[ML] Introduce randomize_seed setting for regression and classificati…

…on (#49990)

This adds a new `randomize_seed` for regression and classification.
When not explicitly set, the seed is randomly generated. One can
reuse the seed in a similar job in order to ensure the same docs
are picked for training.

client/rest-high-level/src/main/java/org/elasticsearch/client/ml/dataframe/Classification.java

@@ -49,6 +49,7 @@ public static Builder builder(String dependentVariable) {
     static final ParseField PREDICTION_FIELD_NAME = new ParseField("prediction_field_name");
     static final ParseField TRAINING_PERCENT = new ParseField("training_percent");
     static final ParseField NUM_TOP_CLASSES = new ParseField("num_top_classes");
+    static final ParseField RANDOMIZE_SEED = new ParseField("randomize_seed");
 
     private static final ConstructingObjectParser<Classification, Void> PARSER =
         new ConstructingObjectParser<>(
@@ -63,7 +64,8 @@ public static Builder builder(String dependentVariable) {
                 (Double) a[5],
                 (String) a[6],
                 (Double) a[7],
-                (Integer) a[8]));
+                (Integer) a[8],
+                (Long) a[9]));
 
     static {
         PARSER.declareString(ConstructingObjectParser.constructorArg(), DEPENDENT_VARIABLE);
@@ -75,6 +77,7 @@ public static Builder builder(String dependentVariable) {
         PARSER.declareString(ConstructingObjectParser.optionalConstructorArg(), PREDICTION_FIELD_NAME);
         PARSER.declareDouble(ConstructingObjectParser.optionalConstructorArg(), TRAINING_PERCENT);
         PARSER.declareInt(ConstructingObjectParser.optionalConstructorArg(), NUM_TOP_CLASSES);
+        PARSER.declareLong(ConstructingObjectParser.optionalConstructorArg(), RANDOMIZE_SEED);
     }
 
     private final String dependentVariable;
@@ -86,10 +89,11 @@ public static Builder builder(String dependentVariable) {
     private final String predictionFieldName;
     private final Double trainingPercent;
     private final Integer numTopClasses;
+    private final Long randomizeSeed;
 
     private Classification(String dependentVariable, @Nullable Double lambda, @Nullable Double gamma, @Nullable Double eta,
                            @Nullable Integer maximumNumberTrees, @Nullable Double featureBagFraction, @Nullable String predictionFieldName,
-                           @Nullable Double trainingPercent, @Nullable Integer numTopClasses) {
+                           @Nullable Double trainingPercent, @Nullable Integer numTopClasses, @Nullable Long randomizeSeed) {
         this.dependentVariable = Objects.requireNonNull(dependentVariable);
         this.lambda = lambda;
         this.gamma = gamma;
@@ -99,6 +103,7 @@ private Classification(String dependentVariable, @Nullable Double lambda, @Nulla
         this.predictionFieldName = predictionFieldName;
         this.trainingPercent = trainingPercent;
         this.numTopClasses = numTopClasses;
+        this.randomizeSeed = randomizeSeed;
     }
 
     @Override
@@ -138,6 +143,10 @@ public Double getTrainingPercent() {
         return trainingPercent;
     }
 
+    public Long getRandomizeSeed() {
+        return randomizeSeed;
+    }
+
     public Integer getNumTopClasses() {
         return numTopClasses;
     }
@@ -167,6 +176,9 @@ public XContentBuilder toXContent(XContentBuilder builder, Params params) throws
         if (trainingPercent != null) {
             builder.field(TRAINING_PERCENT.getPreferredName(), trainingPercent);
         }
+        if (randomizeSeed != null) {
+            builder.field(RANDOMIZE_SEED.getPreferredName(), randomizeSeed);
+        }
         if (numTopClasses != null) {
             builder.field(NUM_TOP_CLASSES.getPreferredName(), numTopClasses);
         }
@@ -177,7 +189,7 @@ public XContentBuilder toXContent(XContentBuilder builder, Params params) throws
     @Override
     public int hashCode() {
         return Objects.hash(dependentVariable, lambda, gamma, eta, maximumNumberTrees, featureBagFraction, predictionFieldName,
-            trainingPercent, numTopClasses);
+            trainingPercent, randomizeSeed, numTopClasses);
     }
 
     @Override
@@ -193,6 +205,7 @@ public boolean equals(Object o) {
             && Objects.equals(featureBagFraction, that.featureBagFraction)
             && Objects.equals(predictionFieldName, that.predictionFieldName)
             && Objects.equals(trainingPercent, that.trainingPercent)
+            && Objects.equals(randomizeSeed, that.randomizeSeed)
             && Objects.equals(numTopClasses, that.numTopClasses);
     }
 
@@ -211,6 +224,7 @@ public static class Builder {
         private String predictionFieldName;
         private Double trainingPercent;
         private Integer numTopClasses;
+        private Long randomizeSeed;
 
         private Builder(String dependentVariable) {
             this.dependentVariable = Objects.requireNonNull(dependentVariable);
@@ -251,14 +265,19 @@ public Builder setTrainingPercent(Double trainingPercent) {
             return this;
         }
 
+        public Builder setRandomizeSeed(Long randomizeSeed) {
+            this.randomizeSeed = randomizeSeed;
+            return this;
+        }
+
         public Builder setNumTopClasses(Integer numTopClasses) {
             this.numTopClasses = numTopClasses;
             return this;
         }
 
         public Classification build() {
             return new Classification(dependentVariable, lambda, gamma, eta, maximumNumberTrees, featureBagFraction, predictionFieldName,
-                trainingPercent, numTopClasses);
+                trainingPercent, numTopClasses, randomizeSeed);
         }
     }
 }

client/rest-high-level/src/main/java/org/elasticsearch/client/ml/dataframe/Regression.java

@@ -48,6 +48,7 @@ public static Builder builder(String dependentVariable) {
     static final ParseField FEATURE_BAG_FRACTION = new ParseField("feature_bag_fraction");
     static final ParseField PREDICTION_FIELD_NAME = new ParseField("prediction_field_name");
     static final ParseField TRAINING_PERCENT = new ParseField("training_percent");
+    static final ParseField RANDOMIZE_SEED = new ParseField("randomize_seed");
 
     private static final ConstructingObjectParser<Regression, Void> PARSER =
         new ConstructingObjectParser<>(
@@ -61,7 +62,8 @@ public static Builder builder(String dependentVariable) {
                 (Integer) a[4],
                 (Double) a[5],
                 (String) a[6],
-                (Double) a[7]));
+                (Double) a[7],
+                (Long) a[8]));
 
     static {
         PARSER.declareString(ConstructingObjectParser.constructorArg(), DEPENDENT_VARIABLE);
@@ -72,6 +74,7 @@ public static Builder builder(String dependentVariable) {
         PARSER.declareDouble(ConstructingObjectParser.optionalConstructorArg(), FEATURE_BAG_FRACTION);
         PARSER.declareString(ConstructingObjectParser.optionalConstructorArg(), PREDICTION_FIELD_NAME);
         PARSER.declareDouble(ConstructingObjectParser.optionalConstructorArg(), TRAINING_PERCENT);
+        PARSER.declareLong(ConstructingObjectParser.optionalConstructorArg(), RANDOMIZE_SEED);
     }
 
     private final String dependentVariable;
@@ -82,10 +85,11 @@ public static Builder builder(String dependentVariable) {
     private final Double featureBagFraction;
     private final String predictionFieldName;
     private final Double trainingPercent;
+    private final Long randomizeSeed;
 
     private Regression(String dependentVariable,  @Nullable Double lambda, @Nullable Double gamma, @Nullable Double eta,
                        @Nullable Integer maximumNumberTrees, @Nullable Double featureBagFraction, @Nullable String predictionFieldName,
-                       @Nullable Double trainingPercent) {
+                       @Nullable Double trainingPercent, @Nullable Long randomizeSeed) {
         this.dependentVariable = Objects.requireNonNull(dependentVariable);
         this.lambda = lambda;
         this.gamma = gamma;
@@ -94,6 +98,7 @@ private Regression(String dependentVariable,  @Nullable Double lambda, @Nullable
         this.featureBagFraction = featureBagFraction;
         this.predictionFieldName = predictionFieldName;
         this.trainingPercent = trainingPercent;
+        this.randomizeSeed = randomizeSeed;
     }
 
     @Override
@@ -133,6 +138,10 @@ public Double getTrainingPercent() {
         return trainingPercent;
     }
 
+    public Long getRandomizeSeed() {
+        return randomizeSeed;
+    }
+
     @Override
     public XContentBuilder toXContent(XContentBuilder builder, Params params) throws IOException {
         builder.startObject();
@@ -158,14 +167,17 @@ public XContentBuilder toXContent(XContentBuilder builder, Params params) throws
         if (trainingPercent != null) {
             builder.field(TRAINING_PERCENT.getPreferredName(), trainingPercent);
         }
+        if (randomizeSeed != null) {
+            builder.field(RANDOMIZE_SEED.getPreferredName(), randomizeSeed);
+        }
         builder.endObject();
         return builder;
     }
 
     @Override
     public int hashCode() {
         return Objects.hash(dependentVariable, lambda, gamma, eta, maximumNumberTrees, featureBagFraction, predictionFieldName,
-            trainingPercent);
+            trainingPercent, randomizeSeed);
     }
 
     @Override
@@ -180,7 +192,8 @@ public boolean equals(Object o) {
             && Objects.equals(maximumNumberTrees, that.maximumNumberTrees)
             && Objects.equals(featureBagFraction, that.featureBagFraction)
             && Objects.equals(predictionFieldName, that.predictionFieldName)
-            && Objects.equals(trainingPercent, that.trainingPercent);
+            && Objects.equals(trainingPercent, that.trainingPercent)
+            && Objects.equals(randomizeSeed, that.randomizeSeed);
     }
 
     @Override
@@ -197,6 +210,7 @@ public static class Builder {
         private Double featureBagFraction;
         private String predictionFieldName;
         private Double trainingPercent;
+        private Long randomizeSeed;
 
         private Builder(String dependentVariable) {
             this.dependentVariable = Objects.requireNonNull(dependentVariable);
@@ -237,9 +251,14 @@ public Builder setTrainingPercent(Double trainingPercent) {
             return this;
         }
 
+        public Builder setRandomizeSeed(Long randomizeSeed) {
+            this.randomizeSeed = randomizeSeed;
+            return this;
+        }
+
         public Regression build() {
             return new Regression(dependentVariable, lambda, gamma, eta, maximumNumberTrees, featureBagFraction, predictionFieldName,
-                trainingPercent);
+                trainingPercent, randomizeSeed);
         }
     }
 }

client/rest-high-level/src/test/java/org/elasticsearch/client/MachineLearningIT.java

@@ -1291,6 +1291,7 @@ public void testPutDataFrameAnalyticsConfig_GivenRegression() throws Exception {
             .setAnalysis(org.elasticsearch.client.ml.dataframe.Regression.builder("my_dependent_variable")
                 .setPredictionFieldName("my_dependent_variable_prediction")
                 .setTrainingPercent(80.0)
+                .setRandomizeSeed(42L)
                 .build())
             .setDescription("this is a regression")
             .build();
@@ -1326,6 +1327,7 @@ public void testPutDataFrameAnalyticsConfig_GivenClassification() throws Excepti
             .setAnalysis(org.elasticsearch.client.ml.dataframe.Classification.builder("my_dependent_variable")
                 .setPredictionFieldName("my_dependent_variable_prediction")
                 .setTrainingPercent(80.0)
+                .setRandomizeSeed(42L)
                 .setNumTopClasses(1)
                 .build())
             .setDescription("this is a classification")

client/rest-high-level/src/test/java/org/elasticsearch/client/documentation/MlClientDocumentationIT.java

@@ -2975,7 +2975,8 @@ public void testPutDataFrameAnalytics() throws Exception {
                 .setFeatureBagFraction(0.4) // <6>
                 .setPredictionFieldName("my_prediction_field_name") // <7>
                 .setTrainingPercent(50.0) // <8>
-                .setNumTopClasses(1) // <9>
+                .setRandomizeSeed(1234L) // <9>
+                .setNumTopClasses(1) // <10>
                 .build();
             // end::put-data-frame-analytics-classification
 
@@ -2988,6 +2989,7 @@ public void testPutDataFrameAnalytics() throws Exception {
                 .setFeatureBagFraction(0.4) // <6>
                 .setPredictionFieldName("my_prediction_field_name") // <7>
                 .setTrainingPercent(50.0) // <8>
+                .setRandomizeSeed(1234L) // <9>
                 .build();
             // end::put-data-frame-analytics-regression
 

client/rest-high-level/src/test/java/org/elasticsearch/client/ml/dataframe/ClassificationTests.java

@@ -34,6 +34,7 @@ public static Classification randomClassification() {
             .setFeatureBagFraction(randomBoolean() ? null : randomDoubleBetween(0.0, 1.0, false))
             .setPredictionFieldName(randomBoolean() ? null : randomAlphaOfLength(10))
             .setTrainingPercent(randomBoolean() ? null : randomDoubleBetween(1.0, 100.0, true))
+            .setRandomizeSeed(randomBoolean() ? null : randomLong())
             .setNumTopClasses(randomBoolean() ? null : randomIntBetween(0, 10))
             .build();
     }

docs/java-rest/high-level/ml/put-data-frame-analytics.asciidoc

@@ -119,7 +119,8 @@ include-tagged::{doc-tests-file}[{api}-classification]
 <6> The fraction of features which will be used when selecting a random bag for each candidate split. A double in (0, 1].
 <7> The name of the prediction field in the results object.
 <8> The percentage of training-eligible rows to be used in training. Defaults to 100%.
-<9> The number of top classes to be reported in the results. Defaults to 2.
+<9> The seed to be used by the random generator that picks which rows are used in training.
+<10> The number of top classes to be reported in the results. Defaults to 2.
 
 ===== Regression
 
@@ -138,6 +139,7 @@ include-tagged::{doc-tests-file}[{api}-regression]
 <6> The fraction of features which will be used when selecting a random bag for each candidate split. A double in (0, 1].
 <7> The name of the prediction field in the results object.
 <8> The percentage of training-eligible rows to be used in training. Defaults to 100%.
+<9> The seed to be used by the random generator that picks which rows are used in training.
 
 ==== Analyzed fields
 

docs/reference/ml/df-analytics/apis/dfanalyticsresources.asciidoc

@@ -204,6 +204,8 @@ include::{docdir}/ml/ml-shared.asciidoc[tag=prediction_field_name]
 
 include::{docdir}/ml/ml-shared.asciidoc[tag=training_percent]
 
+include::{docdir}/ml/ml-shared.asciidoc[tag=randomize_seed]
+
 
 [float]
 [[regression-resources-advanced]]
@@ -252,6 +254,8 @@ include::{docdir}/ml/ml-shared.asciidoc[tag=prediction_field_name]
 
 include::{docdir}/ml/ml-shared.asciidoc[tag=training_percent]
 
+include::{docdir}/ml/ml-shared.asciidoc[tag=randomize_seed]
+
 
 [float]
 [[classification-resources-advanced]]

docs/reference/ml/df-analytics/apis/put-dfanalytics.asciidoc

@@ -397,7 +397,8 @@ PUT _ml/data_frame/analytics/student_performance_mathematics_0.3
    {
      "regression": {
        "dependent_variable": "G3",
-       "training_percent": 70  <1>
+       "training_percent": 70,  <1>
+       "randomize_seed": 19673948271  <2>
      }
    }
 }
@@ -406,6 +407,7 @@ PUT _ml/data_frame/analytics/student_performance_mathematics_0.3
 
 <1> The `training_percent` defines the percentage of the data set that will be used 
 for training the model.
+<2> The `randomize_seed` is the seed used to randomly pick which data is used for training.
 
 
 [[ml-put-dfanalytics-example-c]]

docs/reference/ml/ml-shared.asciidoc

@@ -681,6 +681,15 @@ those that contain arrays) won’t be included in the calculation for used
 percentage. Defaults to `100`.
 end::training_percent[]
 
+tag::randomize_seed[]
+`randomize_seed`::
+(Optional, long) Defines the seed to the random generator that is used to pick
+which documents will be used for training. By default it is randomly generated.
+Set it to a specific value to ensure the same documents are used for training
+assuming other related parameters (e.g. `source`, `analyzed_fields`, etc.) are the same.
+end::randomize_seed[]
+
+
 tag::use-null[]
 Defines whether a new series is used as the null series when there is no value 
 for the by or partition fields. The default value is `false`.

x-pack/plugin/core/src/main/java/org/elasticsearch/xpack/core/ml/dataframe/DataFrameAnalyticsConfig.java

@@ -225,7 +225,8 @@ public XContentBuilder toXContent(XContentBuilder builder, Params params) throws
         builder.field(DEST.getPreferredName(), dest);
 
         builder.startObject(ANALYSIS.getPreferredName());
-        builder.field(analysis.getWriteableName(), analysis);
+        builder.field(analysis.getWriteableName(), analysis,
+            new MapParams(Collections.singletonMap(VERSION.getPreferredName(), version == null ? null : version.toString())));
         builder.endObject();
 
         if (params.paramAsBoolean(ToXContentParams.FOR_INTERNAL_STORAGE, false)) {

x-pack/plugin/core/src/main/java/org/elasticsearch/xpack/core/ml/dataframe/analyses/BoostedTreeParams.java

@@ -49,7 +49,7 @@ static void declareFields(AbstractObjectParser<?, Void> parser) {
     private final Integer maximumNumberTrees;
     private final Double featureBagFraction;
 
-    BoostedTreeParams(@Nullable Double lambda,
+    public BoostedTreeParams(@Nullable Double lambda,
                       @Nullable Double gamma,
                       @Nullable Double eta,
                       @Nullable Integer maximumNumberTrees,
@@ -76,7 +76,7 @@ static void declareFields(AbstractObjectParser<?, Void> parser) {
         this.featureBagFraction = featureBagFraction;
     }
 
-    BoostedTreeParams() {
+    public BoostedTreeParams() {
         this(null, null, null, null, null);
     }