org.elasticsearch.search.aggregations.pipeline.movavg.MovAvgPipelineAggregator Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of elasticsearch Show documentation
Show all versions of elasticsearch Show documentation
Elasticsearch - Open Source, Distributed, RESTful Search Engine
/*
* Licensed to Elasticsearch under one or more contributor
* license agreements. See the NOTICE file distributed with
* this work for additional information regarding copyright
* ownership. Elasticsearch 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.elasticsearch.search.aggregations.pipeline.movavg;
import com.google.common.base.Function;
import com.google.common.collect.EvictingQueue;
import org.elasticsearch.common.io.stream.StreamInput;
import org.elasticsearch.common.io.stream.StreamOutput;
import org.elasticsearch.search.aggregations.Aggregation;
import org.elasticsearch.search.aggregations.AggregationExecutionException;
import org.elasticsearch.search.aggregations.AggregatorFactory;
import org.elasticsearch.search.aggregations.InternalAggregation;
import org.elasticsearch.search.aggregations.InternalAggregation.ReduceContext;
import org.elasticsearch.search.aggregations.InternalAggregation.Type;
import org.elasticsearch.search.aggregations.InternalAggregations;
import org.elasticsearch.search.aggregations.bucket.histogram.HistogramAggregator;
import org.elasticsearch.search.aggregations.bucket.histogram.InternalHistogram;
import org.elasticsearch.search.aggregations.pipeline.BucketHelpers.GapPolicy;
import org.elasticsearch.search.aggregations.pipeline.InternalSimpleValue;
import org.elasticsearch.search.aggregations.pipeline.PipelineAggregator;
import org.elasticsearch.search.aggregations.pipeline.PipelineAggregatorFactory;
import org.elasticsearch.search.aggregations.pipeline.PipelineAggregatorStreams;
import org.elasticsearch.search.aggregations.pipeline.movavg.models.MovAvgModel;
import org.elasticsearch.search.aggregations.pipeline.movavg.models.MovAvgModelStreams;
import org.elasticsearch.search.aggregations.support.format.ValueFormatter;
import org.elasticsearch.search.aggregations.support.format.ValueFormatterStreams;
import org.joda.time.DateTime;
import java.io.IOException;
import java.util.ArrayList;
import java.util.List;
import java.util.ListIterator;
import java.util.Map;
import static org.elasticsearch.common.util.CollectionUtils.eagerTransform;
import static org.elasticsearch.search.aggregations.pipeline.BucketHelpers.resolveBucketValue;
public class MovAvgPipelineAggregator extends PipelineAggregator {
public final static Type TYPE = new Type("moving_avg");
public final static PipelineAggregatorStreams.Stream STREAM = new PipelineAggregatorStreams.Stream() {
@Override
public MovAvgPipelineAggregator readResult(StreamInput in) throws IOException {
MovAvgPipelineAggregator result = new MovAvgPipelineAggregator();
result.readFrom(in);
return result;
}
};
public static void registerStreams() {
PipelineAggregatorStreams.registerStream(STREAM, TYPE.stream());
}
private static final Function FUNCTION = new Function() {
@Override
public InternalAggregation apply(Aggregation input) {
return (InternalAggregation) input;
}
};
private ValueFormatter formatter;
private GapPolicy gapPolicy;
private int window;
private MovAvgModel model;
private int predict;
private boolean minimize;
public MovAvgPipelineAggregator() {
}
public MovAvgPipelineAggregator(String name, String[] bucketsPaths, ValueFormatter formatter, GapPolicy gapPolicy,
int window, int predict, MovAvgModel model, boolean minimize, Map metadata) {
super(name, bucketsPaths, metadata);
this.formatter = formatter;
this.gapPolicy = gapPolicy;
this.window = window;
this.model = model;
this.predict = predict;
this.minimize = minimize;
}
@Override
public Type type() {
return TYPE;
}
@Override
public InternalAggregation reduce(InternalAggregation aggregation, ReduceContext reduceContext) {
InternalHistogram histo = (InternalHistogram) aggregation;
List buckets = histo.getBuckets();
InternalHistogram.Factory factory = histo.getFactory();
List newBuckets = new ArrayList<>();
EvictingQueue values = EvictingQueue.create(this.window);
long lastValidKey = 0;
int lastValidPosition = 0;
int counter = 0;
// Do we need to fit the model parameters to the data?
if (minimize) {
assert (model.canBeMinimized());
model = minimize(buckets, histo, model);
}
for (InternalHistogram.Bucket bucket : buckets) {
Double thisBucketValue = resolveBucketValue(histo, bucket, bucketsPaths()[0], gapPolicy);
// Default is to reuse existing bucket. Simplifies the rest of the logic,
// since we only change newBucket if we can add to it
InternalHistogram.Bucket newBucket = bucket;
if (!(thisBucketValue == null || thisBucketValue.equals(Double.NaN))) {
// Some models (e.g. HoltWinters) have certain preconditions that must be met
if (model.hasValue(values.size())) {
double movavg = model.next(values);
List aggs = new ArrayList<>(eagerTransform(bucket.getAggregations().asList(), AGGREGATION_TRANFORM_FUNCTION));
aggs.add(new InternalSimpleValue(name(), movavg, formatter, new ArrayList(), metaData()));
newBucket = factory.createBucket(bucket.getKey(), bucket.getDocCount(), new InternalAggregations(
aggs), bucket.getKeyed(), bucket.getFormatter());
}
if (predict > 0) {
if (bucket.getKey() instanceof Number) {
lastValidKey = ((Number) bucket.getKey()).longValue();
} else if (bucket.getKey() instanceof DateTime) {
lastValidKey = ((DateTime) bucket.getKey()).getMillis();
} else {
throw new AggregationExecutionException("Expected key of type Number or DateTime but got [" + lastValidKey + "]");
}
lastValidPosition = counter;
}
values.offer(thisBucketValue);
}
counter += 1;
newBuckets.add(newBucket);
}
if (buckets.size() > 0 && predict > 0) {
boolean keyed;
ValueFormatter formatter;
keyed = buckets.get(0).getKeyed();
formatter = buckets.get(0).getFormatter();
double[] predictions = model.predict(values, predict);
for (int i = 0; i < predictions.length; i++) {
List aggs;
long newKey = histo.getRounding().nextRoundingValue(lastValidKey);
if (lastValidPosition + i + 1 < newBuckets.size()) {
InternalHistogram.Bucket bucket = (InternalHistogram.Bucket) newBuckets.get(lastValidPosition + i + 1);
// Get the existing aggs in the bucket so we don't clobber data
aggs = new ArrayList<>(eagerTransform(bucket.getAggregations().asList(), AGGREGATION_TRANFORM_FUNCTION));
aggs.add(new InternalSimpleValue(name(), predictions[i], formatter, new ArrayList(), metaData()));
InternalHistogram.Bucket newBucket = factory.createBucket(newKey, 0, new InternalAggregations(
aggs), keyed, formatter);
// Overwrite the existing bucket with the new version
newBuckets.set(lastValidPosition + i + 1, newBucket);
} else {
// Not seen before, create fresh
aggs = new ArrayList<>();
aggs.add(new InternalSimpleValue(name(), predictions[i], formatter, new ArrayList(), metaData()));
InternalHistogram.Bucket newBucket = factory.createBucket(newKey, 0, new InternalAggregations(
aggs), keyed, formatter);
// Since this is a new bucket, simply append it
newBuckets.add(newBucket);
}
lastValidKey = newKey;
}
}
return factory.create(newBuckets, histo);
}
private MovAvgModel minimize(List buckets, InternalHistogram histo, MovAvgModel model) {
int counter = 0;
EvictingQueue values = EvictingQueue.create(window);
double[] test = new double[window];
ListIterator iter = buckets.listIterator(buckets.size());
// We have to walk the iterator backwards because we don't know if/how many buckets are empty.
while (iter.hasPrevious() && counter < window) {
Double thisBucketValue = resolveBucketValue(histo, iter.previous(), bucketsPaths()[0], gapPolicy);
if (!(thisBucketValue == null || thisBucketValue.equals(Double.NaN))) {
test[window - counter - 1] = thisBucketValue;
counter += 1;
}
}
// If we didn't fill the test set, we don't have enough data to minimize.
// Just return the model with the starting coef
if (counter < window) {
return model;
}
//And do it again, for the train set. Unfortunately we have to fill an array and then
//fill an evicting queue backwards :(
counter = 0;
double[] train = new double[window];
while (iter.hasPrevious() && counter < window) {
Double thisBucketValue = resolveBucketValue(histo, iter.previous(), bucketsPaths()[0], gapPolicy);
if (!(thisBucketValue == null || thisBucketValue.equals(Double.NaN))) {
train[window - counter - 1] = thisBucketValue;
counter += 1;
}
}
// If we didn't fill the train set, we don't have enough data to minimize.
// Just return the model with the starting coef
if (counter < window) {
return model;
}
for (double v : train) {
values.add(v);
}
return SimulatedAnealingMinimizer.minimize(model, values, test);
}
@Override
public void doReadFrom(StreamInput in) throws IOException {
formatter = ValueFormatterStreams.readOptional(in);
gapPolicy = GapPolicy.readFrom(in);
window = in.readVInt();
predict = in.readVInt();
model = MovAvgModelStreams.read(in);
minimize = in.readBoolean();
}
@Override
public void doWriteTo(StreamOutput out) throws IOException {
ValueFormatterStreams.writeOptional(formatter, out);
gapPolicy.writeTo(out);
out.writeVInt(window);
out.writeVInt(predict);
model.writeTo(out);
out.writeBoolean(minimize);
}
public static class Factory extends PipelineAggregatorFactory {
private final ValueFormatter formatter;
private GapPolicy gapPolicy;
private int window;
private MovAvgModel model;
private int predict;
private boolean minimize;
public Factory(String name, String[] bucketsPaths, ValueFormatter formatter, GapPolicy gapPolicy,
int window, int predict, MovAvgModel model, boolean minimize) {
super(name, TYPE.name(), bucketsPaths);
this.formatter = formatter;
this.gapPolicy = gapPolicy;
this.window = window;
this.model = model;
this.predict = predict;
this.minimize = minimize;
}
@Override
protected PipelineAggregator createInternal(Map metaData) throws IOException {
return new MovAvgPipelineAggregator(name, bucketsPaths, formatter, gapPolicy, window, predict, model, minimize, metaData);
}
@Override
public void doValidate(AggregatorFactory parent, AggregatorFactory[] aggFactories,
List pipelineAggregatoractories) {
if (bucketsPaths.length != 1) {
throw new IllegalStateException(PipelineAggregator.Parser.BUCKETS_PATH.getPreferredName()
+ " must contain a single entry for aggregation [" + name + "]");
}
if (!(parent instanceof HistogramAggregator.Factory)) {
throw new IllegalStateException("moving average aggregation [" + name
+ "] must have a histogram or date_histogram as parent");
} else {
HistogramAggregator.Factory histoParent = (HistogramAggregator.Factory) parent;
if (histoParent.minDocCount() != 0) {
throw new IllegalStateException("parent histogram of moving average aggregation [" + name
+ "] must have min_doc_count of 0");
}
}
}
}
}