Status

Current state: "Under Discussion"

...

JIRA: https://issues.apache.org/jira/browse/FLINK-22915

Released: 1.1415

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

NOTE: we have decided to replace this option with the design proposed here. This design doc will be deleted later.

Motivation and Use-cases

The existing Flink ML library allows users to compose an Estimator/Transformer from a pipeline (i.e. linear sequence) of Estimator/Transformer, and each Estimator/Transformer has one input and one output.

...

For example, some graph embedding algorithms (e.g., MetaPath2Vec) need to take two tables as inputs. These two tables represent nodes labels and edges of the graph respectively. This logic can be expressed as an Estimator with 2 input tables.

And some workflow may need to split 1 table into 2 tables, and use these tables for training and validation respectively. This logic can be expressed by a Transformer with 1 input table and 2 output tables.

2) Express an Estimator/Transformer pair whose input schemas are different, and still be able to compose an Estimator from a linear chain (and additionally DAG) of Estimator/Transformer.

For example, the Word2Vec algorithm uses a list of words (i.e. sentence) as the input schema for training. And the fitted algorithm should be able to use a word (instead of a list of words) as the input schema. 

Also, the GraphEmbedding algorithm is expected to take 2 tables as input for training, and take 1 table as input for inference. The input schemas are different between the corresponding Estimator and Transformer.

The existing Pipeline API can not be used to compose such an Estimator with other Transformer/Estimator because it is expected to use the same inputs for the same stage in the Pipeline.

3) Compose a directed-acyclic-graph (i.e DAG) Estimator/Transformer into an Estimator/Transformer.

For example, the workflow may involve the join of 2 tables, where each table could be generated by a chain of Estimator/Transformer. The entire workflow is therefore a DAG of Estimator/Transformer.

4) Online learning where a long-running instance Transformer needs to be updated by the latest model data generated by another long-running instance of Estimator.

In this scenario, we need to allow the Estimator to be run on a different machine than the Transformer. So that Estimator could consume sufficient computation resource in a cluster while the Transformer could be deployed on edge devices.

a generic machine learning function that has multiple inputs/outputs and make sure the class used for this purpose does not have any historical association with Estimator/Model concepts.

For example, some machine learning-related algorithms like model evaluation, computation of some statistics (e.g., ChiSquareTest), graph algorithms (e.g., PageRank) also take tables as input and produce tables as output. Though the functionality of Transformer API

can express this algorithm, some machine learning practitioners have the concern that Transformer is historically associated with the Estimator/Model/Feature Engineering concepts such that it feels weird to use Transformer for these use-cases.  

3) Online learning where a long-running instance Transformer needs to be updated by the latest model data generated by another long-running instance of Estimator.

In this scenario, we need to allow the Estimator to be run on a different machine than the Transformer. So that Estimator could consume sufficient computation resource in a cluster while the Transformer could be deployed on edge devices.

4) Provide APIs to allow Estimator/Transformer to be efficiently saved/loaded even if state (e.g. model data) of Estimator/Transformer is more than 10s of GBs.

The existing PipelineStage::toJson basically requires developer of Transformer/Estimator to serialize all model data into an in-memory string, which could be very inefficient (or practically impossible) if the model data is very large (e.g 10s of GBs).

In addition In addition to addressing the above use-cases, this FLIP also proposes a few more changes to simplify the class hierarchy and improve API usability. The existing Flink ML library has the following usability issues:

5) The Model interface (which currently simply extends the Transformer interface without adding any extra logic) does not provide any added value. The added class hierarchy complexity is not justified.

...

7) A Pipeline is both a Transformer and an Estimator. The experience of using Pipeline is therefore different from the experience of using Estimator (with the needFit API).

8) There is no API provided by the Estimator/Transformer interface to validate the schema consistency of a Pipeline. Users would have to instantiate Tables (with I/O logics) and run fit/transform to know whether the stages in the Pipeline are compatible with each other.

Background

Note: Readers who are familiar with the existing Estimator/Transformer/Pipeline APIs can skip this section.

Background

Note: Readers who are familiar with the existing Estimator/Transformer/Pipeline APIs can skip this section.

The The design proposed in this doc is built on top of the design proposed in FLIP-39: Flink ML pipeline and ML libs, which in turn is motivated by the Estimator/Transformer/Pipeline design proposed in Scikit-learn. Please feel free to read FLIP-39 and Scikit-learn paper for more detail. In this section, we explain the key background information to help better understand the motivation and benefits of the proposed design, as well as the definition of some terminology.

1) What is the motivation of defining the Estimator /and the Transformer interfaces? How are these interfaces related to machine learning?

We expect most classic machine learning algorithms to have a training logic and inference logic. The training logic of the algorithm reads training data (e.g. labeled data) and updates its variables. The inference logic of the algorithm uses the updated variables to make makes prediction (e.g. label) for the unseen data.

The inference logic of this algorithm could be represented as a subclass of Transformer (say TransformerA). The Transformer subclass has a transform() method, which reads data (as Table) and outputs the prediction result data (as Table).

The training logic of this algorithm could be represented as a subclass of Estimator (say EstimatorA). The Estimator subclass has a fit() method, which reads data (as Table) and outputs an instance of the TransformerA (defined above). The TransformerA instance contains the updated variables of this algorithm and can be used to make prediction for the unseen dataTransformer (e.g. TransformerA in this case). The TransformerA instance could be constructed using variables from the EstimatorA instance.

Then, in order to do training and prediction using this algorithm, a user could write the following code:

Estimator estimator = new EstimatorA(...)

Table prediction_result = estimator.fit(training_data).transform(inference_data)

...

2) How does Pipeline work? What is the motivation of the Pipeline class?

The Pipeline is essentially created and used as from a linear chain of Estimator/Transformer (referred to as stage below). And it is used as an Estimator (since it implements Estimator). 

...

It is important to make the following observation: if we don't provide the Pipeline class, users can still accomplish the same use-cases targeted by Pipeline by explicitly writing the training logic and inference logic separately using Estimator/Transformer APIs. But users would have to write construct this chain of Estimator/Transformer twice (for training and inference respectively).

Public Interfaces

This FLIP proposes quite a few changes and addition additions to the existing Flink ML APIs. We first describe the proposed API additions and changes, followed by the API code of interfaces and classes after making the proposed changes.

API additions and changes

Here we list the additions and the changes to the Flink ML API.

The following changes are the most important changes proposed by this doc:

1)  Removed TableEnvironment from the parameter list of fit/transform APIs.

This change simplifies the usage of fit/transform APIs.

2) Added PipelineModel and let Pipeline implement only the Estimator. Pipeline is no longer a Transformer.

This change makes the experience of using Pipeline consistent with the experience of using Estimator/Transformer, where a class is either an Estimator or a Transformer.

3) Removed Pipeline::appendStage from the Pipeline class.

This change makes the concept of Pipeline consistent with that of Graph/GraphBuilder. Neither Graph nor Pipeline provides the API to construct themselves.

4) Removed the Model interface.

This change simplifies the class hierarchy by removing a redundant class. It follows the philosophy of only adding complexity when we have explicit use-case for it.

5) Renamed PipelineStage to Stage and add the PublicEvolving tag to the Stage interface.

This change is reasonable because we will now compose Graph (not just Pipeline) using this class.

6) Added transformSchemas()  to the Stage interface.

This is needed to validate the compatibility of input schemas with a given Estimator/Transformer instance.

7) Updated Transformer/Estimator to take list of tables as inputs and return list of tables as output.

This change addresses the use-cases described in the motivation section, e.g. a graph embedding Estimator needs to take 2 tables as inputs.

8) Added setStateStreams and getStateStreams to the Transformer interface.

This change addresses the use-cases described in the motivation section, where a running Transformer needs to ingest the model state streams emitted by a Estimator, which could be running on a different machine.

9) Added Graph, GraphModel and GraphBuilder.

This change addresses the use-cases described in the motivation section, where we need to compose an Estimator from a DAG of Estimator/Transformer. Note that the Graph/GraphBuilder supports Estimator class whose input schemas are different from its fitted Transformer.

Interfaces and classes after the proposed API changes

The following code block shows the interface of Stage, Transformer, Estimator, Pipeline and PipelineModel after the proposed changes.

Added the AlgoOperator class. AlgoOperator class has the same interface as the existing Transformer (i.e. has the transform method).

This change address the need to encode a generic multi-input multi-output machine learning function. 

2) Updated Transformer/Estimator to take list of tables as inputs and return list of tables as output.

This change addresses the use-cases described in the motivation section, e.g. a graph embedding Estimator needs to take 2 tables as inputs.

3) Added setStateStreams and getStateStreams to the Transformer interface.

This change addresses the use-cases described in the motivation section, where a running Transformer needs to ingest the model state streams emitted by a Estimator, which could be running on a different machine.

4) Removed the methods PipelineStage::toJson and PipelineStage::loadJson. Add methods save(...) and load(...) to the Stage interface.

The following changes are relatively minor:

5) Removed TableEnvironment from the parameter list of fit/transform APIs.

This change simplifies the usage of fit/transform APIs.

6) Added pipelineTransformer and let Pipeline implement only the Estimator. Pipeline is no longer a Transformer.

This change makes the experience of using Pipeline consistent with the experience of using Estimator/Transformer, where a class is either an Estimator or a Transformer.

7) Removed Pipeline::appendStage from the Pipeline class.

8) Removed the Model interface. And renamed PipelineModel to PipelineTransformer.

This change simplifies the class hierarchy by removing a redundant class. It follows the philosophy of only adding complexity when we have explicit use-case for it.

9) Renamed PipelineStage to Stage and add the PublicEvolving tag to the Stage interface.

Interfaces and classes after the proposed API changes

The following code block shows the interface of Stage, Transformer, Estimator, Pipeline and pipelineTransformer after making the changes listed above.

/**
 * Base class for a stage in a Pipeline. The interface is only a concept, and does not have any actual
 * functionality. Its subclasses could be Estimator, Transformer or AlgoOperator. No other classes should inherit this
 * interface directly.
 *
 * <p>Each stage is with parameters, and requires a public empty constructor for restoration.
 *
 * @param <T> The class type of the Stage implementation itself.
 * @see WithParams
 */
@PublicEvolving
interface Stage<T extends Stage<T>> extends WithParams<T>, Serializable {
    /*

@PublicEvolving
interface Stage<T extends Stage<T>> extends WithParams<T>, Serializable {
    /**
     * This method checks the compatibility between input schemas, stage's parameters and stage's
     * logic. It should raise an exception if there is any mismatch, e.g. the number of input
     * schemas is wrong, or if a required field is missing from a schema.
     *
     * <p>IfSaves therethis isstage noto mismatch,the the method derives and returns the output schemas from the inputgiven path.
     */
    void * schemas.save(String path);

     /**
     * <p>NoteLoads thatthis thestage output schemas of afrom the given Estimatorpath.
 instance should equal the output*/
 schemas
   void load(String path);
}

/**
 of the Transformer instance fitted by this Estimator instance, suppose the same* A AlgoOperator is a Stage that takes a list of tables as inputs and produces a list of input
 * tables as results. *It schemascan arebe used as inputs to theencode fit/transforma methods respectively.
    generic multi-input multi-output machine learning function.
 *
 * @param <T> The *class @paramtype schemasof the list of schemas of the input tables. AlgoOperator implementation itself.
 */
@PublicEvolving
public interface AlgoOperator<T extends AlgoOperator<T>> extends Stage<T> {
    /**
     * @return Applies the AlgoOperator on the given listinput oftables, schemasand ofreturns the outputresult tables.
     */
    TableSchema[] transformSchemas(TableSchema... schemas);

    /** Skipped */
    default String toJson() {...}
 * @param inputs a list of tables
     * @return a list of tables
    /** Skipped */
    default void loadJson(String json) {Table[] transform(Table...} inputs);
}

@PublicEvolving/**
public interface* Transformer<TA extendsTransformer Transformer<T>>is extendsa Stage<T>AlgoOperator {

with additional support for /**
     * Applies the Transformer on the given input tables, and returns the result tables.
     *
     * @param inputs a list of tables
     * @return a list of tables
     */
    Table[] transform(Table... inputs);

state streams, which could be set by the Estimator that fitted
 * this Transformer. Unlike AlgoOperator, a Transformer is typically associated with an Estimator.
 *
 * @param <T> The class type of the Transformer implementation itself.
 */
@PublicEvolving
public interface Transformer<T extends Transformer<T>> extends AlgoOperator<T> {
    /**
     * Uses the given list of tables to update internal states. This can be useful for e.g. online
     * learning where an Estimator fits an infinite stream of training samples and streams the model
     * diff data to this Transformer.
     *
     * <p>This method may be called at most once.
     *
     * @param inputs a list of tables
     */
    default void setStateStreams(Table... inputs) {
        throw new UnsupportedOperationException("this method is not implemented");
    }

    /**
     * Gets a list of tables representing changes of internal states of this Transformer. These
     * tables might come from the Estimator that instantiated this Transformer.
     *
     * @return a list of tables
     */
    default Table[] getStateStreams() {
        throw new UnsupportedOperationException("this method is not implemented");
    }
}

@PublicEvolving/**
public interface* Estimator<EAn extendsEstimator Estimator<E,is M>,a MStage extendsthat Transformer<M>>takes extendsa Stage<E>list {

of tables as inputs /**
and produces   a Transformer.
 *
 * Trains@param on<E> theclass giventype inputsof andthe producesEstimator aimplementation Transformeritself.
 If* this@param Estimator<M> mayclass betype usedof to
the Transformer this Estimator produces.
 * compose a Pipeline, the transform method of the returned Transformer should be able to accept/
@PublicEvolving
public interface Estimator<E extends Estimator<E, M>, M extends Transformer<M>> extends Stage<E> {
    /**
     * a list of tables of Trains on the samegiven lengthinputs and schemasproduces as the fit method of this Estimatora Transformer.
     *
     * @param inputs a list of tables
     * @return a Transformer
     */
    M fit(Table... inputs);
}

@PublicEvolving/**
public final* classA Pipeline implementsacts Estimator<Pipeline,as PipelineModel> {

    public Pipeline(List<Stage<?>> stages) {...}

    an Estimator. It consists of an ordered list of stages, each of which could be
 * an Estimator, Transformer or AlgoOperator.
 */
@PublicEvolving
public final class Pipeline implements Estimator<Pipeline, pipelineTransformer> {

    public Pipeline(List<Stage<?>> stages) {...}

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

    /** Skipped a few methods, including the implementations of the Estimator APIs. */
}

@PublicEvolving/**
public final class PipelineModel implements Transformer<PipelineModel>* A pipelineTransformer acts as a Transformer. It consists of an ordered list of Transformers or AlgoOperators.
 */
@PublicEvolving
public final class pipelineTransformer implements Transformer<pipelineTransformer> {

    public PipelineModelpipelineTransformer(List<Transformer<?>> transformers) {...}

    /** Skipped a few methods, including the implementations of the Transformer APIs. */
}

The following code block shows the interface of Graph, GraphModel and GraphBuilder that we propose to add.

/**
 * A Graph acts as an Estimator. It consists of a DAG of stages, each of which is either an
 * Estimator or Transformer.
 */
@PublicEvolving
public final class Graph implements Estimator<Graph, GraphModel> {
    public Graph(...) {...}

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

    @Override
    public TableSchema[] transformSchemas(TableSchema... schemas) {
        return schemas;
    }

    /** Skipped a few methods, including the implementations of some Estimator APIs. */
}

/** A GraphModel acts as a Transformer. It consists of a DAG of Transformers. */
@PublicEvolving
public final class GraphModel implements Transformer<GraphModel> {
    /** Skipped a few methods, including the implementations of the Transformer APIs. */
}

/** A GraphBuilder helps connect Stage instances into a Graph or GraphModel. */
@PublicEvolving
public final class GraphBuilder {
    /**
     * Specifies the upper bound (could be loose) of the number of output tables that can be
     * returned by the Transformer::getStateStreams and Transformer::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.
     */
    public TableId createTableId() {...}

    /**
     * If the stage is an Estimator, both its fit method and the transform method of its fitted Transformer would be
     * invoked with the given inputs when the graph runs.
     *
     * If this stage is a Transformer, its transform method would be invoked with the given inputs when the graph runs.
     *
     * Returns a list of TableIds, which represents outputs of the Transformer::transform invocation.
     */
    public TableId[] getOutputs(Stage<?> stage, TableId... inputs) {...}

    /**
     * If this stage is an Estimator, its fit method would be invoked with estimatorInputs, and the transform method
     * of its fitted Transformer would be invoked with transformerInputs, when the graph runs.
     *
     * This method throws Exception if the stage is a Transformer.
     *
     * This method is useful when the state is an Estimator AND the Estimator::fit needs to take a different list of
     * Tables from the Transformer::transform of the fitted Transformer.
     *
     * Returns a list of TableIds, which represents outputs of the Transformer::transform invocation.
     */
    public TableId[] getOutputs(Stage<?> stage, TableId[] estimatorInputs, TableId[] transformerInputs) {...}

    /**
     * The GraphModel::setStateStreams should invoke the setStateStreams of the corresponding stage
     * with the corresponding inputs.
     */
    void setStateStreams(Stage<?> stage, TableId... inputs) {...}

    /**
     * The GraphModel::getStateStreams should invoke the getStateStreams of the corresponding stage.
     *
     * <p>Returns a list of TableIds, which represents outputs of the getStateStreams invocation.
     */
    TableId[] getStateStreams(Stage<?> stage) {...}

    /**
     * Returns a Graph instance which the following API specification:
     *
     * 1) Graph::fit should take inputs and returns a GraphModel with the following specification.
     *
     * 2) GraphModel::transform should take inputs and return outputs.
     *
     * 3) GraphModel::setStateStreams should take inputStates.
     *
     * 4) GraphModel::getStateStreams should return outputStates.
     *
     * The fit/transform/setStateStreams/getStateStreams should invoke the APIs of the internal stages in the order specified by the DAG of stages.
     */
    Graph build(TableId[] inputs, TableId[] outputs, TableId[] inputStates, TableId[] outputStates) {...}

    /**
     * Returns a Graph instance which the following API specification:
     *
     * 1) Graph::fit should take estimatorInputs and returns a GraphModel with the following specification.
     *
     * 2) GraphModel::transform should take transformerInputs and return outputs.
     *
     * 3) GraphModel::setStateStreams should take inputStates.
     *
     * 4) GraphModel::getStateStreams should return outputStates.
     *
     * The fit/transform/setStateStreams/getStateStreams should invoke the APIs of the internal stages in the order specified by the DAG of stages.
     *
     * This method is useful when the Graph::fit needs to take a different list of Tables from the GraphModel::transform of the fitted GraphModel.
     */
    Graph build(TableId[] estimatorInputs, TableId[] transformerInputs, TableId[] outputs, TableId[] inputStates, TableId[] outputStates) {...}

    /**
     * Returns a GraphModel instance which the following API specification:
     *
     * 1) GraphModel::transform should take inputs and returns outputs.
     *
     * 2) GraphModel::setStateStreams should take inputStates.
     *
     * 3) GraphModel::getStateStreams should return outputStates.
     *
     * The transform/setStateStreams/getStateStreams should invoke the APIs of the internal  stages in the order specified by the DAG of stages.
     *
     * This method throws Exception if any stage of this graph is an Estimator.
     */
    GraphModel buildModel(TableId[] inputs, TableId[] outputs, TableId[] inputStates, TableId[] outputStates) {...}

    // The TableId is necessary to pass the inputs/outputs of various API calls across the
    // Graph/GraphModel stags.
    static class TableId {}

}

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.

Composing an Estimator from a DAG of Estimator/Transformer

Suppose we have the following Transformer and Estimator classes:

• TransformerA whose transform(...) takes 1 input table and has 1 output table.
• TransformerB whose transform(...) takes 2 input tables and has 1 output table.
• EstimatorB whose fit(...) takes 2 input tables and returns an instance of TransformerB.

And we want to compose an Estimator (e.g. Graph) from the following DAG of Transformer/Estimator.

Image Removed

The resulting Graph::fit is expected to have the following behavior:

• The method takes 2 input tables. The 1st input table is given to a TransformerA instance. And the 2nd input table is given to another TransformerA instance.
• An EstimatorB instance fits the output tables of these two TransformerA instances and generates a new TransformerB instance.
• Returns a GraphModel instance which contains 2 TransformerA instance and 1 TransformerB instance, connected using the same DAG as shown above.

...

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.

Online learning by running Transformer and Estimator concurrently on different machines

Here is an online learning scenario:

• We have an infinite stream of tagged data that can be used for training.
• We have an algorithm that can be trained using this infinite stream of data. This algorithm (with its latest states/parameters) can be used to do inference. And the accuracy of the algorithm increases with the increasing amount of training data it has seen.
• We would like to train this algorithm using the given data stream on clusterA. And uses this algorithm with the update-to-date states/parameters to do inference on 10 different web servers.

In order to address this use-case, we can write the training and inference logics of this algorithm into an EstimatorA class and a TransformerA class with the following API behaviors:

• EstimatorA::fit takes a table as input and returns an instance of TransformerA. Before this method returns this transformerA, it calls transformerA.setStateStreams(state_table), where the state_table represents the stream of algorithm parameters changes produced by EstimatorA.
• TransformerA::setStateStreams(...) takes a table as input. Its implementation reads the data from this table to continuously update its algorithm parameters.
• TransformerA::getStateStreams(...) returns the same table instance that has been provided via TransformerA::setStateStreams(...).
• TransformerA::transform takes a table as input and returns a table. The returned table represents the inference results.

Here are the code snippets that address this use-case by using the proposed APIs.

First run the following code on clusterA:

void runTrainingOnClusterA(...) {
  // Creates the training stream from a Kafka topic.
  Table training_stream = ...;

  Estimator estimator = new EstimatorA(...);
  Transformer transformer = estimator.fit(training_stream);
  Table state_stream = transformer.getStateStreams()[0];

  // Writes the state_stream to a Kafka topicA.
  state_stream.sinkTable(...);
  // Saves transformer's state/metadata to a remote path.
  transformer.save(remote_path);

  // Executes the operators generated by the Estimator::fit(...), which reads from training_stream and writes to state_stream.
  env.execute()
}

Then run the following code on each web server:

void runInferenceOnWebServer(...) {
  // Creates the state stream from Kafka topicA which is written by the above code snippet. 
  Table state_stream = ...;
  // Creates the input stream that needs inference.
  Table input_stream = ...;

  Transformer transformer = new Transformer(...);
  transformer.load(remote_path);
  transformer.setStateStreams(new Table[]{state_stream});
  Table output_stream = transformer.transform(input_stream);

  // Do something with the output_stream.

  // Executes the operators generated by the Transformer::transform(...), which reads from state_stream to update its parameters. 
  // It also does inference on input_stream and produces results to the output_stream.
  env.execute()
}

GraphBuilder builder = new GraphBuilder();

// Creates nodes
Stage<?> stage1 = new TransformerA();
Stage<?> stage2 = new TransformerA();
Stage<?> 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.getOutputs(stage1, input1)[0];
TableId output2 = builder.getOutputs(stage2, input2)[0];
TableId output3 = builder.getOutputs(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 GraphModel APIs.
TableId[] inputs = new TableId[] {input1, input2};
TableId[] outputs = new TableId[] {output3};

// Generates the Graph instance.
Graph graph = builder.build(inputs, outputs, new TableId[]{}, new TableId[]{});
// The fit method takes 2 tables which are mapped to input1 and input2.
GraphModel model = graph.fit(...);
// The transform method takes 2 tables which are mapped to input1 and input2.
Table[] results = model.transform(...);

Online learning by running Transformer and Estimator concurrently on different machines

Here is an online learning scenario:

• We have an infinite stream of tagged data that can be used for training.
• We have an algorithm that can be trained using this infinite stream of data. This algorithm (with its latest states/parameters) can be used to do inference. And the accuracy of the algorithm increases with the increasing amount of training data it has seen.
• We would like to train this algorithm using the given data stream on clusterA. And uses this algorithm with the update-to-date states/parameters to do inference on 10 different web servers.

In order to address this use-case, we can write the training and inference logics of this algorithm into an EstimatorA class and a TransformerA class with the following API behaviors:

• EstimatorA::fit takes a table as input and returns an instance of TransformerA. Before this method returns this transformerA, it calls transformerA.setStateStreams(state_table), where the state_table represents the stream of algorithm parameters changes produced by EstimatorA.
• TransformerA::setStateStreams(...) takes a table as input. Its implementation reads the data from this table to continuously update its algorithm parameters.
• TransformerA::getStateStreams(...) returns the same table instance that has been provided via TransformerA::setStateStreams(...).
• TransformerA::transform takes a table as input and returns a table. The returned table represents the inference results.

...

First run the following code on clusterA:

void runTrainingOnClusterA(...) {
  // Creates the training stream from a Kafka topic.
  Table training_stream = ...;

  Estimator estimator = new EstimatorA(...);
  Transformer transformer = estimator.fit(training_stream);
  Table state_stream = transformer.getStateStreams()[0];
  String transformer_json = transformer.toJson();

  // Writes the state_stream to a Kafka topicA.
  // Writes transformer_json to a remote file.

  // Executes the operators generated by the Estimator::fit(...), which reads from training_stream and writes to state_stream.
  env.execute()
}

Then run the following code on each web server:

void runInferenceOnWebServer(...) {

  // Reads the transformer_json from the same remote file written by the above code snippet.
  String transformer_json = ...;
  // Creates the state stream from Kafka topicA which is written by the above code snippet. 
  Table state_stream = ...;
  // Creates the input stream that needs inference.
  Table input_stream = ...;

  Transformer transformer = new Transformer(...);
  transformer.loadJson(transformer_json);
  transformer.setStateStreams(new Table[]{state_stream});
  Table output_stream = transformer.transform(input_stream);

  // Do something with the output_stream.

  // Executes the operators generated by the Transformer::transform(...), which reads from state_stream to update its parameters. It also does inference on input_stream and produces results to the output_stream.
  env.execute()
}

Compose an Estimator from a chain of Estimator/Transformer whose input schemas are different from its fitted Transformer 

Suppose we have the following Estimator and Transformer classes where an Estimator's input schemas could be different from the input schema of its fitted Transformer:

• TransformerA whose transform(...) takes 1 input table and has 1 output table.
• EstimatorA whose fit(...) takes 2 input tables and returns an instance of TransformerA.
• TransformerB whose transform(...) takes 1 input table and has 1 output table.

And we want to compose an Estimator (e.g. Graph) from the following DAG of Transformer/Estimator.

Image Removed

The resulting Graph::fit is expected to have the following behavior:

• The method takes 2 input tables. Both tables are given to EstimatorA::fit.
• EstimatorA fits the input tables and generates a TransformerA instance. The TransformerA instance takes 1 table input, which is different from the 2 tables given to the EstimatorA.
• Returns a GraphModel instance which contains a TransformerA instance and a TransformerB instance, which are connected as a chain.

Notes:

• The fitted GraphModel takes only 1 table as input whereas the Graph takes 2 tables as inputs.
• The proposed APIs also support composing an Estimator from a DAG of Estimator/Transformer whose input schemas are different from its fitted Transformer. 

Here is the code snippet that addresses this use-case by using the proposed APIs:

GraphBuilder builder = new GraphBuilder();

// Creates nodes
Stage<?> stage1 = new EstimatorA();
Stage<?> 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.getOutputs(stage1, new TableId[] {estimatorInput1, estimatorInput2}, new TableId[] {transformerInput1})[0];
TableId output2 = builder.getOutputs(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 GraphModel APIs.
TableId[] estimatorInputs = new TableId[] {estimatorInput1, estimatorInput2};
TableId[] transformerInputs = new TableId[] {transformerInput1};
TableId[] outputs = new TableId[] {output2};
TableId[] inputStates = new TableId[] {};
TableId[] outputStates = new TableId[] {};

// Generates the Graph instance.
Graph graph = builder.build(estimatorInputs, transformerInputs, outputs, inputStates, outputStates);
// The fit method takes 2 tables which are mapped to estimatorInput1 and estimatorInput2.
GraphModel model = graph.fit(...);
// The transform method takes 1 table which is mapped to transformerInput1.
Table[] results = model.transform(...);

Compatibility, Deprecation, and Migration Plan

The changes proposed in this FLIP is backward incompatible with the existing APIs. We propose to change the APIs directly without deprecation period. And we will manually migrate the existing open source projects which use the existing Flink ML API to use the proposed APIs.

...

To our best knowledge, the only open source project that uses the Flink ML API is https://github.com/alibaba/Alink. We will work together with Alink developers to migrate the existing code to use the proposed API. Furthermore, we will migrate Alink's Estimator/Transformer implementation to the Flink ML library codebase as much as possible.

Test Plan

We will provide unit tests to validate the proposed changes.

Rejected Alternatives

There is no rejected alternatives to be listed here yet.

...