org.tribuo.interop.tensorflow

Class TensorFlowSavedModelExternalModel<T extends Output<T>>

java.lang.Object

org.tribuo.Model<T>

org.tribuo.interop.ExternalModel<T,TensorMap,TensorMap>

org.tribuo.interop.tensorflow.TensorFlowSavedModelExternalModel<T>

All Implemented Interfaces:

com.oracle.labs.mlrg.olcut.provenance.Provenancable<ModelProvenance>, Closeable, Serializable, AutoCloseable

public final class TensorFlowSavedModelExternalModel<T extends Output<T>>
extends ExternalModel<T,TensorMap,TensorMap>
implements Closeable

A Tribuo wrapper around a TensorFlow saved model bundle.

The model's serialVersionUID is set to the major TensorFlow version number times 100.

N.B. TensorFlow support is experimental and may change without a major version bump.

See Also:

Serialized Form

Field Summary

Fields inherited from class org.tribuo.interop.ExternalModel

DEFAULT_BATCH_SIZE, featureBackwardMapping, featureForwardMapping

Fields inherited from class org.tribuo.Model

ALL_OUTPUTS, BIAS_FEATURE, featureIDMap, generatesProbabilities, name, outputIDInfo, provenance, provenanceOutput

Method Summary

Modifier and Type	Method and Description
void	close()
protected TensorMap	convertFeatures(SparseVector input) Converts from a SparseVector using the external model's indices into the ingestion format for the external model.
protected TensorMap	convertFeaturesList(List<SparseVector> input) Converts from a list of SparseVector using the external model's indices into the ingestion format for the external model.
protected List<Prediction<T>>	convertOutput(TensorMap output, int[] numValidFeatures, List<Example<T>> examples) Converts a tensor into a prediction.
protected Prediction<T>	convertOutput(TensorMap output, int numValidFeatures, Example<T> example) Converts a tensor into a prediction.
protected Model<T>	copy(String newName, ModelProvenance newProvenance) Copies a model, replacing it's provenance and name with the supplied values.
static <T extends Output<T>> TensorFlowSavedModelExternalModel<T>	createTensorflowModel(OutputFactory<T> factory, Map<String,Integer> featureMapping, Map<T,Integer> outputMapping, String outputName, FeatureConverter featureConverter, OutputConverter<T> outputConverter, String bundleDirectory) Creates a TensorflowSavedModelExternalModel by loading in a SavedModelBundle.
protected TensorMap	externalPrediction(TensorMap input) Runs the session to make a prediction.
Map<String,List<com.oracle.labs.mlrg.olcut.util.Pair<String,Double>>>	getTopFeatures(int n) Gets the top n features associated with this model.

Methods inherited from class org.tribuo.interop.ExternalModel

createFeatureMap, createOutputInfo, getBatchSize, getExcuse, innerPredict, predict, setBatchSize

Methods inherited from class org.tribuo.Model

copy, generatesProbabilities, getExcuses, getFeatureIDMap, getName, getOutputIDInfo, getProvenance, predict, predict, setName, toString, validate

Methods inherited from class java.lang.Object

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

Method Detail

convertFeatures

protected TensorMap convertFeatures(SparseVector input)

Description copied from class: ExternalModel

Converts from a SparseVector using the external model's indices into the ingestion format for the external model.

Specified by:

convertFeatures in class ExternalModel<T extends Output<T>,TensorMap,TensorMap>

Parameters:

input - The features using external indices.

Returns:

The ingestion format for the external model.

convertFeaturesList

protected TensorMap convertFeaturesList(List<SparseVector> input)

Description copied from class: ExternalModel

Converts from a list of SparseVector using the external model's indices into the ingestion format for the external model.

Specified by:

convertFeaturesList in class ExternalModel<T extends Output<T>,TensorMap,TensorMap>

Parameters:

input - The features using external indices.

Returns:

The ingestion format for the external model.

externalPrediction

protected TensorMap externalPrediction(TensorMap input)

Runs the session to make a prediction.

Closes the input tensor after the prediction has been made.

Specified by:

externalPrediction in class ExternalModel<T extends Output<T>,TensorMap,TensorMap>

Parameters:

input - The input in the external model's format.

Returns:

A tensor representing the output.

convertOutput

protected Prediction<T> convertOutput(TensorMap output,
                                      int numValidFeatures,
                                      Example<T> example)

Converts a tensor into a prediction. Closes the output tensor after it's been converted.

Specified by:

convertOutput in class ExternalModel<T extends Output<T>,TensorMap,TensorMap>

Parameters:

output - The output of the external model.

numValidFeatures - The number of valid features in the input.

example - The input example, used to construct the Prediction.

Returns:

A Prediction representing this tensor output.

convertOutput

protected List<Prediction<T>> convertOutput(TensorMap output,
                                            int[] numValidFeatures,
                                            List<Example<T>> examples)

Converts a tensor into a prediction. Closes the output tensor after it's been converted.

Specified by:

convertOutput in class ExternalModel<T extends Output<T>,TensorMap,TensorMap>

Parameters:

output - The output of the external model.

numValidFeatures - An array with the number of valid features in each example.

examples - The input examples, used to construct the Predictions.

Returns:

A list of Prediction representing this tensor output.

getTopFeatures

public Map<String,List<com.oracle.labs.mlrg.olcut.util.Pair<String,Double>>> getTopFeatures(int n)

Description copied from class: Model

Gets the top n features associated with this model.

If the model does not produce per output feature lists, it returns a map with a single element with key Model.ALL_OUTPUTS.

If the model cannot describe it's top features then it returns Collections.emptyMap().

Specified by:

getTopFeatures in class Model<T extends Output<T>>

Parameters:

n - the number of features to return. If this value is less than 0, all features should be returned for each class, unless the model cannot score it's features.

Returns:

a map from string outputs to an ordered list of pairs of feature names and weights associated with that feature in the model

copy

protected Model<T> copy(String newName,
                        ModelProvenance newProvenance)

Description copied from class: Model

Copies a model, replacing it's provenance and name with the supplied values.

Used to provide the provenance removal functionality.

Specified by:

copy in class Model<T extends Output<T>>

Parameters:

newName - The new name.

newProvenance - The new provenance.

Returns:

A copy of the model.

close

public void close()

Specified by:

close in interface Closeable

Specified by:

close in interface AutoCloseable

createTensorflowModel

public static <T extends Output<T>> TensorFlowSavedModelExternalModel<T> createTensorflowModel(OutputFactory<T> factory,
                                                                                               Map<String,Integer> featureMapping,
                                                                                               Map<T,Integer> outputMapping,
                                                                                               String outputName,
                                                                                               FeatureConverter featureConverter,
                                                                                               OutputConverter<T> outputConverter,
                                                                                               String bundleDirectory)

Creates a TensorflowSavedModelExternalModel by loading in a SavedModelBundle.

Throws IllegalArgumentException if the model bundle could not be loaded.

Type Parameters:

T - The type of the output.

Parameters:

factory - The output factory.

featureMapping - The feature mapping between Tribuo's names and the TF integer ids.

outputMapping - The output mapping between Tribuo's names and the TF integer ids.

outputName - The name of the output tensor.

featureConverter - The feature transformation function.

outputConverter - The output transformation function.

bundleDirectory - The path to load the saved model bundle from.

Returns:

The TF model wrapped in a Tribuo ExternalModel.