Expose half-floats. #18887

They have been implemented in https://issues.apache.org/jira/browse/LUCENE-7289.
Ranges are implemented so that the accuracy loss only occurs at index time,
which means that if you are searching for values between A and B, the query will
match exactly all documents whose value rounded to the closest half-float point
is between A and B.

core/src/main/java/org/elasticsearch/index/fielddata/IndexNumericFieldData.java

@@ -29,6 +29,7 @@ public interface IndexNumericFieldData extends IndexFieldData<AtomicNumericField
         SHORT(false),
         INT(false),
         LONG(false),
+        HALF_FLOAT(true),
         FLOAT(true),
         DOUBLE(true);
 

core/src/main/java/org/elasticsearch/index/fielddata/plain/SortedNumericDVIndexFieldData.java

@@ -19,6 +19,7 @@
 
 package org.elasticsearch.index.fielddata.plain;
 
+import org.apache.lucene.document.HalfFloatPoint;
 import org.apache.lucene.index.DocValues;
 import org.apache.lucene.index.DocValuesType;
 import org.apache.lucene.index.LeafReader;
@@ -61,6 +62,7 @@ public class SortedNumericDVIndexFieldData extends DocValuesIndexFieldData imple
     @Override
     public org.elasticsearch.index.fielddata.IndexFieldData.XFieldComparatorSource comparatorSource(Object missingValue, MultiValueMode sortMode, Nested nested) {
         switch (numericType) {
+            case HALF_FLOAT:
             case FLOAT:
                 return new FloatValuesComparatorSource(this, missingValue, sortMode, nested);
             case DOUBLE:
@@ -87,6 +89,8 @@ public class SortedNumericDVIndexFieldData extends DocValuesIndexFieldData imple
         final String field = fieldName;
 
         switch (numericType) {
+            case HALF_FLOAT:
+                return new SortedNumericHalfFloatFieldData(reader, field);
             case FLOAT:
                 return new SortedNumericFloatFieldData(reader, field);
             case DOUBLE:
@@ -134,6 +138,95 @@ public class SortedNumericDVIndexFieldData extends DocValuesIndexFieldData imple
         }
     }
 
+    /**
+     * FieldData implementation for 16-bit float values.
+     * <p>
+     * Order of values within a document is consistent with
+     * {@link Float#compareTo(Float)}, hence the following reversible
+     * transformation is applied at both index and search:
+     * {@code bits ^ (bits >> 15) & 0x7fff}
+     * <p>
+     * Although the API is multi-valued, most codecs in Lucene specialize
+     * for the case where documents have at most one value. In this case
+     * {@link FieldData#unwrapSingleton(SortedNumericDoubleValues)} will return
+     * the underlying single-valued NumericDoubleValues representation, and
+     * {@link FieldData#unwrapSingletonBits(SortedNumericDoubleValues)} will return
+     * a Bits matching documents that have a real value (as opposed to missing).
+     */
+    static final class SortedNumericHalfFloatFieldData extends AtomicDoubleFieldData {
+        final LeafReader reader;
+        final String field;
+
+        SortedNumericHalfFloatFieldData(LeafReader reader, String field) {
+            super(0L);
+            this.reader = reader;
+            this.field = field;
+        }
+
+        @Override
+        public SortedNumericDoubleValues getDoubleValues() {
+            try {
+                SortedNumericDocValues raw = DocValues.getSortedNumeric(reader, field);
+
+                NumericDocValues single = DocValues.unwrapSingleton(raw);
+                if (single != null) {
+                    return FieldData.singleton(new SingleHalfFloatValues(single), DocValues.unwrapSingletonBits(raw));
+                } else {
+                    return new MultiHalfFloatValues(raw);
+                }
+            } catch (IOException e) {
+                throw new IllegalStateException("Cannot load doc values", e);
+            }
+        }
+
+        @Override
+        public Collection<Accountable> getChildResources() {
+            return Collections.emptyList();
+        }
+    }
+
+    /**
+     * Wraps a NumericDocValues and exposes a single 16-bit float per document.
+     */
+    static final class SingleHalfFloatValues extends NumericDoubleValues {
+        final NumericDocValues in;
+
+        SingleHalfFloatValues(NumericDocValues in) {
+            this.in = in;
+        }
+
+        @Override
+        public double get(int docID) {
+            return HalfFloatPoint.sortableShortToHalfFloat((short) in.get(docID));
+        }
+    }
+
+    /**
+     * Wraps a SortedNumericDocValues and exposes multiple 16-bit floats per document.
+     */
+    static final class MultiHalfFloatValues extends SortedNumericDoubleValues {
+        final SortedNumericDocValues in;
+
+        MultiHalfFloatValues(SortedNumericDocValues in) {
+            this.in = in;
+        }
+
+        @Override
+        public void setDocument(int doc) {
+            in.setDocument(doc);
+        }
+
+        @Override
+        public double valueAt(int index) {
+            return HalfFloatPoint.sortableShortToHalfFloat((short) in.valueAt(index));
+        }
+
+        @Override
+        public int count() {
+            return in.count();
+        }
+    }
+
     /**
      * FieldData implementation for 32-bit float values.
      * <p>

core/src/main/java/org/elasticsearch/index/mapper/core/NumberFieldMapper.java

@@ -22,6 +22,7 @@ package org.elasticsearch.index.mapper.core;
 import org.apache.lucene.document.DoublePoint;
 import org.apache.lucene.document.Field;
 import org.apache.lucene.document.FloatPoint;
+import org.apache.lucene.document.HalfFloatPoint;
 import org.apache.lucene.document.IntPoint;
 import org.apache.lucene.document.LongPoint;
 import org.apache.lucene.document.SortedNumericDocValuesField;
@@ -180,6 +181,86 @@ public class NumberFieldMapper extends FieldMapper implements AllFieldMapper.Inc
     }
 
     public enum NumberType {
+        HALF_FLOAT("half_float", NumericType.HALF_FLOAT) {
+            @Override
+            Float parse(Object value) {
+                return (Float) FLOAT.parse(value);
+            }
+
+            @Override
+            Float parse(XContentParser parser, boolean coerce) throws IOException {
+                return parser.floatValue(coerce);
+            }
+
+            @Override
+            Query termQuery(String field, Object value) {
+                float v = parse(value);
+                return HalfFloatPoint.newExactQuery(field, v);
+            }
+
+            @Override
+            Query termsQuery(String field, List<Object> values) {
+                float[] v = new float[values.size()];
+                for (int i = 0; i < values.size(); ++i) {
+                    v[i] = parse(values.get(i));
+                }
+                return HalfFloatPoint.newSetQuery(field, v);
+            }
+
+            @Override
+            Query rangeQuery(String field, Object lowerTerm, Object upperTerm,
+                             boolean includeLower, boolean includeUpper) {
+                float l = Float.NEGATIVE_INFINITY;
+                float u = Float.POSITIVE_INFINITY;
+                if (lowerTerm != null) {
+                    l = parse(lowerTerm);
+                    if (includeLower) {
+                        l = Math.nextDown(l);
+                    }
+                    l = HalfFloatPoint.nextUp(l);
+                }
+                if (upperTerm != null) {
+                    u = parse(upperTerm);
+                    if (includeUpper) {
+                        u = Math.nextUp(u);
+                    }
+                    u = HalfFloatPoint.nextDown(u);
+                }
+                return HalfFloatPoint.newRangeQuery(field, l, u);
+            }
+
+            @Override
+            public List<Field> createFields(String name, Number value,
+                                            boolean indexed, boolean docValued, boolean stored) {
+                List<Field> fields = new ArrayList<>();
+                if (indexed) {
+                    fields.add(new HalfFloatPoint(name, value.floatValue()));
+                }
+                if (docValued) {
+                    fields.add(new SortedNumericDocValuesField(name,
+                        HalfFloatPoint.halfFloatToSortableShort(value.floatValue())));
+                }
+                if (stored) {
+                    fields.add(new StoredField(name, value.floatValue()));
+                }
+                return fields;
+            }
+
+            @Override
+            FieldStats.Double stats(IndexReader reader, String fieldName,
+                                    boolean isSearchable, boolean isAggregatable) throws IOException {
+                long size = XPointValues.size(reader, fieldName);
+                if (size == 0) {
+                    return null;
+                }
+                int docCount = XPointValues.getDocCount(reader, fieldName);
+                byte[] min = XPointValues.getMinPackedValue(reader, fieldName);
+                byte[] max = XPointValues.getMaxPackedValue(reader, fieldName);
+                return new FieldStats.Double(reader.maxDoc(),docCount, -1L, size,
+                    isSearchable, isAggregatable,
+                    HalfFloatPoint.decodeDimension(min, 0), HalfFloatPoint.decodeDimension(max, 0));
+            }
+        },
         FLOAT("float", NumericType.FLOAT) {
             @Override
             Float parse(Object value) {

core/src/test/java/org/elasticsearch/fieldstats/FieldStatsIntegrationIT.java

@@ -19,6 +19,7 @@
 
 package org.elasticsearch.fieldstats;
 
+import org.apache.lucene.document.HalfFloatPoint;
 import org.apache.lucene.util.BytesRef;
 import org.elasticsearch.action.ActionRequestValidationException;
 import org.elasticsearch.action.fieldstats.FieldStats;
@@ -54,6 +55,7 @@ public class FieldStatsIntegrationIT extends ESIntegTestCase {
                     "string", "type=text",
                     "date", "type=date",
                     "double", "type=double",
+                    "half_float", "type=half_float",
                     "float", "type=float",
                     "long", "type=long",
                     "integer", "type=integer",
@@ -67,6 +69,7 @@ public class FieldStatsIntegrationIT extends ESIntegTestCase {
             "string", "type=text,index=false",
             "date", "type=date,index=false",
             "double", "type=double,index=false",
+            "half_float", "type=half_float",
             "float", "type=float,index=false",
             "long", "type=long,index=false",
             "integer", "type=integer,index=false",
@@ -81,6 +84,7 @@ public class FieldStatsIntegrationIT extends ESIntegTestCase {
             "string", "type=text,index=false",
             "date", "type=date,index=false",
             "double", "type=double,index=false",
+            "half_float", "type=half_float",
             "float", "type=float,index=false",
             "long", "type=long,index=false",
             "integer", "type=integer,index=false",
@@ -97,10 +101,12 @@ public class FieldStatsIntegrationIT extends ESIntegTestCase {
         long maxInt = Integer.MIN_VALUE;
         long minLong = Long.MAX_VALUE;
         long maxLong = Long.MIN_VALUE;
-        double minFloat = Float.MAX_VALUE;
+        double minHalfFloat = Double.POSITIVE_INFINITY;
-        double maxFloat = Float.MIN_VALUE;
+        double maxHalfFloat = Double.NEGATIVE_INFINITY;
-        double minDouble = Double.MAX_VALUE;
+        double minFloat = Double.POSITIVE_INFINITY;
-        double maxDouble = Double.MIN_VALUE;
+        double maxFloat = Double.NEGATIVE_INFINITY;
+        double minDouble = Double.POSITIVE_INFINITY;
+        double maxDouble = Double.NEGATIVE_INFINITY;
         String minString = new String(Character.toChars(1114111));
         String maxString = "0";
 
@@ -119,6 +125,10 @@ public class FieldStatsIntegrationIT extends ESIntegTestCase {
             long l = randomLong();
             minLong = Math.min(minLong, l);
             maxLong = Math.max(maxLong, l);
+            float hf = randomFloat();
+            hf = HalfFloatPoint.sortableShortToHalfFloat(HalfFloatPoint.halfFloatToSortableShort(hf));
+            minHalfFloat = Math.min(minHalfFloat, hf);
+            maxHalfFloat = Math.max(maxHalfFloat, hf);
             float f = randomFloat();
             minFloat = Math.min(minFloat, f);
             maxFloat = Math.max(maxFloat, f);
@@ -138,6 +148,7 @@ public class FieldStatsIntegrationIT extends ESIntegTestCase {
                                 "short", s,
                                 "integer", i,
                                 "long", l,
+                                "half_float", hf,
                                 "float", f,
                                 "double", d,
                                 "string", str)
@@ -147,7 +158,7 @@ public class FieldStatsIntegrationIT extends ESIntegTestCase {
 
         FieldStatsResponse response = client()
             .prepareFieldStats()
-            .setFields("byte", "short", "integer", "long", "float", "double", "string").get();
+            .setFields("byte", "short", "integer", "long", "half_float", "float", "double", "string").get();
         assertAllSuccessful(response);
 
         for (FieldStats<?> stats : response.getAllFieldStats().values()) {
@@ -164,6 +175,8 @@ public class FieldStatsIntegrationIT extends ESIntegTestCase {
         assertThat(response.getAllFieldStats().get("integer").getMaxValue(), equalTo(maxInt));
         assertThat(response.getAllFieldStats().get("long").getMinValue(), equalTo(minLong));
         assertThat(response.getAllFieldStats().get("long").getMaxValue(), equalTo(maxLong));
+        assertThat(response.getAllFieldStats().get("half_float").getMinValue(), equalTo(minHalfFloat));
+        assertThat(response.getAllFieldStats().get("half_float").getMaxValue(), equalTo(maxHalfFloat));
         assertThat(response.getAllFieldStats().get("float").getMinValue(), equalTo(minFloat));
         assertThat(response.getAllFieldStats().get("float").getMaxValue(), equalTo(maxFloat));
         assertThat(response.getAllFieldStats().get("double").getMinValue(), equalTo(minDouble));

core/src/test/java/org/elasticsearch/fieldstats/FieldStatsTests.java

@@ -108,6 +108,24 @@ public class FieldStatsTests extends ESSingleNodeTestCase {
         assertThat(result.getAllFieldStats().get(fieldName).getMinValueAsString(), equalTo(Double.toString(-1)));
     }
 
+    public void testHalfFloat() {
+        String fieldName = "field";
+        createIndex("test", Settings.EMPTY, "test", fieldName, "type=half_float");
+        for (float value = -1; value <= 9; value++) {
+            client().prepareIndex("test", "test").setSource(fieldName, value).get();
+        }
+        client().admin().indices().prepareRefresh().get();
+
+        FieldStatsResponse result = client().prepareFieldStats().setFields(fieldName).get();
+        assertThat(result.getAllFieldStats().get(fieldName).getMaxDoc(), equalTo(11L));
+        assertThat(result.getAllFieldStats().get(fieldName).getDocCount(), equalTo(11L));
+        assertThat(result.getAllFieldStats().get(fieldName).getDensity(), equalTo(100));
+        assertThat(result.getAllFieldStats().get(fieldName).getMinValue(), equalTo(-1d));
+        assertThat(result.getAllFieldStats().get(fieldName).getMaxValue(), equalTo(9d));
+        assertThat(result.getAllFieldStats().get(fieldName).getMinValueAsString(), equalTo(Float.toString(-1)));
+        assertThat(result.getAllFieldStats().get(fieldName).getMaxValueAsString(), equalTo(Float.toString(9)));
+    }
+
     public void testFloat() {
         String fieldName = "field";
         createIndex("test", Settings.EMPTY, "test", fieldName, "type=float");

core/src/test/java/org/elasticsearch/index/fielddata/plain/HalfFloatFielddataTests.java

@@ -0,0 +1,82 @@
+/*
+ * 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.index.fielddata.plain;
+
+import org.apache.lucene.document.Document;
+import org.apache.lucene.index.DirectoryReader;
+import org.apache.lucene.index.IndexWriter;
+import org.apache.lucene.index.IndexWriterConfig;
+import org.apache.lucene.index.IndexableField;
+import org.apache.lucene.index.LeafReader;
+import org.apache.lucene.store.Directory;
+import org.apache.lucene.util.IOUtils;
+import org.apache.lucene.util.TestUtil;
+import org.elasticsearch.index.fielddata.FieldData;
+import org.elasticsearch.index.fielddata.SortedNumericDoubleValues;
+import org.elasticsearch.index.mapper.core.NumberFieldMapper;
+import org.elasticsearch.test.ESTestCase;
+
+import java.io.IOException;
+
+public class HalfFloatFielddataTests extends ESTestCase {
+
+    public void testSingleValued() throws IOException {
+        Directory dir = newDirectory();
+        // we need the default codec to check for singletons
+        IndexWriter w = new IndexWriter(dir, new IndexWriterConfig(null).setCodec(TestUtil.getDefaultCodec()));
+        Document doc = new Document();
+        for (IndexableField f : NumberFieldMapper.NumberType.HALF_FLOAT.createFields("half_float", 3f, false, true, false)) {
+            doc.add(f);
+        }
+        w.addDocument(doc);
+        final DirectoryReader dirReader = DirectoryReader.open(w);
+        LeafReader reader = getOnlyLeafReader(dirReader);
+        SortedNumericDoubleValues values = new SortedNumericDVIndexFieldData.SortedNumericHalfFloatFieldData(
+                reader, "half_float").getDoubleValues();
+        assertNotNull(FieldData.unwrapSingleton(values));
+        values.setDocument(0);
+        assertEquals(1, values.count());
+        assertEquals(3f, values.valueAt(0), 0f);
+        IOUtils.close(dirReader, w, dir);
+    }
+
+    public void testMultiValued() throws IOException {
+        Directory dir = newDirectory();
+        IndexWriter w = new IndexWriter(dir, new IndexWriterConfig(null));
+        Document doc = new Document();
+        for (IndexableField f : NumberFieldMapper.NumberType.HALF_FLOAT.createFields("half_float", 3f, false, true, false)) {
+            doc.add(f);
+        }
+        for (IndexableField f : NumberFieldMapper.NumberType.HALF_FLOAT.createFields("half_float", 2f, false, true, false)) {
+            doc.add(f);
+        }
+        w.addDocument(doc);
+        final DirectoryReader dirReader = DirectoryReader.open(w);
+        LeafReader reader = getOnlyLeafReader(dirReader);
+        SortedNumericDoubleValues values = new SortedNumericDVIndexFieldData.SortedNumericHalfFloatFieldData(
+                reader, "half_float").getDoubleValues();
+        assertNull(FieldData.unwrapSingleton(values));
+        values.setDocument(0);
+        assertEquals(2, values.count());
+        assertEquals(2f, values.valueAt(0), 0f);
+        assertEquals(3f, values.valueAt(1), 0f);
+        IOUtils.close(dirReader, w, dir);
+    }
+}

core/src/test/java/org/elasticsearch/index/mapper/core/NumberFieldTypeTests.java

@@ -21,8 +21,18 @@ package org.elasticsearch.index.mapper.core;
 
 import com.carrotsearch.randomizedtesting.generators.RandomPicks;
 
+import org.apache.lucene.document.Document;
+import org.apache.lucene.document.FloatPoint;
+import org.apache.lucene.document.HalfFloatPoint;
 import org.apache.lucene.document.LongPoint;
+import org.apache.lucene.index.DirectoryReader;
 import org.apache.lucene.index.IndexOptions;
+import org.apache.lucene.index.IndexWriter;
+import org.apache.lucene.index.IndexWriterConfig;
+import org.apache.lucene.search.IndexSearcher;
+import org.apache.lucene.search.Query;
+import org.apache.lucene.store.Directory;
+import org.apache.lucene.util.IOUtils;
 import org.elasticsearch.index.mapper.FieldTypeTestCase;
 import org.elasticsearch.index.mapper.MappedFieldType;
 import org.elasticsearch.index.mapper.MappedFieldType.Relation;
@@ -81,6 +91,7 @@ public class NumberFieldTypeTests extends FieldTypeTestCase {
         assertEquals((short) 3, NumberType.SHORT.parse(3d));
         assertEquals(3, NumberType.INTEGER.parse(3d));
         assertEquals(3L, NumberType.LONG.parse(3d));
+        assertEquals(3f, NumberType.HALF_FLOAT.parse(3d));
         assertEquals(3f, NumberType.FLOAT.parse(3d));
         assertEquals(3d, NumberType.DOUBLE.parse(3d));
 
@@ -103,7 +114,39 @@ public class NumberFieldTypeTests extends FieldTypeTestCase {
         assertEquals("Value [2147483648] is out of range for an integer", e.getMessage());
         e = expectThrows(IllegalArgumentException.class, () -> NumberType.LONG.parse(10000000000000000000d));
         assertEquals("Value [1.0E19] is out of range for a long", e.getMessage());
-        assertEquals(1.1f, NumberType.FLOAT.parse(1.1)); // accuracy loss is expected
+        assertEquals(1.1f, NumberType.HALF_FLOAT.parse(1.1));
+        assertEquals(1.1f, NumberType.FLOAT.parse(1.1));
         assertEquals(1.1d, NumberType.DOUBLE.parse(1.1));
     }
+
+    public void testHalfFloatRange() throws IOException {
+        // make sure the accuracy loss of half floats only occurs at index time
+        // this test checks that searching half floats yields the same results as
+        // searching floats that are rounded to the closest half float
+        Directory dir = newDirectory();
+        IndexWriter w = new IndexWriter(dir, new IndexWriterConfig(null));
+        final int numDocs = 10000;
+        for (int i = 0; i < numDocs; ++i) {
+            Document doc = new Document();
+            float value = (randomFloat() * 2 - 1) * 70000;
+            float rounded = HalfFloatPoint.sortableShortToHalfFloat(HalfFloatPoint.halfFloatToSortableShort(value));
+            doc.add(new HalfFloatPoint("half_float", value));
+            doc.add(new FloatPoint("float", rounded));
+            w.addDocument(doc);
+        }
+        final DirectoryReader reader = DirectoryReader.open(w);
+        w.close();
+        IndexSearcher searcher = newSearcher(reader);
+        final int numQueries = 1000;
+        for (int i = 0; i < numQueries; ++i) {
+            float l = (randomFloat() * 2 - 1) * 70000;
+            float u = (randomFloat() * 2 - 1) * 70000;
+            boolean includeLower = randomBoolean();
+            boolean includeUpper = randomBoolean();
+            Query floatQ = NumberFieldMapper.NumberType.FLOAT.rangeQuery("float", l, u, includeLower, includeUpper);
+            Query halfFloatQ = NumberFieldMapper.NumberType.HALF_FLOAT.rangeQuery("half_float", l, u, includeLower, includeUpper);
+            assertEquals(searcher.count(floatQ), searcher.count(halfFloatQ));
+        }
+        IOUtils.close(reader, dir);
+    }
 }

docs/reference/mapping/types/numeric.asciidoc

@@ -4,12 +4,13 @@
 The following numeric types are supported:
 
 [horizontal]
-`long`::    A signed 64-bit integer with a minimum value of +-2^63^+ and a maximum value of +2^63^-1+.
+`long`::       A signed 64-bit integer with a minimum value of +-2^63^+ and a maximum value of +2^63^-1+.
-`integer`:: A signed 32-bit integer with a minimum value of +-2^31^+ and a maximum value of +2^31^-1+.
+`integer`::    A signed 32-bit integer with a minimum value of +-2^31^+ and a maximum value of +2^31^-1+.
-`short`::   A signed 16-bit integer with a minimum value of +-32,768+ and a maximum value of +32,767+.
+`short`::      A signed 16-bit integer with a minimum value of +-32,768+ and a maximum value of +32,767+.
-`byte`::    A signed 8-bit integer with a minimum value of +-128+ and a maximum value of +127+.
+`byte`::       A signed 8-bit integer with a minimum value of +-128+ and a maximum value of +127+.
-`double`::  A double-precision 64-bit IEEE 754 floating point.
+`double`::     A double-precision 64-bit IEEE 754 floating point.
-`float`::   A single-precision 32-bit IEEE 754 floating point.
+`float`::      A single-precision 32-bit IEEE 754 floating point.
+`half_float`:: A half-precision 16-bit IEEE 754 floating point.
 
 Below is an example of configuring a mapping with numeric fields:
 
@@ -33,6 +34,34 @@ PUT my_index
 --------------------------------------------------
 // CONSOLE
 
+==== Which type should I use?
+
+As far as integer types (`byte`, `short`, `integer` and `long`) are concerned,
+you should pick the smallest type which is enough for your use-case. This will
+help indexing and searching be more efficient. Note however that given that
+storage is optimized based on the actual values that are stored, picking one
+type over another one will have no impact on storage requirements.
+
+For floating-point types, picking the smallest type that is enough for the
+use-case will still help indexing and searching be more efficient. However,
+given that floating-point data is hard to compress, it might also have a
+significant impact on storage requirements. Here is a table that compares the
+3 floating-point types that are available in order to help make a decision.
+
+[cols="<,<,<,<",options="header",]
+|=======================================================================
+|Type |Minimum value |Maximum value |Significant bits / digits
+|`double`|+2^-1074^+ |+(2-2^-52^)·2^1023^+ |+53+ / +15.95+
+|`float`|+2^-149^+ |+(2-2^-23^)·2^127^+ |+24+ / +7.22+
+|`half_float`|+2^-24^+ |+65504+ |+11+ / +3.31+
+|=======================================================================
+
+When possible, it is often more efficient to store floating-point data into an
+integer using a scaling factor. For instance, it is more efficient to store
+percentages as integers between 0 and 100 than as floating-point numbers between 0
+and 1. Another example would be prices: it will be more efficient to store prices
+as a number of cents, which is an integer, than as a floating-point number.
+
 [[number-params]]
 ==== Parameters for numeric fields