SOLR-10082: JSON Facet API, add stddev and variance functions

This commit is contained in:
yonik 2017-04-17 22:30:29 -04:00
parent d286864d80
commit 3145f781b3
7 changed files with 311 additions and 50 deletions

View File

@ -167,6 +167,10 @@ New Features
initializing CloudSolrClient would not work if you have collection aliases on older versions of Solr
server that doesn't support LISTALIASES. (Ishan Chattopadhyaya, Noble Paul)
* SOLR-10082: Variance and Standard Deviation aggregators for the JSON Facet API.
Example: json.facet={x:"stddev(field1)", y:"variance(field2)"}
(Rustam Hashimov, yonik)
Optimizations
----------------------

View File

@ -58,9 +58,11 @@ import org.apache.solr.search.facet.HLLAgg;
import org.apache.solr.search.facet.MaxAgg;
import org.apache.solr.search.facet.MinAgg;
import org.apache.solr.search.facet.PercentileAgg;
import org.apache.solr.search.facet.StddevAgg;
import org.apache.solr.search.facet.SumAgg;
import org.apache.solr.search.facet.SumsqAgg;
import org.apache.solr.search.facet.UniqueAgg;
import org.apache.solr.search.facet.VarianceAgg;
import org.apache.solr.search.function.CollapseScoreFunction;
import org.apache.solr.search.function.OrdFieldSource;
import org.apache.solr.search.function.ReverseOrdFieldSource;
@ -931,14 +933,21 @@ public abstract class ValueSourceParser implements NamedListInitializedPlugin {
}
});
/***
addParser("agg_stdev", new ValueSourceParser() {
@Override
public ValueSource parse(FunctionQParser fp) throws SyntaxError {
return null;
}
addParser("agg_variance", new ValueSourceParser() {
@Override
public ValueSource parse(FunctionQParser fp) throws SyntaxError {
return new VarianceAgg(fp.parseValueSource());
}
});
addParser("agg_stddev", new ValueSourceParser() {
@Override
public ValueSource parse(FunctionQParser fp) throws SyntaxError {
return new StddevAgg(fp.parseValueSource());
}
});
/***
addParser("agg_multistat", new ValueSourceParser() {
@Override
public ValueSource parse(FunctionQParser fp) throws SyntaxError {

View File

@ -33,7 +33,7 @@ import java.util.Iterator;
import java.util.List;
/**
* Accumulates statistics separated by a slot number.
* Accumulates statistics separated by a slot number.
* There is a separate statistic per slot. The slot is usually an ordinal into a set of values, e.g. tracking a count
* frequency <em>per term</em>.
* Sometimes there doesn't need to be a slot distinction, in which case there is just one nominal slot.
@ -46,8 +46,7 @@ public abstract class SlotAcc implements Closeable {
this.fcontext = fcontext;
}
public void setNextReader(LeafReaderContext readerContext) throws IOException {
}
public void setNextReader(LeafReaderContext readerContext) throws IOException {}
public abstract void collect(int doc, int slot) throws IOException;
@ -61,7 +60,7 @@ public abstract class SlotAcc implements Closeable {
int segBase = 0;
int segMax;
int adjustedMax = 0;
for (DocIterator docsIt = docs.iterator(); docsIt.hasNext(); ) {
for (DocIterator docsIt = docs.iterator(); docsIt.hasNext();) {
final int doc = docsIt.nextDoc();
if (doc >= adjustedMax) {
do {
@ -78,12 +77,11 @@ public abstract class SlotAcc implements Closeable {
setNextReader(ctx);
}
count++;
collect(doc - segBase, slot); // per-seg collectors
collect(doc - segBase, slot); // per-seg collectors
}
return count;
}
public abstract int compare(int slotA, int slotB);
public abstract Object getValue(int slotNum) throws IOException;
@ -101,8 +99,7 @@ public abstract class SlotAcc implements Closeable {
public abstract void resize(Resizer resizer);
@Override
public void close() throws IOException {
}
public void close() throws IOException {}
public static abstract class Resizer {
public abstract int getNewSize();
@ -181,15 +178,14 @@ abstract class FuncSlotAcc extends SlotAcc {
}
}
// have a version that counts the number of times a Slot has been hit? (for avg... what else?)
// have a version that counts the number of times a Slot has been hit? (for avg... what else?)
// TODO: make more sense to have func as the base class rather than double?
// double-slot-func -> func-slot -> slot -> acc
// double-slot-func -> double-slot -> slot -> acc
abstract class DoubleFuncSlotAcc extends FuncSlotAcc {
double[] result; // TODO: use DoubleArray
double[] result; // TODO: use DoubleArray
double initialValue;
public DoubleFuncSlotAcc(ValueSource values, FacetContext fcontext, int numSlots) {
@ -210,7 +206,6 @@ abstract class DoubleFuncSlotAcc extends FuncSlotAcc {
return Double.compare(result[slotA], result[slotB]);
}
@Override
public Object getValue(int slot) {
return result[slot];
@ -228,7 +223,7 @@ abstract class DoubleFuncSlotAcc extends FuncSlotAcc {
}
abstract class IntSlotAcc extends SlotAcc {
int[] result; // use LongArray32
int[] result; // use LongArray32
int initialValue;
public IntSlotAcc(FacetContext fcontext, int numSlots, int initialValue) {
@ -261,15 +256,13 @@ abstract class IntSlotAcc extends SlotAcc {
}
}
class SumSlotAcc extends DoubleFuncSlotAcc {
public SumSlotAcc(ValueSource values, FacetContext fcontext, int numSlots) {
super(values, fcontext, numSlots);
}
public void collect(int doc, int slotNum) throws IOException {
double val = values.doubleVal(doc); // todo: worth trying to share this value across multiple stats that need it?
double val = values.doubleVal(doc); // todo: worth trying to share this value across multiple stats that need it?
result[slotNum] += val;
}
}
@ -287,8 +280,6 @@ class SumsqSlotAcc extends DoubleFuncSlotAcc {
}
}
class MinSlotAcc extends DoubleFuncSlotAcc {
public MinSlotAcc(ValueSource values, FacetContext fcontext, int numSlots) {
super(values, fcontext, numSlots, Double.NaN);
@ -297,10 +288,10 @@ class MinSlotAcc extends DoubleFuncSlotAcc {
@Override
public void collect(int doc, int slotNum) throws IOException {
double val = values.doubleVal(doc);
if (val == 0 && !values.exists(doc)) return; // depend on fact that non existing values return 0 for func query
if (val == 0 && !values.exists(doc)) return; // depend on fact that non existing values return 0 for func query
double currMin = result[slotNum];
if (!(val >= currMin)) { // val>=currMin will be false for staring value: val>=NaN
if (!(val >= currMin)) { // val>=currMin will be false for staring value: val>=NaN
result[slotNum] = val;
}
}
@ -314,17 +305,16 @@ class MaxSlotAcc extends DoubleFuncSlotAcc {
@Override
public void collect(int doc, int slotNum) throws IOException {
double val = values.doubleVal(doc);
if (val == 0 && !values.exists(doc)) return; // depend on fact that non existing values return 0 for func query
if (val == 0 && !values.exists(doc)) return; // depend on fact that non existing values return 0 for func query
double currMax = result[slotNum];
if (!(val <= currMax)) { // reversed order to handle NaN
if (!(val <= currMax)) { // reversed order to handle NaN
result[slotNum] = val;
}
}
}
class AvgSlotAcc extends DoubleFuncSlotAcc {
int[] counts;
@ -336,7 +326,7 @@ class AvgSlotAcc extends DoubleFuncSlotAcc {
@Override
public void reset() {
super.reset();
for (int i=0; i<counts.length; i++) {
for (int i = 0; i < counts.length; i++) {
counts[i] = 0;
}
}
@ -351,11 +341,12 @@ class AvgSlotAcc extends DoubleFuncSlotAcc {
}
private double avg(double tot, int count) {
return count==0 ? 0 : tot/count; // returns 0 instead of NaN.. todo - make configurable? if NaN, we need to handle comparisons though...
return count == 0 ? 0 : tot / count; // returns 0 instead of NaN.. todo - make configurable? if NaN, we need to
// handle comparisons though...
}
private double avg(int slot) {
return avg(result[slot], counts[slot]); // calc once and cache in result?
return avg(result[slot], counts[slot]); // calc once and cache in result?
}
@Override
@ -367,8 +358,8 @@ class AvgSlotAcc extends DoubleFuncSlotAcc {
public Object getValue(int slot) {
if (fcontext.isShard()) {
ArrayList lst = new ArrayList(2);
lst.add( counts[slot] );
lst.add( result[slot] );
lst.add(counts[slot]);
lst.add(result[slot]);
return lst;
} else {
return avg(slot);
@ -382,32 +373,157 @@ class AvgSlotAcc extends DoubleFuncSlotAcc {
}
}
class VarianceSlotAcc extends DoubleFuncSlotAcc {
int[] counts;
double[] sum;
public VarianceSlotAcc(ValueSource values, FacetContext fcontext, int numSlots) {
super(values, fcontext, numSlots);
counts = new int[numSlots];
sum = new double[numSlots];
}
@Override
public void reset() {
super.reset();
Arrays.fill(counts, 0);
Arrays.fill(sum, 0);
}
@Override
public void resize(Resizer resizer) {
super.resize(resizer);
this.counts = resizer.resize(this.counts, 0);
this.sum = resizer.resize(this.sum, 0);
}
private double variance(double sumSq, double sum, int count) {
double val = count == 0 ? 0 : (sumSq / count) - Math.pow(sum / count, 2);
return val;
}
private double variance(int slot) {
return variance(result[slot], sum[slot], counts[slot]); // calc once and cache in result?
}
@Override
public int compare(int slotA, int slotB) {
return Double.compare(this.variance(slotA), this.variance(slotB));
}
@Override
public Object getValue(int slot) {
if (fcontext.isShard()) {
ArrayList lst = new ArrayList(3);
lst.add(counts[slot]);
lst.add(result[slot]);
lst.add(sum[slot]);
return lst;
} else {
return this.variance(slot);
}
}
@Override
public void collect(int doc, int slot) throws IOException {
double val = values.doubleVal(doc);
if (values.exists(doc)) {
counts[slot]++;
result[slot] += val * val;
sum[slot] += val;
}
}
}
class StddevSlotAcc extends DoubleFuncSlotAcc {
int[] counts;
double[] sum;
public StddevSlotAcc(ValueSource values, FacetContext fcontext, int numSlots) {
super(values, fcontext, numSlots);
counts = new int[numSlots];
sum = new double[numSlots];
}
@Override
public void reset() {
super.reset();
Arrays.fill(counts, 0);
Arrays.fill(sum, 0);
}
@Override
public void resize(Resizer resizer) {
super.resize(resizer);
this.counts = resizer.resize(this.counts, 0);
this.result = resizer.resize(this.result, 0);
}
private double stdDev(double sumSq, double sum, int count) {
double val = count == 0 ? 0 : Math.sqrt((sumSq / count) - Math.pow(sum / count, 2));
return val;
}
private double stdDev(int slot) {
return stdDev(result[slot], sum[slot], counts[slot]); // calc once and cache in result?
}
@Override
public int compare(int slotA, int slotB) {
return Double.compare(this.stdDev(slotA), this.stdDev(slotB));
}
@Override
public Object getValue(int slot) {
if (fcontext.isShard()) {
ArrayList lst = new ArrayList(3);
lst.add(counts[slot]);
lst.add(result[slot]);
lst.add(sum[slot]);
return lst;
} else {
return this.stdDev(slot);
}
}
@Override
public void collect(int doc, int slot) throws IOException {
double val = values.doubleVal(doc);
if (values.exists(doc)) {
counts[slot]++;
result[slot] += val * val;
sum[slot] += val;
}
}
}
abstract class CountSlotAcc extends SlotAcc {
public CountSlotAcc(FacetContext fcontext) {
super(fcontext);
}
public abstract void incrementCount(int slot, int count);
public abstract int getCount(int slot);
}
class CountSlotArrAcc extends CountSlotAcc {
int[] result;
public CountSlotArrAcc(FacetContext fcontext, int numSlots) {
super(fcontext);
result = new int[numSlots];
}
@Override
public void collect(int doc, int slotNum) { // TODO: count arrays can use fewer bytes based on the number of docs in the base set (that's the upper bound for single valued) - look at ttf?
public void collect(int doc, int slotNum) { // TODO: count arrays can use fewer bytes based on the number of docs in
// the base set (that's the upper bound for single valued) - look at ttf?
result[slotNum]++;
}
@Override
public int compare(int slotA, int slotB) {
return Integer.compare( result[slotA], result[slotB] );
return Integer.compare(result[slotA], result[slotB]);
}
@Override
@ -439,7 +555,6 @@ class CountSlotArrAcc extends CountSlotAcc {
}
}
class SortSlotAcc extends SlotAcc {
public SortSlotAcc(FacetContext fcontext) {
super(fcontext);

View File

@ -0,0 +1,66 @@
/*
* Licensed to the Apache Software Foundation (ASF) under one or more
* contributor license agreements. See the NOTICE file distributed with
* this work for additional information regarding copyright ownership.
* The ASF licenses this file to You under the Apache License, Version 2.0
* (the "License"); you may not use this file except in compliance with
* the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.solr.search.facet;
import java.io.IOException;
import java.util.List;
import org.apache.lucene.queries.function.ValueSource;
public class StddevAgg extends SimpleAggValueSource {
public StddevAgg(ValueSource vs) {
super("stddev", vs);
}
@Override
public SlotAcc createSlotAcc(FacetContext fcontext, int numDocs, int numSlots) throws IOException {
return new StddevSlotAcc(getArg(), fcontext, numSlots);
}
@Override
public FacetMerger createFacetMerger(Object prototype) {
return new Merger();
}
private static class Merger extends FacetDoubleMerger {
long count;
double sumSq;
double sum;
@Override
@SuppressWarnings("unchecked")
public void merge(Object facetResult, Context mcontext1) {
List<Number> numberList = (List<Number>)facetResult;
this.count += numberList.get(0).longValue();
this.sumSq += numberList.get(1).doubleValue();
this.sum += numberList.get(2).doubleValue();
}
@Override
public Object getMergedResult() {
return this.getDouble();
}
@Override
protected double getDouble() {
double val = count == 0 ? 0.0d : Math.sqrt((sumSq/count)-Math.pow(sum/count, 2));
return val;
}
};
}

View File

@ -0,0 +1,65 @@
/*
* Licensed to the Apache Software Foundation (ASF) under one or more
* contributor license agreements. See the NOTICE file distributed with
* this work for additional information regarding copyright ownership.
* The ASF licenses this file to You under the Apache License, Version 2.0
* (the "License"); you may not use this file except in compliance with
* the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.solr.search.facet;
import java.io.IOException;
import java.util.List;
import org.apache.lucene.queries.function.ValueSource;
public class VarianceAgg extends SimpleAggValueSource {
public VarianceAgg(ValueSource vs) {
super("variance", vs);
}
@Override
public SlotAcc createSlotAcc(FacetContext fcontext, int numDocs, int numSlots) throws IOException {
return new VarianceSlotAcc(getArg(), fcontext, numSlots);
}
@Override
public FacetMerger createFacetMerger(Object prototype) {
return new Merger();
}
private static class Merger extends FacetDoubleMerger {
long count;
double sumSq;
double sum;
@Override
@SuppressWarnings("unchecked")
public void merge(Object facetResult, Context mcontext1) {
List<Number> numberList = (List<Number>)facetResult;
this.count += numberList.get(0).longValue();
this.sumSq += numberList.get(1).doubleValue();
this.sum += numberList.get(2).doubleValue();
}
@Override
public Object getMergedResult() {
return this.getDouble();
}
@Override
protected double getDouble() {
double val = count == 0 ? 0.0d : (sumSq/count)-Math.pow(sum/count, 2);
return val;
}
};
}

View File

@ -1099,7 +1099,8 @@ public class QueryEqualityTest extends SolrTestCaseJ4 {
assertFuncEquals("agg_hll(foo_i)", "agg_hll(foo_i)");
assertFuncEquals("agg_sumsq(foo_i)", "agg_sumsq(foo_i)");
assertFuncEquals("agg_percentile(foo_i,50)", "agg_percentile(foo_i,50)");
// assertFuncEquals("agg_stdev(foo_i)", "agg_stdev(foo_i)");
assertFuncEquals("agg_variance(foo_i)", "agg_variance(foo_i)");
assertFuncEquals("agg_stddev(foo_i)", "agg_stddev(foo_i)");
// assertFuncEquals("agg_multistat(foo_i)", "agg_multistat(foo_i)");
}

View File

@ -529,6 +529,7 @@ public class TestJsonFacets extends SolrTestCaseHS {
" , f2:{${terms} type:terms, field:'${cat_s}', sort:'x desc', facet:{x:'max(${num_d})'} } " +
" , f3:{${terms} type:terms, field:'${cat_s}', sort:'x desc', facet:{x:'unique(${where_s})'} } " +
" , f4:{${terms} type:terms, field:'${cat_s}', sort:'x desc', facet:{x:'hll(${where_s})'} } " +
" , f5:{${terms} type:terms, field:'${cat_s}', sort:'x desc', facet:{x:'variance(${num_d})'} } " +
"}"
)
, "facets=={ 'count':6, " +
@ -536,6 +537,7 @@ public class TestJsonFacets extends SolrTestCaseHS {
", f2:{ 'buckets':[{ val:'B', count:3, x:11.0 }, { val:'A', count:2, x:4.0 }]} " +
", f3:{ 'buckets':[{ val:'A', count:2, x:2 }, { val:'B', count:3, x:2 }]} " +
", f4:{ 'buckets':[{ val:'A', count:2, x:2 }, { val:'B', count:3, x:2 }]} " +
", f5:{ 'buckets':[{ val:'B', count:3, x:74.6666666666666 }, { val:'A', count:2, x:1.0 }]} " +
"}"
);
@ -845,19 +847,18 @@ public class TestJsonFacets extends SolrTestCaseHS {
);
// stats at top level
client.testJQ(params(p, "q", "*:*"
, "json.facet", "{ sum1:'sum(${num_d})', sumsq1:'sumsq(${num_d})', avg1:'avg(${num_d})', avg2:'avg(def(${num_d},0))', min1:'min(${num_d})', max1:'max(${num_d})'" +
", numwhere:'unique(${where_s})', unique_num_i:'unique(${num_i})', unique_num_d:'unique(${num_d})', unique_date:'unique(${date})'" +
", where_hll:'hll(${where_s})', hll_num_i:'hll(${num_i})', hll_num_d:'hll(${num_d})', hll_date:'hll(${date})'" +
", med:'percentile(${num_d},50)', perc:'percentile(${num_d},0,50.0,100)' }"
", med:'percentile(${num_d},50)', perc:'percentile(${num_d},0,50.0,100)', variance:'variance(${num_d})', stddev:'stddev(${num_d})' }"
)
, "facets=={ 'count':6, " +
"sum1:3.0, sumsq1:247.0, avg1:0.6, avg2:0.5, min1:-9.0, max1:11.0" +
", numwhere:2, unique_num_i:4, unique_num_d:5, unique_date:5" +
", where_hll:2, hll_num_i:4, hll_num_d:5, hll_date:5" +
", med:2.0, perc:[-9.0,2.0,11.0] }"
", med:2.0, perc:[-9.0,2.0,11.0], variance:49.04, stddev:7.002856560004639}"
);
// stats at top level, no matches
@ -865,21 +866,20 @@ public class TestJsonFacets extends SolrTestCaseHS {
, "json.facet", "{ sum1:'sum(${num_d})', sumsq1:'sumsq(${num_d})', avg1:'avg(${num_d})', min1:'min(${num_d})', max1:'max(${num_d})'" +
", numwhere:'unique(${where_s})', unique_num_i:'unique(${num_i})', unique_num_d:'unique(${num_d})', unique_date:'unique(${date})'" +
", where_hll:'hll(${where_s})', hll_num_i:'hll(${num_i})', hll_num_d:'hll(${num_d})', hll_date:'hll(${date})'" +
", med:'percentile(${num_d},50)', perc:'percentile(${num_d},0,50.0,100)' }"
", med:'percentile(${num_d},50)', perc:'percentile(${num_d},0,50.0,100)', variance:'variance(${num_d})', stddev:'stddev(${num_d})' }"
)
, "facets=={count:0 " +
"/* ,sum1:0.0, sumsq1:0.0, avg1:0.0, min1:'NaN', max1:'NaN', numwhere:0 */" +
"\n// ,sum1:0.0, sumsq1:0.0, avg1:0.0, min1:'NaN', max1:'NaN', numwhere:0 \n" +
" }"
);
// stats at top level, matching documents, but no values in the field
// NOTE: this represents the current state of what is returned, not the ultimate desired state.
client.testJQ(params(p, "q", "id:3"
, "json.facet", "{ sum1:'sum(${num_d})', sumsq1:'sumsq(${num_d})', avg1:'avg(${num_d})', min1:'min(${num_d})', max1:'max(${num_d})'" +
", numwhere:'unique(${where_s})', unique_num_i:'unique(${num_i})', unique_num_d:'unique(${num_d})', unique_date:'unique(${date})'" +
", where_hll:'hll(${where_s})', hll_num_i:'hll(${num_i})', hll_num_d:'hll(${num_d})', hll_date:'hll(${date})'" +
", med:'percentile(${num_d},50)', perc:'percentile(${num_d},0,50.0,100)' }"
", med:'percentile(${num_d},50)', perc:'percentile(${num_d},0,50.0,100)', variance:'variance(${num_d})', stddev:'stddev(${num_d})' }"
)
, "facets=={count:1 " +
",sum1:0.0," +
@ -894,11 +894,12 @@ public class TestJsonFacets extends SolrTestCaseHS {
" where_hll:0," +
" hll_num_i:0," +
" hll_num_d:0," +
" hll_date:0" +
" hll_date:0," +
" variance:0.0," +
" stddev:0.0" +
" }"
);
//
// tests on a multi-valued field with actual multiple values, just to ensure that we are
// using a multi-valued method for the rest of the tests when appropriate.