const searchScore =
|
||||
checkSearchScoreDetails(assertFieldPresent, searchDetails, "search", 2, true);
|
||||
const vectorDetails = subDetails[1];
|
||||
const vectorSearchScore =
|
||||
checkVectorScoreDetails(assertFieldPresent, vectorDetails, "vector", 1);
|
||||
assert.eq(score, searchScore + vectorSearchScore);
|
||||
}
|
||||
})();
|
||||
|
||||
/**
|
||||
* Test vectorSearch/vectorSearch where neither has score details.
|
||||
* "score" : 0.054098360655737705,
|
||||
* "details" : {
|
||||
"value" : 0.054098360655737705,
|
||||
"description" : "value output by reciprocal rank fusion algorithm...",
|
||||
"details" : [
|
||||
{
|
||||
"inputPipelineName" : "secondVector",
|
||||
"rank" : 1,
|
||||
"weight" : 2.8,
|
||||
"value" : 1,
|
||||
"details" : [ ]
|
||||
},
|
||||
{
|
||||
"inputPipelineName" : "vector",
|
||||
"rank" : 1,
|
||||
"weight" : 0.5,
|
||||
"value" : 1,
|
||||
"details" : [ ]
|
||||
}
|
||||
]
|
||||
}
|
||||
*/
|
||||
(function testVectorSearchWithRankFusionScoreDetailsTwoInputPipelines() {
|
||||
const testQuery = [
|
||||
{
|
||||
$rankFusion: {
|
||||
input: {pipelines: {vector: [vectorStage], secondVector: [vectorStage]}},
|
||||
combination: {weights: {vector: 0.5, secondVector: 2.8}},
|
||||
scoreDetails: true,
|
||||
},
|
||||
},
|
||||
{$project: {score: {$meta: "score"}, details: {$meta: "scoreDetails"}}}
|
||||
];
|
||||
|
||||
const results = coll.aggregate(testQuery).toArray();
|
||||
|
||||
for (const foundDoc of results) {
|
||||
const [assertFieldPresent, subDetails, score] = checkOuterScoreDetails(foundDoc, 2);
|
||||
|
||||
const secondVectorDetails = subDetails[0];
|
||||
const secondVectorSearchScore =
|
||||
checkVectorScoreDetails(assertFieldPresent, secondVectorDetails, "secondVector", 2.8);
|
||||
const vectorDetails = subDetails[1];
|
||||
const vectorSearchScore =
|
||||
checkVectorScoreDetails(assertFieldPresent, vectorDetails, "vector", 0.5);
|
||||
assert.eq(score, secondVectorSearchScore + vectorSearchScore);
|
||||
}
|
||||
})();
|
||||
|
||||
/**
|
||||
* Test search/vectorSearch where search scoreDetails is off but $rankFusion's scoreDetails is on.
|
||||
* "score" : 0.03278688524590164,
|
||||
* "details" : {
|
||||
"value" : 0.03278688524590164,
|
||||
"description" : "value output by reciprocal rank fusion algorithm...",
|
||||
"details" : [
|
||||
{
|
||||
"inputPipelineName" : "search",
|
||||
"rank" : 1,
|
||||
"weight" : 1,
|
||||
"value" : 1.5521023273468018,
|
||||
"details" : [ ]
|
||||
},
|
||||
{
|
||||
"inputPipelineName" : "vector",
|
||||
"rank" : 1,
|
||||
"weight" : 1,
|
||||
"value" : 1,
|
||||
"details" : [ ]
|
||||
}
|
||||
]
|
||||
}
|
||||
*/
|
||||
(function testVectorSearchAndSearchNoScoreDetailsWithRankFusionScoreDetailsTwoInputPipelines() {
|
||||
const testQuery = [
|
||||
{
|
||||
$rankFusion: {
|
||||
input: {
|
||||
pipelines:
|
||||
{vector: [vectorStage], search: [searchStageNoDetails, {$limit: limit}]}
|
||||
},
|
||||
scoreDetails: true,
|
||||
},
|
||||
},
|
||||
{$project: {score: {$meta: "score"}, details: {$meta: "scoreDetails"}}}
|
||||
];
|
||||
|
||||
const results = coll.aggregate(testQuery).toArray();
|
||||
|
||||
for (const foundDoc of results) {
|
||||
const [assertFieldPresent, subDetails, score] = checkOuterScoreDetails(foundDoc, 2);
|
||||
|
||||
const searchDetails = subDetails[0];
|
||||
const searchScore =
|
||||
checkSearchScoreDetails(assertFieldPresent, searchDetails, "search", 1, false);
|
||||
const vectorDetails = subDetails[1];
|
||||
const vectorSearchScore =
|
||||
checkVectorScoreDetails(assertFieldPresent, vectorDetails, "vector", 1);
|
||||
assert.eq(score, searchScore + vectorSearchScore);
|
||||
}
|
||||
})();
|
||||
|
||||
/**
|
||||
* Test $rankFusion with scoreDetails with 1 search input pipeline that has scoreDetails.
|
||||
* "score" : 0.01639344262295082,
|
||||
* "details" : {
|
||||
"value" : 0.01639344262295082,
|
||||
"description" : "value output by reciprocal rank fusion algorithm...",
|
||||
"details" : [
|
||||
{
|
||||
"inputPipelineName" : "search",
|
||||
"rank" : 1,
|
||||
"weight" : 1,
|
||||
"value" : 1.5521023273468018,
|
||||
"description" : "sum of:",
|
||||
"details" : [
|
||||
{...} // search's scoreDetails
|
||||
]
|
||||
}
|
||||
]
|
||||
}
|
||||
*/
|
||||
(function testSearchWithScoreDetailsWithRankFusionScoreDetailsOneInputPipeline() {
|
||||
const testQuery = [
|
||||
{
|
||||
$rankFusion: {
|
||||
input: {pipelines: {search: [searchStage, {$limit: limit}]}},
|
||||
scoreDetails: true,
|
||||
},
|
||||
},
|
||||
{$project: {score: {$meta: "score"}, details: {$meta: "scoreDetails"}}}
|
||||
];
|
||||
|
||||
const results = coll.aggregate(testQuery).toArray();
|
||||
|
||||
for (const foundDoc of results) {
|
||||
const [assertFieldPresent, subDetails, score] = checkOuterScoreDetails(foundDoc, 1);
|
||||
|
||||
const searchDetails = subDetails[0];
|
||||
const searchScore =
|
||||
checkSearchScoreDetails(assertFieldPresent, searchDetails, "search", 1, true);
|
||||
assert.eq(score, searchScore);
|
||||
}
|
||||
})();
|
||||
|
||||
/**
|
||||
* Test search/vectorSearch where search scoreDetails is off and $rankFusion's scoreDetails is off.
|
||||
* { "_id" : 6, "score" : 0.03278688524590164 }
|
||||
*/
|
||||
(function testVectorSearchAndSearchNoScoreDetailsWithRankFusionNoScoreDetailsTwoInputPipelines() {
|
||||
const testQuery = [
|
||||
{
|
||||
$rankFusion: {
|
||||
input: {
|
||||
pipelines:
|
||||
{vector: [vectorStage], search: [searchStageNoDetails, {$limit: limit}]}
|
||||
},
|
||||
scoreDetails: false,
|
||||
},
|
||||
},
|
||||
{$project: {score: {$meta: "score"}}}
|
||||
];
|
||||
|
||||
const results = coll.aggregate(testQuery).toArray();
|
||||
|
||||
for (const foundDoc of results) {
|
||||
// Assert that the score metadata has been set.
|
||||
assert(fieldPresent("score", foundDoc), foundDoc);
|
||||
const score = foundDoc["score"];
|
||||
assert.gte(score, 0);
|
||||
}
|
||||
})();
|
||||
|
||||
/**
|
||||
* Verify that when $rankFusion.scoreDetails is false and an input pipeline ($search) has
|
||||
* scoreDetails set to true, the aggregation fails when scoreDetails metadata is projected out.
|
||||
*/
|
||||
testQuery = [
|
||||
{
|
||||
$rankFusion: {
|
||||
input: {pipelines: {vector: [vectorStage], search: [searchStage, {$limit: limit}]}},
|
||||
combination: {weights: {search: 2}},
|
||||
scoreDetails: false,
|
||||
(function testScoreDetailsMetadataProjectionFailsWhenRankFusionHasNoScoreDetails() {
|
||||
const testQuery = [
|
||||
{
|
||||
$rankFusion: {
|
||||
input: {pipelines: {vector: [vectorStage], search: [searchStage, {$limit: limit}]}},
|
||||
combination: {weights: {search: 2}},
|
||||
scoreDetails: false,
|
||||
},
|
||||
},
|
||||
},
|
||||
{$project: {score: {$meta: "score"}, details: {$meta: "scoreDetails"}}}
|
||||
];
|
||||
{$project: {score: {$meta: "score"}, details: {$meta: "scoreDetails"}}}
|
||||
];
|
||||
|
||||
assertErrCodeAndErrMsgContains(coll, testQuery, 40218, "query requires scoreDetails metadata");
|
||||
assertErrCodeAndErrMsgContains(coll, testQuery, 40218, "query requires scoreDetails metadata");
|
||||
})();
|
||||
|
||||
/**
|
||||
* Verify that when $rankFusion.scoreDetails is false and an input pipeline ($search) has
|
||||
* scoreDetails set to true, the aggregation succeeds when scoreDetails metadata is NOT projected
|
||||
* out.
|
||||
*/
|
||||
testQuery = [
|
||||
{
|
||||
$rankFusion: {
|
||||
input: {pipelines: {search: [searchStage, {$limit: limit}]}},
|
||||
combination: {weights: {search: 2}},
|
||||
scoreDetails: false,
|
||||
(function testQueryWithoutScoreDetailsMetadataProjectionWorksWhenRankFusionHasNoScoreDetails() {
|
||||
const testQuery = [
|
||||
{
|
||||
$rankFusion: {
|
||||
input: {pipelines: {search: [searchStage, {$limit: limit}]}},
|
||||
combination: {weights: {search: 2}},
|
||||
scoreDetails: false,
|
||||
},
|
||||
},
|
||||
},
|
||||
{$project: {plot_embedding: 0}}
|
||||
];
|
||||
{$project: {plot_embedding: 0}}
|
||||
];
|
||||
|
||||
assert.commandWorked(db.runCommand({aggregate: collName, pipeline: testQuery, cursor: {}}));
|
||||
assert.commandWorked(db.runCommand({aggregate: collName, pipeline: testQuery, cursor: {}}));
|
||||
})();
|
||||
|
||||
/**
|
||||
* Verify that when $rankFusion.scoreDetails is true and an input pipeline doesn't set
|
||||
* score/scoreDetails metadata, the projected scoreDetails is empty.
|
||||
*/
|
||||
(function testQueryWithScoreDetailsForNoScoreOrScoreDetailsGeneratingPipeline() {
|
||||
const testQuery = [
|
||||
{
|
||||
$rankFusion: {
|
||||
input: {pipelines: {matchAndSort: [{$match: {title: "ape"}}, {$sort: {title: 1}}]}},
|
||||
combination: {weights: {matchAndSort: 2}},
|
||||
scoreDetails: true,
|
||||
},
|
||||
},
|
||||
{$project: {plot_embedding: 0}}
|
||||
];
|
||||
assert.commandWorked(db.runCommand({aggregate: collName, pipeline: testQuery, cursor: {}}));
|
||||
const results = coll.aggregate(testQuery).toArray();
|
||||
assert.eq(results, []);
|
||||
})();
|
||||
|
||||
// TODO SERVER-93218 Test scoreDetails with nested rankFusion.
|
||||
dropSearchIndex(coll, {name: getMovieSearchIndexSpec().name});
|
||||
dropSearchIndex(coll, {name: getMovieVectorSearchIndexSpec().name});
|
||||
|
||||
/**
|
||||
* Verify scoreDetails correctly projected when $rankFusion takes a $geoNear input pipeline.
|
||||
*
|
||||
"_id" : 41,
|
||||
"score" : 0.04891591750396616,
|
||||
"details" : {
|
||||
"value" : 0.04891591750396616,
|
||||
"description" : "value output by reciprocal rank fusion algorithm...",
|
||||
"details" : [
|
||||
{
|
||||
"inputPipelineName" : "geoNear",
|
||||
"rank" : 1,
|
||||
"weight" : 2,
|
||||
"details" : [ ]
|
||||
},
|
||||
{
|
||||
"inputPipelineName" : "search",
|
||||
"rank" : 2,
|
||||
"weight" : 1,
|
||||
"value" : 2.7601585388183594,
|
||||
"details" : [ ]
|
||||
}
|
||||
]
|
||||
}
|
||||
}
|
||||
*/
|
||||
(function testQueryWithScoreDetailsForGeoNearInputPipeline() {
|
||||
coll.drop();
|
||||
|
||||
assert.commandWorked(coll.insertMany(getRentalData()));
|
||||
|
||||
// Index is blocking by default so that the query is only run after index has been made.
|
||||
createSearchIndex(coll, getRentalSearchIndexSpec());
|
||||
|
||||
assert.commandWorked(coll.createIndex({"address.location.coordinates": "2d"}));
|
||||
|
||||
const testQuery = [
|
||||
{
|
||||
$rankFusion: {
|
||||
input: {
|
||||
pipelines: {
|
||||
search: [
|
||||
{
|
||||
$search: {
|
||||
index: getRentalSearchIndexSpec().name,
|
||||
text: {
|
||||
query: "brooklyn",
|
||||
path: [
|
||||
"name",
|
||||
"summary",
|
||||
"description",
|
||||
"neighborhood_overview",
|
||||
],
|
||||
},
|
||||
}
|
||||
},
|
||||
{$limit: limit}
|
||||
],
|
||||
geoNear: [{
|
||||
$geoNear: {
|
||||
near: [-73.97713, 40.68675],
|
||||
}
|
||||
}],
|
||||
}
|
||||
},
|
||||
combination: {weights: {geoNear: 2}},
|
||||
scoreDetails: true,
|
||||
},
|
||||
},
|
||||
{$project: {score: {$meta: "score"}, details: {$meta: "scoreDetails"}}}
|
||||
];
|
||||
assert.commandWorked(db.runCommand({aggregate: collName, pipeline: testQuery, cursor: {}}));
|
||||
const results = coll.aggregate(testQuery).toArray();
|
||||
for (const foundDoc of results) {
|
||||
const [assertFieldPresent, subDetails, score] = checkOuterScoreDetails(foundDoc, 2);
|
||||
|
||||
// Check geoNear input pipeline details.
|
||||
const geoNearScore =
|
||||
checkGeoNearScoreDetails(assertFieldPresent, subDetails[0], "geoNear", 2);
|
||||
|
||||
// Check search input pipeline details.
|
||||
const searchScore =
|
||||
checkSearchScoreDetails(assertFieldPresent, subDetails[1], "search", 1, false);
|
||||
|
||||
assert.eq(score, geoNearScore + searchScore);
|
||||
}
|
||||
})();
|
||||
|
||||
@ -4423,5 +4423,677 @@ TEST_F(DocumentSourceRankFusionTest, CheckTwoPipelineRankFusionFullDesugaring) {
|
||||
})",
|
||||
asOneObj);
|
||||
}
|
||||
|
||||
TEST_F(DocumentSourceRankFusionTest, CheckFourPipelinesScoreDetailsDesugaring) {
|
||||
RAIIServerParameterControllerForTest featureFlagController("featureFlagRankFusionFull", true);
|
||||
auto expCtx = getExpCtx();
|
||||
expCtx->setResolvedNamespaces(ResolvedNamespaceMap{
|
||||
{expCtx->getNamespaceString(), {expCtx->getNamespaceString(), std::vector<BSONObj>()}}});
|
||||
auto spec = fromjson(R"({
|
||||
$rankFusion: {
|
||||
input: {
|
||||
pipelines: {
|
||||
matchWithTextScore: [
|
||||
{ $match: { $text: { $search: "Agatha Christie" } } },
|
||||
{ $sort: {author: 1} }
|
||||
],
|
||||
searchPipe: [
|
||||
{
|
||||
$search: {
|
||||
index: "search_index",
|
||||
text: {
|
||||
query: "mystery",
|
||||
path: "genres"
|
||||
},
|
||||
scoreDetails: true
|
||||
}
|
||||
}
|
||||
],
|
||||
vectorSearchPipe: [
|
||||
{
|
||||
$vectorSearch: {
|
||||
queryVector: [1.0, 2.0, 3.0],
|
||||
path: "plot_embedding",
|
||||
numCandidates: 300,
|
||||
index: "vector_index",
|
||||
limit: 10
|
||||
}
|
||||
}
|
||||
],
|
||||
matchWithoutTextScore: [
|
||||
{ $match : { author : "Agatha Christie" } },
|
||||
{ $sort: {author: 1} }
|
||||
]
|
||||
}
|
||||
},
|
||||
combination: {
|
||||
weights: {
|
||||
matchWithTextScore: 3,
|
||||
searchPipe: 2,
|
||||
vectorSearchPipe: 4,
|
||||
matchWithoutTextScore: 5
|
||||
}
|
||||
},
|
||||
scoreDetails: true
|
||||
}
|
||||
})");
|
||||
|
||||
const auto desugaredList =
|
||||
DocumentSourceRankFusion::createFromBson(spec.firstElement(), getExpCtx());
|
||||
const auto pipeline = Pipeline::create(desugaredList, getExpCtx());
|
||||
BSONObj asOneObj = BSON("expectedStages" << pipeline->serializeToBson());
|
||||
const std::string expectedStages = std::string(R"({
|
||||
"expectedStages": [)") +
|
||||
std::string(R"({
|
||||
"$match": {
|
||||
"$text": {
|
||||
"$search": "Agatha Christie"
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"$sort": {
|
||||
"author": 1,
|
||||
"$_internalOutputSortKeyMetadata": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"$replaceRoot": {
|
||||
"newRoot": {
|
||||
"docs": "$$ROOT"
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"$_internalSetWindowFields": {
|
||||
"sortBy": {
|
||||
"order": 1
|
||||
},
|
||||
"output": {
|
||||
"matchWithTextScore_rank": {
|
||||
"$rank": {}
|
||||
}
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"$addFields": {
|
||||
"matchWithTextScore_score": {
|
||||
"$multiply": [
|
||||
{
|
||||
"$divide": [
|
||||
{
|
||||
"$const": 1
|
||||
},
|
||||
{
|
||||
"$add": [
|
||||
"$matchWithTextScore_rank",
|
||||
{
|
||||
"$const": 60
|
||||
}
|
||||
]
|
||||
}
|
||||
]
|
||||
},
|
||||
{
|
||||
"$const": 3
|
||||
}
|
||||
]
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"$addFields": {
|
||||
"matchWithTextScore_scoreDetails": {
|
||||
"value": {
|
||||
"$meta": "score"
|
||||
},
|
||||
"details": []
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"$unionWith": {
|
||||
"coll": "pipeline_test",
|
||||
"pipeline": [
|
||||
{
|
||||
"$match": {
|
||||
"author": "Agatha Christie"
|
||||
}
|
||||
},
|
||||
{
|
||||
"$sort": {
|
||||
"author": 1,
|
||||
"$_internalOutputSortKeyMetadata": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"$replaceRoot": {
|
||||
"newRoot": {
|
||||
"docs": "$$ROOT"
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"$_internalSetWindowFields": {
|
||||
"sortBy": {
|
||||
"order": 1
|
||||
},
|
||||
"output": {
|
||||
"matchWithoutTextScore_rank": {
|
||||
"$rank": {}
|
||||
}
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"$addFields": {
|
||||
"matchWithoutTextScore_score": {
|
||||
"$multiply": [
|
||||
{
|
||||
"$divide": [
|
||||
{
|
||||
"$const": 1
|
||||
},
|
||||
{
|
||||
"$add": [
|
||||
"$matchWithoutTextScore_rank",
|
||||
{
|
||||
"$const": 60
|
||||
}
|
||||
]
|
||||
}
|
||||
]
|
||||
},
|
||||
{
|
||||
"$const": 5
|
||||
}
|
||||
]
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"$addFields": {
|
||||
"matchWithoutTextScore_scoreDetails": {
|
||||
"details": []
|
||||
}
|
||||
}
|
||||
}
|
||||
]
|
||||
}
|
||||
},)") +
|
||||
std::string(R"(
|
||||
{
|
||||
"$unionWith": {
|
||||
"coll": "pipeline_test",
|
||||
"pipeline": [
|
||||
{
|
||||
"$search": {
|
||||
"mongotQuery": {
|
||||
"index": "search_index",
|
||||
"text": {
|
||||
"query": "mystery",
|
||||
"path": "genres"
|
||||
},
|
||||
"scoreDetails": true
|
||||
},
|
||||
"requiresSearchSequenceToken": false,
|
||||
"requiresSearchMetaCursor": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"$replaceRoot": {
|
||||
"newRoot": {
|
||||
"docs": "$$ROOT"
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"$_internalSetWindowFields": {
|
||||
"sortBy": {
|
||||
"order": 1
|
||||
},
|
||||
"output": {
|
||||
"searchPipe_rank": {
|
||||
"$rank": {}
|
||||
}
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"$addFields": {
|
||||
"searchPipe_score": {
|
||||
"$multiply": [
|
||||
{
|
||||
"$divide": [
|
||||
{
|
||||
"$const": 1
|
||||
},
|
||||
{
|
||||
"$add": [
|
||||
"$searchPipe_rank",
|
||||
{
|
||||
"$const": 60
|
||||
}
|
||||
]
|
||||
}
|
||||
]
|
||||
},
|
||||
{
|
||||
"$const": 2
|
||||
}
|
||||
]
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"$addFields": {
|
||||
"searchPipe_scoreDetails": {
|
||||
"$meta": "scoreDetails"
|
||||
}
|
||||
}
|
||||
}
|
||||
]
|
||||
}
|
||||
},
|
||||
{
|
||||
"$unionWith": {
|
||||
"coll": "pipeline_test",
|
||||
"pipeline": [
|
||||
{
|
||||
"$vectorSearch": {
|
||||
"queryVector": [
|
||||
1,
|
||||
2,
|
||||
3
|
||||
],
|
||||
"path": "plot_embedding",
|
||||
"numCandidates": 300,
|
||||
"index": "vector_index",
|
||||
"limit": 10
|
||||
}
|
||||
},
|
||||
{
|
||||
"$replaceRoot": {
|
||||
"newRoot": {
|
||||
"docs": "$$ROOT"
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"$_internalSetWindowFields": {
|
||||
"sortBy": {
|
||||
"order": 1
|
||||
},
|
||||
"output": {
|
||||
"vectorSearchPipe_rank": {
|
||||
"$rank": {}
|
||||
}
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"$addFields": {
|
||||
"vectorSearchPipe_score": {
|
||||
"$multiply": [
|
||||
{
|
||||
"$divide": [
|
||||
{
|
||||
"$const": 1
|
||||
},
|
||||
{
|
||||
"$add": [
|
||||
"$vectorSearchPipe_rank",
|
||||
{
|
||||
"$const": 60
|
||||
}
|
||||
]
|
||||
}
|
||||
]
|
||||
},
|
||||
{
|
||||
"$const": 4
|
||||
}
|
||||
]
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"$addFields": {
|
||||
"vectorSearchPipe_scoreDetails": {
|
||||
"value": {
|
||||
"$meta": "score"
|
||||
},
|
||||
"details": []
|
||||
}
|
||||
}
|
||||
}
|
||||
]
|
||||
}
|
||||
},)") +
|
||||
std::string(R"(
|
||||
{
|
||||
"$group": {
|
||||
"_id": "$docs._id",
|
||||
"docs": {
|
||||
"$first": "$docs"
|
||||
},
|
||||
"matchWithTextScore_score": {
|
||||
"$max": {
|
||||
"$ifNull": [
|
||||
"$matchWithTextScore_score",
|
||||
{
|
||||
"$const": 0
|
||||
}
|
||||
]
|
||||
}
|
||||
},
|
||||
"matchWithTextScore_rank": {
|
||||
"$max": {
|
||||
"$ifNull": [
|
||||
"$matchWithTextScore_rank",
|
||||
{
|
||||
"$const": 0
|
||||
}
|
||||
]
|
||||
}
|
||||
},
|
||||
"matchWithTextScore_scoreDetails": {
|
||||
"$mergeObjects": "$matchWithTextScore_scoreDetails"
|
||||
},
|
||||
"matchWithoutTextScore_score": {
|
||||
"$max": {
|
||||
"$ifNull": [
|
||||
"$matchWithoutTextScore_score",
|
||||
{
|
||||
"$const": 0
|
||||
}
|
||||
]
|
||||
}
|
||||
},
|
||||
"matchWithoutTextScore_rank": {
|
||||
"$max": {
|
||||
"$ifNull": [
|
||||
"$matchWithoutTextScore_rank",
|
||||
{
|
||||
"$const": 0
|
||||
}
|
||||
]
|
||||
}
|
||||
},
|
||||
"matchWithoutTextScore_scoreDetails": {
|
||||
"$mergeObjects": "$matchWithoutTextScore_scoreDetails"
|
||||
},
|
||||
"searchPipe_score": {
|
||||
"$max": {
|
||||
"$ifNull": [
|
||||
"$searchPipe_score",
|
||||
{
|
||||
"$const": 0
|
||||
}
|
||||
]
|
||||
}
|
||||
},
|
||||
"searchPipe_rank": {
|
||||
"$max": {
|
||||
"$ifNull": [
|
||||
"$searchPipe_rank",
|
||||
{
|
||||
"$const": 0
|
||||
}
|
||||
]
|
||||
}
|
||||
},
|
||||
"searchPipe_scoreDetails": {
|
||||
"$mergeObjects": "$searchPipe_scoreDetails"
|
||||
},
|
||||
"vectorSearchPipe_score": {
|
||||
"$max": {
|
||||
"$ifNull": [
|
||||
"$vectorSearchPipe_score",
|
||||
{
|
||||
"$const": 0
|
||||
}
|
||||
]
|
||||
}
|
||||
},
|
||||
"vectorSearchPipe_rank": {
|
||||
"$max": {
|
||||
"$ifNull": [
|
||||
"$vectorSearchPipe_rank",
|
||||
{
|
||||
"$const": 0
|
||||
}
|
||||
]
|
||||
}
|
||||
},
|
||||
"vectorSearchPipe_scoreDetails": {
|
||||
"$mergeObjects": "$vectorSearchPipe_scoreDetails"
|
||||
},
|
||||
"$willBeMerged": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"$addFields": {
|
||||
"score": {
|
||||
"$add": [
|
||||
"$matchWithTextScore_score",
|
||||
"$matchWithoutTextScore_score",
|
||||
"$searchPipe_score",
|
||||
"$vectorSearchPipe_score"
|
||||
]
|
||||
}
|
||||
}
|
||||
},)") +
|
||||
std::string(R"(
|
||||
{
|
||||
"$addFields": {
|
||||
"matchWithTextScore_rank": {
|
||||
"$cond": [
|
||||
{
|
||||
"$eq": [
|
||||
"$matchWithTextScore_rank",
|
||||
{
|
||||
"$const": 0
|
||||
}
|
||||
]
|
||||
},
|
||||
{
|
||||
"$const": "NA"
|
||||
},
|
||||
"$matchWithTextScore_rank"
|
||||
]
|
||||
},
|
||||
"matchWithoutTextScore_rank": {
|
||||
"$cond": [
|
||||
{
|
||||
"$eq": [
|
||||
"$matchWithoutTextScore_rank",
|
||||
{
|
||||
"$const": 0
|
||||
}
|
||||
]
|
||||
},
|
||||
{
|
||||
"$const": "NA"
|
||||
},
|
||||
"$matchWithoutTextScore_rank"
|
||||
]
|
||||
},
|
||||
"searchPipe_rank": {
|
||||
"$cond": [
|
||||
{
|
||||
"$eq": [
|
||||
"$searchPipe_rank",
|
||||
{
|
||||
"$const": 0
|
||||
}
|
||||
]
|
||||
},
|
||||
{
|
||||
"$const": "NA"
|
||||
},
|
||||
"$searchPipe_rank"
|
||||
]
|
||||
},
|
||||
"vectorSearchPipe_rank": {
|
||||
"$cond": [
|
||||
{
|
||||
"$eq": [
|
||||
"$vectorSearchPipe_rank",
|
||||
{
|
||||
"$const": 0
|
||||
}
|
||||
]
|
||||
},
|
||||
{
|
||||
"$const": "NA"
|
||||
},
|
||||
"$vectorSearchPipe_rank"
|
||||
]
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"$addFields": {
|
||||
"calculatedScoreDetails": [
|
||||
{
|
||||
"$mergeObjects": [
|
||||
{
|
||||
"inputPipelineName": {
|
||||
"$const": "matchWithTextScore"
|
||||
},
|
||||
"rank": "$matchWithTextScore_rank",
|
||||
"weight": {
|
||||
"$cond": [
|
||||
{
|
||||
"$eq": [
|
||||
"$matchWithTextScore_rank",
|
||||
{
|
||||
"$const": "NA"
|
||||
}
|
||||
]
|
||||
},
|
||||
"$$REMOVE",
|
||||
{
|
||||
"$const": 3
|
||||
}
|
||||
]
|
||||
}
|
||||
},
|
||||
"$matchWithTextScore_scoreDetails"
|
||||
]
|
||||
},
|
||||
{
|
||||
"$mergeObjects": [
|
||||
{
|
||||
"inputPipelineName": {
|
||||
"$const": "matchWithoutTextScore"
|
||||
},
|
||||
"rank": "$matchWithoutTextScore_rank",
|
||||
"weight": {
|
||||
"$cond": [
|
||||
{
|
||||
"$eq": [
|
||||
"$matchWithoutTextScore_rank",
|
||||
{
|
||||
"$const": "NA"
|
||||
}
|
||||
]
|
||||
},
|
||||
"$$REMOVE",
|
||||
{
|
||||
"$const": 5
|
||||
}
|
||||
]
|
||||
}
|
||||
},
|
||||
"$matchWithoutTextScore_scoreDetails"
|
||||
]
|
||||
},
|
||||
{
|
||||
"$mergeObjects": [
|
||||
{
|
||||
"inputPipelineName": {
|
||||
"$const": "searchPipe"
|
||||
},
|
||||
"rank": "$searchPipe_rank",
|
||||
"weight": {
|
||||
"$cond": [
|
||||
{
|
||||
"$eq": [
|
||||
"$searchPipe_rank",
|
||||
{
|
||||
"$const": "NA"
|
||||
}
|
||||
]
|
||||
},
|
||||
"$$REMOVE",
|
||||
{
|
||||
"$const": 2
|
||||
}
|
||||
]
|
||||
}
|
||||
},
|
||||
"$searchPipe_scoreDetails"
|
||||
]
|
||||
},
|
||||
{
|
||||
"$mergeObjects": [
|
||||
{
|
||||
"inputPipelineName": {
|
||||
"$const": "vectorSearchPipe"
|
||||
},
|
||||
"rank": "$vectorSearchPipe_rank",
|
||||
"weight": {
|
||||
"$cond": [
|
||||
{
|
||||
"$eq": [
|
||||
"$vectorSearchPipe_rank",
|
||||
{
|
||||
"$const": "NA"
|
||||
}
|
||||
]
|
||||
},
|
||||
"$$REMOVE",
|
||||
{
|
||||
"$const": 4
|
||||
}
|
||||
]
|
||||
}
|
||||
},
|
||||
"$vectorSearchPipe_scoreDetails"
|
||||
]
|
||||
}
|
||||
]
|
||||
}
|
||||
},)") +
|
||||
std::string(R"(
|
||||
{
|
||||
"$setMetadata": {
|
||||
"scoreDetails": {
|
||||
"value": "$score",
|
||||
"description": {
|
||||
"$const": "value output by reciprocal rank fusion algorithm, computed as sum of (weight * (1 / (60 + rank))) across input pipelines from which this document is output, from:"
|
||||
},
|
||||
"details": "$calculatedScoreDetails"
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"$sort": {
|
||||
"score": -1,
|
||||
"_id": 1
|
||||
}
|
||||
},
|
||||
{
|
||||
"$replaceRoot": {
|
||||
"newRoot": "$docs"
|
||||
}
|
||||
}
|
||||
]
|
||||
})");
|
||||
ASSERT_BSONOBJ_EQ_AUTO(expectedStages, asOneObj);
|
||||
}
|
||||
} // namespace
|
||||
} // namespace mongo
|
||||
|
||||
Loading…
Reference in New Issue
Block a user