org.knime.base.node.mine.knn

Class KnnNodeModel

java.lang.Object

org.knime.core.node.NodeModel

org.knime.base.node.mine.knn.KnnNodeModel

public class KnnNodeModel
extends NodeModel

This is the model for the k Nearest Neighbor node. In contrast to most learner/predictor combinations this is "all in one" since the model here really stores all of the training data.

Author:

Michael Berthold, University of Konstanz, Thorsten Meinl, University of Konstanz

Constructor Summary

Constructors

Constructor and Description
KnnNodeModel() Creates a new model for the kNN node.

Method Summary

Modifier and Type	Method and Description
protected DataTableSpec[]	configure(DataTableSpec[] inSpecs) This function is called whenever the derived model should re-configure and generate the expected output specs.
StreamableOperator	createStreamableOperator(PartitionInfo partitionInfo, PortObjectSpec[] inSpecs) Streaming API (pending): Factory method for a streamable operator that is used to execute this node.
protected BufferedDataTable[]	execute(BufferedDataTable[] inData, ExecutionContext exec) This function is invoked by the Node#executeNode() method of the node (through the #executeModel(BufferedDataTable[],ExecutionMonitor) method) only after all predecessor nodes have been successfully executed and all data is therefore available at the input ports.
InputPortRole[]	getInputPortRoles() Streaming API (pending): Defines properties on the input ports when used in a streamed and/or distributed fashion.
OutputPortRole[]	getOutputPortRoles() Streaming API (pending): Similar to NodeModel.getInputPortRoles() describes the role of the output.
protected void	loadInternals(File nodeInternDir, ExecutionMonitor exec) Load internals into the derived NodeModel.
protected void	loadValidatedSettingsFrom(NodeSettingsRO settings) Sets new settings from the passed object in the model.
protected void	reset() Override this function in the derived model and reset your NodeModel.
protected void	saveInternals(File nodeInternDir, ExecutionMonitor exec) Save internals of the derived NodeModel.
protected void	saveSettingsTo(NodeSettingsWO settings) Adds to the given NodeSettings the model specific settings.
protected void	validateSettings(NodeSettingsRO settings) Validates the settings in the passed NodeSettings object.

Methods inherited from class org.knime.core.node.NodeModel

addWarningListener, computeFinalOutputSpecs, configure, continueLoop, createInitialStreamableOperatorInternals, createMergeOperator, execute, finishStreamableExecution, getAvailableFlowVariables, getAvailableInputFlowVariables, getCredentialsProvider, getInHiLiteHandler, getInPortType, getInteractiveNodeView, getLogger, getLoopEndNode, getLoopStartNode, getNrInPorts, getNrOutPorts, getOutHiLiteHandler, getOutPortType, getWarningMessage, iterate, notifyViews, notifyWarningListeners, onDispose, peekFlowVariableDouble, peekFlowVariableInt, peekFlowVariableString, pushFlowVariableDouble, pushFlowVariableInt, pushFlowVariableString, removeWarningListener, resetAndConfigureLoopBody, setInHiLiteHandler, setWarningMessage, stateChanged

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

KnnNodeModel

public KnnNodeModel()

Creates a new model for the kNN node.

Method Detail

configure

protected DataTableSpec[] configure(DataTableSpec[] inSpecs)
                             throws InvalidSettingsException

This function is called whenever the derived model should re-configure and generate the expected output specs. Based on the given input data table spec(s) and the current model's settings, the derived model has to calculate the output data table spec and return them.

For ordinary(*) nodes the passed DataTableSpec elements are never null but can be empty. The model may return null data table spec(s) for the outputs. Note, after the model has been executed this function will not be called anymore, as the output DataTableSpecs are then being pulled from the output DataTables. A derived NodeModel that cannot provide any DataTableSpecs at its outputs before execution (because the table structure is unknown at this point) can return an array containing just null elements. As an example consider a "Transpose" node that flips columns to rows -- there is no way to determine the table spec at time of configuration as the number of rows (which is the number of new columns at the output) is unknown though the node is still executable.

(*)For nodes that support optional inputs or may have inactive outputs it's better to override NodeModel.configure(PortObjectSpec[]).

Implementation note: This method is called from the NodeModel.configure(PortObjectSpec[]) method unless that method is overwritten.

Overrides:

configure in class NodeModel

Parameters:

inSpecs - The input data table specs (as many as this model has inputs). Do NOT modify the contents of this array. If no spec is available for any given port (because the port is not connected or the previous node does not produce a spec) the framework will pass an empty DataTableSpec (no columns) unless the port is marked as optional (in which case the array element is null).

Returns:

An array of DataTableSpecs (as many as this model has outputs) They will be propagated to connected successor nodes. null DataTableSpec elements are changed to empty once.

Throws:

InvalidSettingsException - if the #configure() failed, that is, the settings are inconsistent with given DataTableSpec elements.

execute

protected BufferedDataTable[] execute(BufferedDataTable[] inData,
                                      ExecutionContext exec)
                               throws Exception

This function is invoked by the Node#executeNode() method of the node (through the #executeModel(BufferedDataTable[],ExecutionMonitor) method) only after all predecessor nodes have been successfully executed and all data is therefore available at the input ports. Implement this function with your task in the derived model.

The input data is available in the given array argument inData and is ensured to be neither null nor contain null elements (with few non-standard exception, which are described in more detail in NodeModel.execute(PortObject[], ExecutionContext)).

In order to create output data, you need to create objects of class BufferedDataTable. Use the execution context argument to create BufferedDataTable.

Overrides:

execute in class NodeModel

Parameters:

inData - An array holding DataTable elements, one for each input.

exec - The execution monitor for this execute method. It provides us with means to create new BufferedDataTable. Additionally, it should be asked frequently if the execution should be interrupted and throws an exception then. This exception might me caught, and then after closing all data streams, been thrown again. Also, if you can tell the progress of your task, just set it in this monitor.

Returns:

An array of non- null DataTable elements with the size of the number of outputs. The result of this execution.

Throws:

Exception - If you must fail the execution. Try to provide a meaningful error message in the exception as it will be displayed to the user.Please be advised to check frequently the canceled status by invoking ExecutionMonitor#checkCanceled which will throw an CanceledExcecutionException and abort the execution.

createStreamableOperator

public StreamableOperator createStreamableOperator(PartitionInfo partitionInfo,
                                                   PortObjectSpec[] inSpecs)
                                            throws InvalidSettingsException

Streaming API (pending): Factory method for a streamable operator that is used to execute this node. The default implementation returns a standard operator that wraps the NodeModel.execute(PortObject[], ExecutionContext) method. Subclasses may override it to return a new operator that follows the data handling described by the NodeModel.getInputPortRoles() and NodeModel.getOutputPortRoles() methods.

This method is called by the node executor once or multiple times depending on the input roles. If any input is distributable, the method is called multiple times (for each partition once), possibly on different (remote) clones of this NodeModel.

Overrides:

createStreamableOperator in class NodeModel

Parameters:

partitionInfo - The partition info describing the chunk (if distributable).

inSpecs - The port object specs of the input ports. These are identical to the specs that NodeModel.configure(PortObjectSpec[]) was last called with (also on the remote side).

Returns:

A new operator for the (chunk) execution.

Throws:

InvalidSettingsException - Usually not thrown in the client but still part of the method signature as implementations often run the same methods as during configure. (This method is not being called when configure fails.)

getInputPortRoles

public InputPortRole[] getInputPortRoles()

Streaming API (pending): Defines properties on the input ports when used in a streamed and/or distributed fashion.

A data input is streamed when the node implementation only needs to see each data record once (no iterative access), otherwise it's non-streamed. If a port is streamed the StreamableOperator.runFinal(PortInput[], PortOutput[], ExecutionContext) method will provide the input as a RowInput object, to which the client implementation can safely type-cast to. For non-streamed ports the input is represented by an instance of PortObjectInput. Non-data ports (not BufferedDataTable) are always non-streamed.

An data input may be distributable (= parallelizable), in which case the data is processed in paralleled (possibly scattered in the cloud). Non-data ports are always non-distributable (but the execution may still take place in a distributed fashion if another port is distributed -- any non-distributable port is then simply duplicated as required).

Overrides:

getInputPortRoles in class NodeModel

Returns:

An array with the port role for each input port. Null elements are not allowed. The default implementation marks each input as InputPortRole.NONDISTRIBUTED_NONSTREAMABLE.

getOutputPortRoles

public OutputPortRole[] getOutputPortRoles()

Streaming API (pending): Similar to NodeModel.getInputPortRoles() describes the role of the output. An output is distributable when the (distributed!) input directly maps to the output without any further merge or reduction step (which is otherwise described by the NodeModel.createMergeOperator()). Only data outputs can be distributable, any other (model) output is always non-distributable. The input- and output roles define the place where the output data is generated:

• If the input data (and hence also the output data) is non-distributable, the output will be generated by the StreamableOperator.runFinal(PortInput[], PortOutput[], ExecutionContext) method. Only one instance of the operator is used.
• If both the input and output data is distributable the output is created by the StreamableOperator.runFinal(PortInput[], PortOutput[], ExecutionContext) method, too. Note that in this case there are several instances of a StreamableOperator (either representing different threads in the same JVM or distributed in a compute cluster).
• If the input is distributable but the output is non-distributable, the output is created in the NodeModel.finishStreamableExecution(StreamableOperatorInternals, ExecutionContext, PortOutput[]) implementation. The client implementation must also overwrite the NodeModel.createMergeOperator() method. The implementation of StreamableOperator.runFinal(PortInput[], PortOutput[], ExecutionContext) must not return or push any result into the PortOutput representations.

In case a node has "mixed" outputs (e.g. a distributable data output and a non-distributable model output) the output objects are produced at two different locations: The data in the StreamableOperator instances and the model after the merge in the NodeModel.

Overrides:

getOutputPortRoles in class NodeModel

Returns:

The output roles for each of the output ports, never null. The default implementation returns OutputPortRole.NONDISTRIBUTED.

reset

protected void reset()

Override this function in the derived model and reset your NodeModel. All components should unregister themselves from any observables (at least from the hilite handler right now). All internally stored data structures should be released. User settings should not be deleted/reset though.

Specified by:

reset in class NodeModel

saveSettingsTo

protected void saveSettingsTo(NodeSettingsWO settings)

Adds to the given NodeSettings the model specific settings. The settings don't need to be complete or consistent. If, right after startup, no valid settings are available this method can write either nothing or invalid settings.

Method is called by the Node if the current settings need to be saved or transfered to the node's dialog.

Specified by:

saveSettingsTo in class NodeModel

Parameters:

settings - The object to write settings into.

See Also:

NodeModel.loadValidatedSettingsFrom(NodeSettingsRO), NodeModel.validateSettings(NodeSettingsRO)

validateSettings

protected void validateSettings(NodeSettingsRO settings)
                         throws InvalidSettingsException

Validates the settings in the passed NodeSettings object. The specified settings should be checked for completeness and consistency. It must be possible to load a settings object validated here without any exception in the #loadValidatedSettings(NodeSettings) method. The method must not change the current settings in the model - it is supposed to just check them. If some settings are missing, invalid, inconsistent, or just not right throw an exception with a message useful to the user.

Specified by:

validateSettings in class NodeModel

Parameters:

settings - The settings to validate.

Throws:

InvalidSettingsException - If the validation of the settings failed.

See Also:

NodeModel.saveSettingsTo(NodeSettingsWO), NodeModel.loadValidatedSettingsFrom(NodeSettingsRO)

loadValidatedSettingsFrom

protected void loadValidatedSettingsFrom(NodeSettingsRO settings)
                                  throws InvalidSettingsException

Sets new settings from the passed object in the model. You can safely assume that the object passed has been successfully validated by the #validateSettings(NodeSettings) method. The model must set its internal configuration according to the settings object passed.

Specified by:

loadValidatedSettingsFrom in class NodeModel

Parameters:

settings - The settings to read.

Throws:

InvalidSettingsException - If a property is not available.

See Also:

NodeModel.saveSettingsTo(NodeSettingsWO), NodeModel.validateSettings(NodeSettingsRO)

saveInternals

protected void saveInternals(File nodeInternDir,
                             ExecutionMonitor exec)
                      throws IOException,
                             CanceledExecutionException

Save internals of the derived NodeModel. This method is only called if the Node is executed. Write all your internal structures into the given file directory which are necessary to recreate this model when the workflow is loaded, e.g. view content and/or hilite mapping.

Specified by:

saveInternals in class NodeModel

Parameters:

nodeInternDir - The directory to write into.

exec - Used to report progress and to cancel the save process.

Throws:

IOException - If an error occurs during writing to this dir.

CanceledExecutionException - If the saving has been canceled.

See Also:

NodeModel.loadInternals(File,ExecutionMonitor)

loadInternals

protected void loadInternals(File nodeInternDir,
                             ExecutionMonitor exec)
                      throws IOException,
                             CanceledExecutionException

Load internals into the derived NodeModel. This method is only called if the Node was executed. Read all your internal structures from the given file directory to create your internal data structure which is necessary to provide all node functionalities after the workflow is loaded, e.g. view content and/or hilite mapping.

Specified by:

loadInternals in class NodeModel

Parameters:

nodeInternDir - The directory to read from.

exec - Used to report progress and to cancel the load process.

Throws:

IOException - If an error occurs during reading from this dir.

CanceledExecutionException - If the loading has been canceled.

See Also:

NodeModel.saveInternals(File,ExecutionMonitor)