Versions Compared

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.


Page properties


Current state: Not ready for discussion.

Discussion thread: To be added

...

Jira
serverASF JIRA
serverId5aa69414-a9e9-3523-82ec-879b028fb15b
keyFLINK-23959

...

Releaseml-2.0.0


Please keep the discussion on the mailing list rather than commenting on the wiki (wiki discussions get unwieldy fast).

Table of Contents

[This FLIP proposal is a joint work between Dong Lin and Zhipeng Zhang]

Motivation and Use-cases

The existing Flink ML library allows users to compose an Estimator/Transformer/AlgoOperator from a pipeline (i.e. linear sequence) of Estimator/Transformer/AlgoOperator. Users only need to construct this Pipeline once and generate the corresponding PipelineModel, without having to explicitly construct the fitted PipelineModel as a linear sequence of stages. However, in order to train a DAG of Estimator/Transformer/AlgoOperator and uses the trained model for inference, users currently need to construct the DAG twice, once for the training logic and once for the inference logic. This experience is inferior to the experience of training and using a chain of Estimator/Transformer/AlgoOperator. In addition to requiring more work from users, this approach is more error prone because the DAG for the training logic may be inconsistent from the DAG for the inference logic.

...

This FLIP proposes to add the Graph, GraphModel, GraphBuilder, GraphNode and TableId classes. The following code block shows the public APIs of these classes.

1) Add the TableId class to represent the input/output of a stage.

This class is necessary in order to construct the DAG before we have the concrete Tables available. And this class overrides the equals/hashCode so that it can be used as the key of a hash map.

Code Block
languagejava
public class TableId {
    private final int tableId;

    @Override
    public boolean equals(Object obj) {...}

    @Override
    public int hashCode() {...}
}


2) Add the GraphNode class.

This class contains the stage as well as the input/output of this stage in the form of TableId lists. A DAG can thus be represented as a list of GraphNodes.

Code Block
languagejava
public class GraphNode {
    public final Stage<?> stage;
    public final TableId[] estimatorInputs;
    public final TableId[] algoOpInputs;
    public final TableId[] outputs;
}


3) Add the Graph class to wrap a DAG of Estimator/Model/Transformer/AlgoOperaor into an Estimator.

Code Block
languagejava
/**
 * A Graph acts as an Estimator. A Graph consists of a DAG of stages, each of which could be an
 * Estimator, Model, Transformer or AlgoOperator. When `Graph::fit` is called, the stages are
 * executed in a topologically-sorted order. If a stage is an Estimator, its `Estimator::fit` method
 * will be called on the input tables (from the input edges) to fit a Model. Then the Model will be
 * used to transform the input tables and produce output tables to the output edges. If a stage is
 * an AlgoOperator, its `AlgoOperator::transform` method will be called on the input tables and
 * produce output tables to the output edges. The GraphModel fitted from a Graph consists of the
 * fitted Models and AlgoOperators, corresponding to the Graph's stages.
 */
@PublicEvolving
public final class Graph implements Estimator<Graph, GraphModel> {
    public Graph(List<GraphNode> nodes, TableId[] estimatorInputIds, TableId[] algoOpInputs, TableId[] outputs, TableId[] inputModelData, TableId[] outputModelData) {...}

    @Override
    public GraphModel fit(Table... inputs) {...}

    @Override
    public void save(String path) throws IOException {...}

    @Override
    public static Graph load(StreamTableEnvironment tEnv, String path) throws IOException {...}
}


4) Add the GraphModel class to wrap a DAG of Estimator/Model/Transformer/AlgoOperaor into a Model.

Code Block
languagejava
/**
 * A GraphModel acts as a Model. A GraphModel consists of a DAG of stages, each of which could be an
 * Estimator, Model, Transformer or AlgoOperators. When `GraphModel::transform` is called, the
 * stages are executed in a topologically-sorted order. When a stage is executed, its
 * `AlgoOperator::transform` method will be called on the input tables (from the input edges) and
 * produce output tables to the output edges.
 */
public final class GraphModel implements Model<GraphModel> {

    public GraphModel(List<GraphNode> nodes, TableId[] inputIds, TableId[] outputIds, TableId[] inputModelData, TableId[] outputModelData) {...}

    @Override
    public Table[] transform(Table... inputTables) {...}

    @Override
    public void setModelData(Table... inputs) {...}

    @Override
    public Table[] getModelData() {...}

    @Override
    public void save(String path) throws IOException {...}

    public static GraphModel load(StreamTableEnvironment tEnv, String path) throws IOException {...}
}


5) Add the GraphBuilder class to build GraphModel or Graph from a DAG of stages.

Code Block
languagejava
/**
 * A GraphBuilder provides APIs to build Estimator/Model/AlgoOperator from a DAG of stages, each of
 * which could be an Estimator, Model, Transformer or AlgoOperator.
 */
@PublicEvolving
public final class GraphBuilder {     

    /**
     * Specifies the loose upper bound of the number of output tables that can be returned by the
     * Model::getModelData() and AlgoOperator::transform() methods, for any stage involved in this
     * Graph.
     *
     * <p>The default upper bound is 20.
     */
    public GraphBuilder setMaxOutputTableNum(int maxOutputLength) {...}

    /**
     * Creates a TableId associated with this GraphBuilder. It can be used to specify the passing of
     * tables between stages, as well as the input/output tables of the Graph/GraphModel generated
     * by this builder.
     *
     * @return A TableId.
     */
    public TableId createTableId() {...}

    /**
     * Adds an AlgoOperator in the graph.
     *
     * <p>When the graph runs as Estimator, the transform() of the given AlgoOperator would be
     * invoked with the given inputs. Then when the GraphModel fitted by this graph runs, the
     * transform() of the given AlgoOperator would be invoked with the given inputs.
     *
     * <p>When the graph runs as AlgoOperator or Model, the transform() of the given AlgoOperator
     * would be invoked with the given inputs.
     *
     * <p>NOTE: the number of the returned TableIds does not represent the actual number of Tables
     * outputted by transform(). This number could be configured using {@link
     * #setMaxOutputTableNum(int)}. Users should make sure that this number >= the actual number of
     * Tables outputted by transform().
     *
     * @param algoOp An AlgoOperator instance.
     * @param inputs A list of TableIds which represents inputs to transform() of the given
     *     AlgoOperator.
     * @return A list of TableIds which represents the outputs of transform() of the given
     *     AlgoOperator.
     */
    public TableId[] addAlgoOperator(AlgoOperator<?> algoOp, TableId... inputs) {...}

    /**
     * Adds an Estimator in the graph.
     *
     * <p>When the graph runs as Estimator, the fit() of the given Estimator would be invoked with
     * the given inputs. Then when the GraphModel fitted by this graph runs, the transform() of the
     * Model fitted by the given Estimator would be invoked with the given inputs.
     *
     * <p>When the graph runs as AlgoOperator or Model, the fit() of the given Estimator would be
     * invoked with the given inputs, then the transform() of the Model fitted by the given
     * Estimator would be invoked with the given inputs.
     *
     * <p>NOTE: the number of the returned TableIds does not represent the actual number of Tables
     * outputted by transform(). This number could be configured using {@link
     * #setMaxOutputTableNum(int)}. Users should make sure that this number >= the actual number of
     * Tables outputted by transform().
     *
     * @param estimator An Estimator instance.
     * @param inputs A list of TableIds which represents inputs to fit() of the given Estimator as
     *     well as inputs to transform() of the Model fitted by the given Estimator.
     * @return A list of TableIds which represents the outputs of transform() of the Model fitted by
     *     the given Estimator.
     */
    public TableId[] addEstimator(Estimator<?, ?> estimator, TableId... inputs) {...}

    /**
     * Adds an Estimator in the graph.
     *
     * <p>When the graph runs as Estimator, the fit() of the given Estimator would be invoked with
     * estimatorInputs. Then when the GraphModel fitted by this graph runs, the transform() of the
     * Model fitted by the given Estimator would be invoked with modelInputs.
     *
     * <p>When the graph runs as AlgoOperator or Model, the fit() of the given Estimator would be
     * invoked with estimatorInputs, then the transform() of the Model fitted by the given Estimator
     * would be invoked with modelInputs.
     *
     * <p>NOTE: the number of the returned TableIds does not represent the actual number of Tables
     * outputted by transform(). This number could be configured using {@link
     * #setMaxOutputTableNum(int)}. Users should make sure that this number >= the actual number of
     * Tables outputted by transform().
     *
     * @param estimator An Estimator instance.
     * @param estimatorInputs A list of TableIds which represents inputs to fit() of the given
     *     Estimator.
     * @param modelInputs A list of TableIds which represents inputs to transform() of the Model
     *     fitted by the given Estimator.
     * @return A list of TableIds which represents the outputs of transform() of the Model fitted by
     *     the given Estimator.
     */
    public TableId[] addEstimator(
            Estimator<?, ?> estimator, TableId[] estimatorInputs, TableId[] modelInputs) {...}

    /**
     * When the graph runs as Estimator, it first generates a GraphModel that contains the Model
     * fitted by the given Estimator. Then when this GraphModel runs, the setModelData() of the
     * fitted Model would be invoked with the given inputs before its transform() is invoked.
     *
     * <p>When the graph runs as AlgoOperator or Model, the setModelData() of the Model fitted by
     * the given Estimator would be invoked with the given inputs before its transform() is invoked.
     *
     * @param estimator An Estimator instance.
     * @param inputs A list of TableIds which represents inputs to setModelData() of the Model
     *     fitted by the given Estimator.
     */
    public void setModelDataOnEstimator(Estimator<?, ?> estimator, TableId... inputs) {...}

    /**
     * When the graph runs as Estimator, the setModelData() of the given Model would be invoked with
     * the given inputs before its transform() is invoked. Then when the GraphModel fitted by this
     * graph runs, the setModelData() of the given Model would be invoked with the given inputs.
     *
     * <p>When the graph runs as AlgoOperator or Model, the setModelData() of the given Model would
     * be invoked with the given inputs before its transform() is invoked.
     *
     * @param model A Model instance.
     * @param inputs A list of TableIds which represents inputs to setModelData() of the given
     *     Model.
     */
    public void setModelDataOnModel(Model<?> model, TableId... inputs) {...}

    /**
     * When the graph runs as Estimator, it first generates a GraphModel that contains the Model
     * fitted by the given Estimator. Then when this GraphModel runs, the getModelData() of the
     * fitted Model would be invoked.
     *
     * <p>When the graph runs as AlgoOperator or Model, the getModelData() of the Model fitted by
     * the given Estimator would be invoked.
     *
     * <p>NOTE: the number of the returned TableIds does not represent the actual number of Tables
     * outputted by getModelData(). This number could be configured using {@link
     * #setMaxOutputTableNum(int)}. Users should make sure that this number >= the actual number of
     * Tables outputted by getModelData().
     *
     * @param estimator An Estimator instance.
     * @return A list of TableIds which represents the outputs of getModelData() of the Model fitted
     *     by the given Estimator.
     */
    public TableId[] getModelDataFromEstimator(Estimator<?, ?> estimator
Code Block
languagejava
/**
 * A Graph acts as an Estimator. A Graph consists of a DAG of stages, each of which could be an
 * Estimator, Model, Transformer or AlgoOperator. When `Graph::fit` is called, the stages are
 * executed in a topologically-sorted order. If a stage is an Estimator, its `Estimator::fit` method
 * will be called on the input tables (from the input edges) to fit a Model. Then the Model will be
 * used to transform the input tables and produce output tables to the output edges. If a stage is
 * an AlgoOperator, its `AlgoOperator::transform` method will be called on the input tables and
 * produce output tables to the output edges. The GraphModel fitted from a Graph consists of the
 * fitted Models and AlgoOperators, corresponding to the Graph's stages.
 */
@PublicEvolving
public final class Graph implements Estimator<Graph, GraphModel> {
    public Graph(List<GraphNode> nodes, TableId[] estimatorInputIds, TableId[] modelInputs, TableId[] outputs, TableId[] inputModelData, TableId[] outputModelData) {...}

    @Override
    public GraphModel fit(Table... inputs) {...}

    @Override/**
     public void save(String path) throws IOException {...}

    @Override
    public static Graph load(String path) throws IOException {...}
}

/**
 * A GraphModel acts as a Model. A GraphModel consists of a DAG of stages, each of which could be a
 * Model, Transformer or AlgoOperators. When `GraphModel::transform` is called, the stages are
 * executed in a topologically-sorted order. When a stage is executed, its `AlgoOperator::transform`
 * method will be called on the input tables (from the input edges) and produce output tables to the
 * output edges.
 */
public final class GraphModel implements Model<GraphModel> {

    public GraphModel(List<GraphNode> nodes, TableId[] inputIds, TableId[] outputIds, TableId[] inputModelData, TableId[] outputModelData) {...}

    @Override
    public Table[] transform(Table... inputTables) {...}

    @Override
    public void setModelData(Table... inputs) {...}

    @Override
    public Table[] getModelData() {...}

    @Override
    public void save(String path) throws IOException {...}

    public static GraphModel load(String path) throws IOException {...}
}


/**
 * A GraphBuilder provides APIs to build Graph/Model/AlgoOperator from a DAG of stages, each of
 * which could be an Estimator, Model, Transformer or AlgoOperator.
 */
@PublicEvolving
public final class GraphBuilder {
    private int maxOutputLength = 20;

    public GraphBuilder() {}

    /**
     * Specifies the upper bound (could be loose) of the number of output tables that can be
     * returned by the Transformer::getModelData and AlgoOperator::transform methods, for any stage
     * involved in this Graph.
     *
     * <p>The default upper bound is 20.
     */
    public GraphBuilder setMaxOutputLength(int maxOutputLength) {...}

    /**
     * Creates a TableId associated with this GraphBuilder. It can be used to specify the passing of
     * tables between stages, as well as the input/output tables of the Graph/GraphModel generated
     * by this builder* When the graph runs as Estimator, the getModelData() of the given Model would be invoked.
     * Then when the GraphModel fitted by this graph runs, the getModelData() of the given Model
     * would be invoked.
     *
     * <p>When the graph runs as AlgoOperator or Model, the getModelData() of the given Model would
     * be invoked.
     *
     * <p>NOTE: the number of the returned TableIds does not represent the actual number of Tables
     * outputted by getModelData(). This number could be configured using {@link
     * #setMaxOutputTableNum(int)}. Users should make sure that this number >= the actual number of
     * Tables outputted by getModelData().
     *
     * @param model A Model instance.
     * @return A list of TableIds which represents the outputs of getModelData() of the given Model.
     */
    public TableId[] getModelDataFromModel(Model<?> model) {...}

    /**
     * Wraps nodes of the graph into an Estimator.
     *
     * <p>When the returned Estimator runs, and when the Model fitted by the returned Estimator
     * runs, the sequence of operations recorded by the {@code addAlgoOperator(...)}, {@code
     * addEstimator(...)}, {@code setModelData(...)} and {@code getModelData(...)} would be executed
     * as specified in the Java doc of the corresponding methods.
     *
     * @param inputs A list of TableIds which represents inputs to fit() of the returned Estimator
     *     as well as inputs to transform() of the Model fitted by the returned Estimator.
     * @param outputs A list of TableIds which represents outputs of transform() of the Model fitted
     *     by the returned Estimator.
     * @return An Estimator which wraps the nodes of this graph.
     */
    public Estimator<?, ?> buildEstimator(TableId[] inputs, TableId[] createTableId(outputs) {...}

    /**
     * Wraps nodes Ifof the stagegraph isinto an Estimator,.
   both its fit*
 method and the transform method* of<p>When itsthe fitted
returned Estimator runs, and when *the Model wouldfitted beby invokedthe withreturned theEstimator
 given inputs when the graph* runs.
, the sequence of operations *
recorded by the {@code  * <p>If this stage is a Model, Transformer or AlgoOperator, its transform method would beaddAlgoOperator(...)}, {@code
     * addEstimator(...)}, {@code setModelData(...)} and {@code getModelData(...)} would be executed
     * invokedas specified within the givenJava inputsdoc whenof the graphcorresponding runsmethods.
     *
     * @param <p>Returnsinputs aA list of TableIds, which represents outputsinputs of AlgoOperator::transform of the given stage.
     */
    public TableId[] getOutputs(Stage<?> stage, TableId... inputs) {...}

    /**to fit() of the returned Estimator
     * If this stage is anas Estimator,well itsas fitinputs methodto wouldtransform() beof invokedthe withModel estimatorInputs,fitted andby the returned Estimator.
     * @param outputs A list of TableIds which represents outputs of transform() methodof ofthe itsModel fitted Model would be invoked with modelInputs
     *     by the returned Estimator.
     *
 @param inputModelData A list *of <p>ThisTableIds methodwhich throwsrepresents Exceptioninputs ifto thesetModelData() stage is not an Estimator.of the
     *
     *Model <p>Thisfitted methodby isthe usefulreturned whenEstimator.
 the state is an Estimator* AND@param the Estimator::fit needs to take
     * a different list of Tables from the Model::transform of the fitted Model.
     *outputModelData A list of TableIds which represents outputs of getModelData() of the
     *     Model fitted by the returned Estimator.
     * <p>Returns@return aAn listEstimator of TableIds, which representswraps outputs of Model::transformthe nodes of thethis fitted Modelgraph.
     */
    public TableId[] getOutputs(Stage<?> stage, TableId[] estimatorInputs, TableId[] modelInputs) {...}

Estimator<?, ?> buildEstimator(
        /**
     * The setModelData() of the fitted GraphModel should invoke the setModelData() of the given
TableId[] inputs,
            TableId[] outputs,
       *  stage with the given inputs.TableId[] inputModelData,
     */
    public void setModelData(Stage<?> stage, TableId... inputs[] outputModelData) {...}

    /**
     * The getModelData() of the fitted GraphModel should invoke the getModelData()*
     * Wraps nodes of the given
graph into an   * stageEstimator.
     *
     * <p>Returns<p>When athe listreturned ofEstimator TableIdsruns, whichand representswhen the outputsModel offitted getModelData()by ofthe thereturned givenEstimator
     * stage.
     */ runs, the sequence of operations recorded by the {@code addAlgoOperator(...)}, {@code
    public TableId[]* getModelData(Stage<?> stage) {addEstimator(...)}, {@code setModelData(...)} and {@code getModelData(...)}

 would be  /**executed
     * as specified Returnsin anthe EstimatorJava instancedoc withof the followingcorresponding behavior:methods.
     *
     * <p>1) Estimator::fit should take the given inputs and return a Model with the following@param estimatorInputs A list of TableIds which represents inputs to fit() of the returned
     *     Estimator.
     * behavior.
@param modelInputs A list of *
TableIds which represents inputs to transform() *of <p>2)the Model::transform
 should take the given inputs* and return the given outputs.
fitted by the returned  *Estimator.
     * <p>The@param outputs fitA methodlist of theTableIds returnedwhich Estimatorrepresents andoutputs theof transform() method of the Model fitted Model
     *    should invokeby the corresponding methods of the internal stages as specified by the
     * GraphBuilder.returned Estimator.
     * @param inputModelData A list of TableIds which represents inputs to setModelData() of the
     */
    public Estimator<?, ?> buildEstimator(TableId[] inputs, TableId[] outputs) {...}

Model fitted by the returned Estimator.
     /**
 @param outputModelData A list *of ReturnsTableIds anwhich Estimatorrepresents instanceoutputs withof thegetModelData() followingof behavior:the
     *
     *Model <p>1) Estimator::fit should take the given inputs and returns a Model with the following
     * behaviorfitted by the returned Estimator.
     * @return An Estimator which wraps the nodes of this graph.
     */
     * <p>2) Model::transform should take the given inputs and return the given outputs.public Estimator<?, ?> buildEstimator(
            TableId[] estimatorInputs,
     *
     * <p>3) Model::setModelData should take the given inputModelData.
 TableId[] modelInputs,
         *
   TableId[] outputs,
 * <p>4) Model::getModelData should return the given outputModelData.
    TableId[] *inputModelData,
     *    <p>The fit method of the returned Estimator and the transform/setModelData/getModelDataTableId[] outputModelData) {...}

    /**
     * methodsWraps nodes of the fittedgraph Modelinto shouldan invokeAlgoOperator.
 the corresponding methods of the internal stages as *
     * specified by<p>When the GraphBuilder.
returned AlgoOperator runs, the sequence */
of operations recorded by public Estimator<?, ?> buildEstimator(TableId[] inputs, TableId[] outputs, TableId[] inputModelData, TableId[] outputModelData) {...}

    /**the {@code
     * addAlgoOperator(...)} and {@code addEstimator(...)} would be executed as specified in the
     * ReturnsJava andoc Estimator instanceof with the followingcorresponding behavior:methods.
     *
     * <p>1) Estimator::fit should take the given estimatorInputs and returns a Model with the @param inputs A list of TableIds which represents inputs to transform() of the returned
     *     AlgoOperator.
     * following@param behavior.
outputs A list of TableIds *
which represents outputs of transform() *of <p>2) Model::transform should take the given transformerInputs and return the given outputsthe returned
     *     AlgoOperator.
     *
 @return An AlgoOperator which *wraps <p>3) Model::setModelData should take the given inputModelDatathe nodes of this graph.
     */
     * <p>4) Model::getModelData should return the given outputModelData.
public AlgoOperator<?> buildAlgoOperator(TableId[] inputs, TableId[] outputs) {...}

     /**
     * <p>TheWraps fitnodes method of the returned Estimator and the transform/setModelData/getModelData graph into a Model.
     *
     * methods<p>When of the fittedreturned Model shouldruns, invoke the correspondingsequence methodsof ofoperations therecorded internalby stagesthe as{@code
     * specified by the GraphBuilder.
     */
    public Estimator<?, ?> buildEstimator(TableId[] estimatorInputs, TableId[] modelInputs, TableId[] outputs, TableId[] inputModelData, TableId[] outputModelData) {...}

    /*addAlgoOperator(...)} and {@code addEstimator(...)} would be executed as specified in the
     * Java doc of the corresponding methods.
     *
     * Returns@param aninputs AlgoOperatorA instancelist withof theTableIds followingwhich behavior:
represents inputs to transform() of the *returned
     * <p>1) AlgoOperator::transform should take theModel.
 given inputs and returns the* given@param outputs.
 A list of TableIds *
which represents outputs of transform() *of <p>Thethe transformreturned
 method of the returned AlgoOperator* should invoke the corresponding methodsModel.
     ** @return ofA theModel internalwhich stageswraps asthe specifiednodes byof thethis GraphBuildergraph.
     */
    public AlgoOperator<Model<?> buildAlgoOperatorbuildModel(TableId[] inputs, TableId[] outputs) {...}

    /**
     * ReturnsWraps anodes Modelof instancethe withgraph theinto followinga behavior:Model.
     *
     * <p>1) Model::transform should take <p>When the givenreturned inputsModel andruns, returns the givensequence outputs.
of operations recorded by the *{@code
     * <p>The transform method of the returned Model should invoke the corresponding methods of the* addAlgoOperator(...)}, {@code addEstimator(...)}, {@code setModelData(...)} and {@code
     * internal stages getModelData(...)} would be executed as specified byin the GraphBuilder.
Java doc of the  */corresponding
    public Model<?> buildModel(TableId[] inputs, TableId[] outputs) {...}

 * methods.
     /**
     * Returns a Model instance with the following behavior: @param inputs A list of TableIds which represents inputs to transform() of the returned
     *     Model.
     * <p>1) Model::transform should take the given inputs and returns the given outputs. @param outputs A list of TableIds which represents outputs of transform() of the returned
     *     Model.
     * <p>2) Model::setModelData should take the given inputModelData.
     *@param inputModelData A list of TableIds which represents inputs to setModelData() of the
     * <p>3) Model::getModelData should return the givenreturned outputModelDataModel.
     *
 @param outputModelData A list of TableIds * <p>The transform/setModelData/getModelData methods of the returned Model should invoke thewhich represents outputs of getModelData() of the
     *     returned Model.
     * corresponding@return methodsA ofModel thewhich internalwraps stagesthe asnodes specifiedof bythis the GraphBuildergraph.
     */
    public Model<?> buildModel(TableId[] inputs, TableId[] outputs, TableId[] inputModelData, TableId[] outputModelData) {...}
}

public class GraphNode {
    public final Stage<?> stage;
    public final TableId[] estimatorInputs;inputs,
    public final TableId[] modelInputs;
    public final TableId[] outputs;
},

public class TableId {
    private final int tableId;

    @Override
    public boolean equals(Object obj) {...}

TableId[] inputModelData,
       @Override
    public int hashCode(TableId[] outputModelData) {...}
}


Example Usage

In this section we provide examples code snippets to demonstrate how we can use the APIs proposed in this FLIP to address the use-cases in the motivation section.

...

Code Block
languagejava
GraphBuilder builder = new GraphBuilder();

// Creates nodes
Stage<AlgoOperator<?> stage1 = new TransformerA();
Stage<AlgoOperator<?> stage2 = new TransformerA();
Stage<Estimator<?> stage3 = new EstimatorB();
// Creates inputs and inputStates
TableId input1 = builder.createTableId();
TableId input2 = builder.createTableId();
// Feeds inputs to nodes and gets outputs.
TableId output1 = builder.getOutputsaddAlgoOperator(stage1, input1)[0];
TableId output2 = builder.getOutputsaddAlgoOperator(stage2, input2)[0];
TableId output3 = builder.getOutputsaddEstimator(stage3, output1, output2)[0];

// Specifies the ordered lists of inputs, outputs, input states and output states that will
// be used as the inputs/outputs of the corresponding Graph and GraphTransformer APIs.
TableId[] inputs = new TableId[] {input1, input2};
TableId[] outputs = new TableId[] {output3};

// Generates the Graph instance.
Estimator<?, ?> estimator = builder.buildEstimator(inputs, outputs);
// The fit method takes 2 tables which are mapped to input1 and input2.
Model<?> model = estimator.fit(...);
// The transform method takes 2 tables which are mapped to input1 and input2.
Table[] results = model.transform(...);

...

Code Block
languagejava
GraphBuilder builder = new GraphBuilder();

// Creates nodes
Stage<Estimator<?> stage1 = new EstimatorA();
Stage<AlgoOperator<?> stage2 = new TransformerB();
// Creates inputs
TableId estimatorInput1 = builder.createTableId();
TableId estimatorInput2 = builder.createTableId();
TableId transformerInput1 = builder.createTableId();

// Feeds inputs to nodes and gets outputs.
TableId output1 = builder.getOutputsaddEstimator(stage1, new TableId[] {estimatorInput1, estimatorInput2}, new TableId[] {transformerInput1})[0];
TableId output2 = builder.getOutputsaddAlgoOperator(stage2, output1)[0];

// Specifies the ordered lists of estimator inputs, transformer inputs, outputs, input states and output states
// that will be used as the inputs/outputs of the corresponding Graph and GraphTransformer APIs.
TableId[] estimatorInputs = new TableId[] {estimatorInput1, estimatorInput2};
TableId[] transformerInputs = new TableId[] {transformerInput1};
TableId[] outputs = new TableId[] {output2};
TableId[] inputModelData = new TableId[] {};
TableId[] outputModelData = new TableId[] {};

// Generates the Graph instance.
Estimator<?, ?> estimator = builder.buildEstimator(estimatorInputs, transformerInputs, outputs, inputModelData, outputModelData);
// The fit method takes 2 tables which are mapped to estimatorInput1 and estimatorInput2.
Model<?> model = estimator.fit(...);
// The transform method takes 1 table which is mapped to transformerInput1.
Table[] results = model.transform(...);

...