Skip to content

Commit

Permalink
[ML] Add num_top_feature_importance_values param to regression and cl…
Browse files Browse the repository at this point in the history 
…assi… (elastic#50914)

Adds a new parameter to regression and classification that enables computation
of importance for the top most important features. The computation of the importance
is based on SHAP (SHapley Additive exPlanations) method.
  • Loading branch information
@dimitris-athanasiou @SivagurunathanV
dimitris-athanasiou authored and SivagurunathanV committed Jan 21, 2020
1 parent dfd6a20 commit 7d1e640
Show file tree
Hide file tree
Showing 19 changed files with 266 additions and 80 deletions.
38 changes: 29 additions & 9 deletions ...t/rest-high-level/src/main/java/org/elasticsearch/client/ml/dataframe/Classification.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -46,6 +46,7 @@ public static Builder builder(String dependentVariable) {
static final ParseField ETA = new ParseField("eta");
static final ParseField MAXIMUM_NUMBER_TREES = new ParseField("maximum_number_trees");
static final ParseField FEATURE_BAG_FRACTION = new ParseField("feature_bag_fraction");
static final ParseField NUM_TOP_FEATURE_IMPORTANCE_VALUES = new ParseField("num_top_feature_importance_values");
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");
Expand All @@ -62,10 +63,11 @@ public static Builder builder(String dependentVariable) {
(Double) a[3],
(Integer) a[4],
(Double) a[5],
(String) a[6],
(Double) a[7],
(Integer) a[8],
(Long) a[9]));
(Integer) a[6],
(String) a[7],
(Double) a[8],
(Integer) a[9],
(Long) a[10]));

static {
PARSER.declareString(ConstructingObjectParser.constructorArg(), DEPENDENT_VARIABLE);
Expand All @@ -74,6 +76,7 @@ public static Builder builder(String dependentVariable) {
PARSER.declareDouble(ConstructingObjectParser.optionalConstructorArg(), ETA);
PARSER.declareInt(ConstructingObjectParser.optionalConstructorArg(), MAXIMUM_NUMBER_TREES);
PARSER.declareDouble(ConstructingObjectParser.optionalConstructorArg(), FEATURE_BAG_FRACTION);
PARSER.declareInt(ConstructingObjectParser.optionalConstructorArg(), NUM_TOP_FEATURE_IMPORTANCE_VALUES);
PARSER.declareString(ConstructingObjectParser.optionalConstructorArg(), PREDICTION_FIELD_NAME);
PARSER.declareDouble(ConstructingObjectParser.optionalConstructorArg(), TRAINING_PERCENT);
PARSER.declareInt(ConstructingObjectParser.optionalConstructorArg(), NUM_TOP_CLASSES);
Expand All @@ -86,20 +89,23 @@ public static Builder builder(String dependentVariable) {
private final Double eta;
private final Integer maximumNumberTrees;
private final Double featureBagFraction;
private final Integer numTopFeatureImportanceValues;
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 Integer maximumNumberTrees, @Nullable Double featureBagFraction,
@Nullable Integer numTopFeatureImportanceValues, @Nullable String predictionFieldName,
@Nullable Double trainingPercent, @Nullable Integer numTopClasses, @Nullable Long randomizeSeed) {
this.dependentVariable = Objects.requireNonNull(dependentVariable);
this.lambda = lambda;
this.gamma = gamma;
this.eta = eta;
this.maximumNumberTrees = maximumNumberTrees;
this.featureBagFraction = featureBagFraction;
this.numTopFeatureImportanceValues = numTopFeatureImportanceValues;
this.predictionFieldName = predictionFieldName;
this.trainingPercent = trainingPercent;
this.numTopClasses = numTopClasses;
Expand Down Expand Up @@ -135,6 +141,10 @@ public Double getFeatureBagFraction() {
return featureBagFraction;
}

public Integer getNumTopFeatureImportanceValues() {
return numTopFeatureImportanceValues;
}

public String getPredictionFieldName() {
return predictionFieldName;
}
Expand Down Expand Up @@ -170,6 +180,9 @@ public XContentBuilder toXContent(XContentBuilder builder, Params params) throws
if (featureBagFraction != null) {
builder.field(FEATURE_BAG_FRACTION.getPreferredName(), featureBagFraction);
}
if (numTopFeatureImportanceValues != null) {
builder.field(NUM_TOP_FEATURE_IMPORTANCE_VALUES.getPreferredName(), numTopFeatureImportanceValues);
}
if (predictionFieldName != null) {
builder.field(PREDICTION_FIELD_NAME.getPreferredName(), predictionFieldName);
}
Expand All @@ -188,8 +201,8 @@ public XContentBuilder toXContent(XContentBuilder builder, Params params) throws

@Override
public int hashCode() {
return Objects.hash(dependentVariable, lambda, gamma, eta, maximumNumberTrees, featureBagFraction, predictionFieldName,
trainingPercent, randomizeSeed, numTopClasses);
return Objects.hash(dependentVariable, lambda, gamma, eta, maximumNumberTrees, featureBagFraction, numTopFeatureImportanceValues,
predictionFieldName, trainingPercent, randomizeSeed, numTopClasses);
}

@Override
Expand All @@ -203,6 +216,7 @@ public boolean equals(Object o) {
&& Objects.equals(eta, that.eta)
&& Objects.equals(maximumNumberTrees, that.maximumNumberTrees)
&& Objects.equals(featureBagFraction, that.featureBagFraction)
&& Objects.equals(numTopFeatureImportanceValues, that.numTopFeatureImportanceValues)
&& Objects.equals(predictionFieldName, that.predictionFieldName)
&& Objects.equals(trainingPercent, that.trainingPercent)
&& Objects.equals(randomizeSeed, that.randomizeSeed)
Expand All @@ -221,6 +235,7 @@ public static class Builder {
private Double eta;
private Integer maximumNumberTrees;
private Double featureBagFraction;
private Integer numTopFeatureImportanceValues;
private String predictionFieldName;
private Double trainingPercent;
private Integer numTopClasses;
Expand Down Expand Up @@ -255,6 +270,11 @@ public Builder setFeatureBagFraction(Double featureBagFraction) {
return this;
}

public Builder setNumTopFeatureImportanceValues(Integer numTopFeatureImportanceValues) {
this.numTopFeatureImportanceValues = numTopFeatureImportanceValues;
return this;
}

public Builder setPredictionFieldName(String predictionFieldName) {
this.predictionFieldName = predictionFieldName;
return this;
Expand All @@ -276,8 +296,8 @@ public Builder setNumTopClasses(Integer numTopClasses) {
}

public Classification build() {
return new Classification(dependentVariable, lambda, gamma, eta, maximumNumberTrees, featureBagFraction, predictionFieldName,
trainingPercent, numTopClasses, randomizeSeed);
return new Classification(dependentVariable, lambda, gamma, eta, maximumNumberTrees, featureBagFraction,
numTopFeatureImportanceValues, predictionFieldName, trainingPercent, numTopClasses, randomizeSeed);
}
}
}
38 changes: 29 additions & 9 deletions client/rest-high-level/src/main/java/org/elasticsearch/client/ml/dataframe/Regression.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -46,6 +46,7 @@ public static Builder builder(String dependentVariable) {
static final ParseField ETA = new ParseField("eta");
static final ParseField MAXIMUM_NUMBER_TREES = new ParseField("maximum_number_trees");
static final ParseField FEATURE_BAG_FRACTION = new ParseField("feature_bag_fraction");
static final ParseField NUM_TOP_FEATURE_IMPORTANCE_VALUES = new ParseField("num_top_feature_importance_values");
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");
Expand All @@ -61,9 +62,10 @@ public static Builder builder(String dependentVariable) {
(Double) a[3],
(Integer) a[4],
(Double) a[5],
(String) a[6],
(Double) a[7],
(Long) a[8]));
(Integer) a[6],
(String) a[7],
(Double) a[8],
(Long) a[9]));

static {
PARSER.declareString(ConstructingObjectParser.constructorArg(), DEPENDENT_VARIABLE);
Expand All @@ -72,6 +74,7 @@ public static Builder builder(String dependentVariable) {
PARSER.declareDouble(ConstructingObjectParser.optionalConstructorArg(), ETA);
PARSER.declareInt(ConstructingObjectParser.optionalConstructorArg(), MAXIMUM_NUMBER_TREES);
PARSER.declareDouble(ConstructingObjectParser.optionalConstructorArg(), FEATURE_BAG_FRACTION);
PARSER.declareInt(ConstructingObjectParser.optionalConstructorArg(), NUM_TOP_FEATURE_IMPORTANCE_VALUES);
PARSER.declareString(ConstructingObjectParser.optionalConstructorArg(), PREDICTION_FIELD_NAME);
PARSER.declareDouble(ConstructingObjectParser.optionalConstructorArg(), TRAINING_PERCENT);
PARSER.declareLong(ConstructingObjectParser.optionalConstructorArg(), RANDOMIZE_SEED);
Expand All @@ -83,19 +86,22 @@ public static Builder builder(String dependentVariable) {
private final Double eta;
private final Integer maximumNumberTrees;
private final Double featureBagFraction;
private final Integer numTopFeatureImportanceValues;
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,
private Regression(String dependentVariable, @Nullable Double lambda, @Nullable Double gamma, @Nullable Double eta,
@Nullable Integer maximumNumberTrees, @Nullable Double featureBagFraction,
@Nullable Integer numTopFeatureImportanceValues, @Nullable String predictionFieldName,
@Nullable Double trainingPercent, @Nullable Long randomizeSeed) {
this.dependentVariable = Objects.requireNonNull(dependentVariable);
this.lambda = lambda;
this.gamma = gamma;
this.eta = eta;
this.maximumNumberTrees = maximumNumberTrees;
this.featureBagFraction = featureBagFraction;
this.numTopFeatureImportanceValues = numTopFeatureImportanceValues;
this.predictionFieldName = predictionFieldName;
this.trainingPercent = trainingPercent;
this.randomizeSeed = randomizeSeed;
Expand Down Expand Up @@ -130,6 +136,10 @@ public Double getFeatureBagFraction() {
return featureBagFraction;
}

public Integer getNumTopFeatureImportanceValues() {
return numTopFeatureImportanceValues;
}

public String getPredictionFieldName() {
return predictionFieldName;
}
Expand Down Expand Up @@ -161,6 +171,9 @@ public XContentBuilder toXContent(XContentBuilder builder, Params params) throws
if (featureBagFraction != null) {
builder.field(FEATURE_BAG_FRACTION.getPreferredName(), featureBagFraction);
}
if (numTopFeatureImportanceValues != null) {
builder.field(NUM_TOP_FEATURE_IMPORTANCE_VALUES.getPreferredName(), numTopFeatureImportanceValues);
}
if (predictionFieldName != null) {
builder.field(PREDICTION_FIELD_NAME.getPreferredName(), predictionFieldName);
}
Expand All @@ -176,8 +189,8 @@ public XContentBuilder toXContent(XContentBuilder builder, Params params) throws

@Override
public int hashCode() {
return Objects.hash(dependentVariable, lambda, gamma, eta, maximumNumberTrees, featureBagFraction, predictionFieldName,
trainingPercent, randomizeSeed);
return Objects.hash(dependentVariable, lambda, gamma, eta, maximumNumberTrees, featureBagFraction, numTopFeatureImportanceValues,
predictionFieldName, trainingPercent, randomizeSeed);
}

@Override
Expand All @@ -191,6 +204,7 @@ public boolean equals(Object o) {
&& Objects.equals(eta, that.eta)
&& Objects.equals(maximumNumberTrees, that.maximumNumberTrees)
&& Objects.equals(featureBagFraction, that.featureBagFraction)
&& Objects.equals(numTopFeatureImportanceValues, that.numTopFeatureImportanceValues)
&& Objects.equals(predictionFieldName, that.predictionFieldName)
&& Objects.equals(trainingPercent, that.trainingPercent)
&& Objects.equals(randomizeSeed, that.randomizeSeed);
Expand All @@ -208,6 +222,7 @@ public static class Builder {
private Double eta;
private Integer maximumNumberTrees;
private Double featureBagFraction;
private Integer numTopFeatureImportanceValues;
private String predictionFieldName;
private Double trainingPercent;
private Long randomizeSeed;
Expand Down Expand Up @@ -241,6 +256,11 @@ public Builder setFeatureBagFraction(Double featureBagFraction) {
return this;
}

public Builder setNumTopFeatureImportanceValues(Integer numTopFeatureImportanceValues) {
this.numTopFeatureImportanceValues = numTopFeatureImportanceValues;
return this;
}

public Builder setPredictionFieldName(String predictionFieldName) {
this.predictionFieldName = predictionFieldName;
return this;
Expand All @@ -257,8 +277,8 @@ public Builder setRandomizeSeed(Long randomizeSeed) {
}

public Regression build() {
return new Regression(dependentVariable, lambda, gamma, eta, maximumNumberTrees, featureBagFraction, predictionFieldName,
trainingPercent, randomizeSeed);
return new Regression(dependentVariable, lambda, gamma, eta, maximumNumberTrees, featureBagFraction,
numTopFeatureImportanceValues, predictionFieldName, trainingPercent, randomizeSeed);
}
}
}
12 changes: 12 additions & 0 deletions client/rest-high-level/src/test/java/org/elasticsearch/client/MachineLearningIT.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -1294,6 +1294,12 @@ public void testPutDataFrameAnalyticsConfig_GivenRegression() throws Exception {
.setPredictionFieldName("my_dependent_variable_prediction")
.setTrainingPercent(80.0)
.setRandomizeSeed(42L)
.setLambda(1.0)
.setGamma(1.0)
.setEta(1.0)
.setMaximumNumberTrees(10)
.setFeatureBagFraction(0.5)
.setNumTopFeatureImportanceValues(3)
.build())
.setDescription("this is a regression")
.build();
Expand Down Expand Up @@ -1331,6 +1337,12 @@ public void testPutDataFrameAnalyticsConfig_GivenClassification() throws Excepti
.setTrainingPercent(80.0)
.setRandomizeSeed(42L)
.setNumTopClasses(1)
.setLambda(1.0)
.setGamma(1.0)
.setEta(1.0)
.setMaximumNumberTrees(10)
.setFeatureBagFraction(0.5)
.setNumTopFeatureImportanceValues(3)
.build())
.setDescription("this is a classification")
.build();
Expand Down
18 changes: 10 additions & 8 deletions ...h-level/src/test/java/org/elasticsearch/client/documentation/MlClientDocumentationIT.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2975,10 +2975,11 @@ public void testPutDataFrameAnalytics() throws Exception {
.setEta(5.5) // <4>
.setMaximumNumberTrees(50) // <5>
.setFeatureBagFraction(0.4) // <6>
.setPredictionFieldName("my_prediction_field_name") // <7>
.setTrainingPercent(50.0) // <8>
.setRandomizeSeed(1234L) // <9>
.setNumTopClasses(1) // <10>
.setNumTopFeatureImportanceValues(3) // <7>
.setPredictionFieldName("my_prediction_field_name") // <8>
.setTrainingPercent(50.0) // <9>
.setRandomizeSeed(1234L) // <10>
.setNumTopClasses(1) // <11>
.build();
// end::put-data-frame-analytics-classification

Expand All @@ -2989,9 +2990,10 @@ public void testPutDataFrameAnalytics() throws Exception {
.setEta(5.5) // <4>
.setMaximumNumberTrees(50) // <5>
.setFeatureBagFraction(0.4) // <6>
.setPredictionFieldName("my_prediction_field_name") // <7>
.setTrainingPercent(50.0) // <8>
.setRandomizeSeed(1234L) // <9>
.setNumTopFeatureImportanceValues(3) // <7>
.setPredictionFieldName("my_prediction_field_name") // <8>
.setTrainingPercent(50.0) // <9>
.setRandomizeSeed(1234L) // <10>
.build();
// end::put-data-frame-analytics-regression

Expand Down Expand Up @@ -3670,7 +3672,7 @@ public void testPutTrainedModel() throws Exception {
}
{
PutTrainedModelRequest request = new PutTrainedModelRequest(trainedModelConfig);

// tag::put-trained-model-execute-listener
ActionListener<PutTrainedModelResponse> listener = new ActionListener<>() {
@Override
Expand Down
1 change: 1 addition & 0 deletions ...t-high-level/src/test/java/org/elasticsearch/client/ml/dataframe/ClassificationTests.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -32,6 +32,7 @@ public static Classification randomClassification() {
.setEta(randomBoolean() ? null : randomDoubleBetween(0.001, 1.0, true))
.setMaximumNumberTrees(randomBoolean() ? null : randomIntBetween(1, 2000))
.setFeatureBagFraction(randomBoolean() ? null : randomDoubleBetween(0.0, 1.0, false))
.setNumTopFeatureImportanceValues(randomBoolean() ? null : randomIntBetween(0, Integer.MAX_VALUE))
.setPredictionFieldName(randomBoolean() ? null : randomAlphaOfLength(10))
.setTrainingPercent(randomBoolean() ? null : randomDoubleBetween(1.0, 100.0, true))
.setRandomizeSeed(randomBoolean() ? null : randomLong())
Expand Down
1 change: 1 addition & 0 deletions .../rest-high-level/src/test/java/org/elasticsearch/client/ml/dataframe/RegressionTests.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -32,6 +32,7 @@ public static Regression randomRegression() {
.setEta(randomBoolean() ? null : randomDoubleBetween(0.001, 1.0, true))
.setMaximumNumberTrees(randomBoolean() ? null : randomIntBetween(1, 2000))
.setFeatureBagFraction(randomBoolean() ? null : randomDoubleBetween(0.0, 1.0, false))
.setNumTopFeatureImportanceValues(randomBoolean() ? null : randomIntBetween(0, Integer.MAX_VALUE))
.setPredictionFieldName(randomBoolean() ? null : randomAlphaOfLength(10))
.setTrainingPercent(randomBoolean() ? null : randomDoubleBetween(1.0, 100.0, true))
.build();
Expand Down
Loading

0 comments on commit 7d1e640

Please sign in to comment.