org.tribuo.interop.tensorflow

Class TensorFlowModel<T extends Output<T>>

java.lang.Object

org.tribuo.Model<T>

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

Type Parameters:

T - The output type.

All Implemented Interfaces:

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

Direct Known Subclasses:

TensorFlowCheckpointModel, TensorFlowNativeModel

public abstract class TensorFlowModel<T extends Output<T>>
extends Model<T>
implements AutoCloseable

Base class for a TensorFlow model that operates on Examples.

The subclasses are package private and concern themselves with how the model is stored on disk.

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

See Also:

Serialized Form

Field Summary

Fields

Modifier and Type	Field and Description
protected int	batchSize
protected boolean	closed
protected FeatureConverter	featureConverter
protected org.tensorflow.Graph	modelGraph
protected OutputConverter<T>	outputConverter
protected String	outputName
protected org.tensorflow.Session	session

Fields inherited from class org.tribuo.Model

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

Constructor Summary

Constructors

Modifier	Constructor and Description
protected	TensorFlowModel(String name, ModelProvenance provenance, ImmutableFeatureMap featureIDMap, ImmutableOutputInfo<T> outputIDInfo, org.tensorflow.proto.framework.GraphDef trainedGraphDef, int batchSize, String outputName, FeatureConverter featureConverter, OutputConverter<T> outputConverter) Builds a TFModel.

Method Summary

Modifier and Type	Method and Description
void	close()
void	exportModel(String path) Exports this model as a SavedModelBundle, writing to the supplied directory.
int	getBatchSize() Gets the current testing batch size.
Optional<Excuse<T>>	getExcuse(Example<T> example) Deep learning models don't do excuses.
String	getOutputName() Gets the name of the output operation.
Map<String,List<com.oracle.labs.mlrg.olcut.util.Pair<String,Double>>>	getTopFeatures(int n) Deep learning models don't do feature rankings.
protected List<Prediction<T>>	innerPredict(Iterable<Example<T>> examples) Called by the base implementations of Model.predict(Iterable) and Model.predict(Dataset).
Prediction<T>	predict(Example<T> example) Uses the model to predict the output for a single example.
void	setBatchSize(int batchSize) Sets a new batch size.

Methods inherited from class org.tribuo.Model

copy, 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

Field Detail

batchSize

protected int batchSize

outputName

protected final String outputName

featureConverter

protected final FeatureConverter featureConverter

outputConverter

protected final OutputConverter<T extends Output<T>> outputConverter

modelGraph

protected transient org.tensorflow.Graph modelGraph

session

protected transient org.tensorflow.Session session

closed

protected transient boolean closed

Constructor Detail

TensorFlowModel

protected TensorFlowModel(String name,
                          ModelProvenance provenance,
                          ImmutableFeatureMap featureIDMap,
                          ImmutableOutputInfo<T> outputIDInfo,
                          org.tensorflow.proto.framework.GraphDef trainedGraphDef,
                          int batchSize,
                          String outputName,
                          FeatureConverter featureConverter,
                          OutputConverter<T> outputConverter)

Builds a TFModel. The session should be initialized in the subclass constructor.

Parameters:

name - The model name.

provenance - The model provenance.

featureIDMap - The feature domain.

outputIDInfo - The output domain.

trainedGraphDef - The graph definition.

batchSize - The test time batch size.

outputName - The name of the output operation.

featureConverter - The feature converter.

outputConverter - The output converter.

Method Detail

predict

public Prediction<T> predict(Example<T> example)

Description copied from class: Model

Uses the model to predict the output for a single example.

predict does not mutate the example.

Throws IllegalArgumentException if the example has no features or no feature overlap with the model.

Specified by:

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

Parameters:

example - the example to predict.

Returns:

the result of the prediction.

innerPredict

protected List<Prediction<T>> innerPredict(Iterable<Example<T>> examples)

Description copied from class: Model

Called by the base implementations of Model.predict(Iterable) and Model.predict(Dataset).

Overrides:

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

Parameters:

examples - The examples to predict.

Returns:

The results of the predictions, in the same order as the examples.

getBatchSize

public int getBatchSize()

Gets the current testing batch size.

Returns:

The batch size.

setBatchSize

public void setBatchSize(int batchSize)

Sets a new batch size.

Throws IllegalArgumentException if the batch size isn't positive.

Parameters:

batchSize - The batch size to use.

getTopFeatures

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

Deep learning models don't do feature rankings. Use an Explainer.

This method always returns the empty map.

Specified by:

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

Parameters:

n - the number of features to return.

Returns:

The empty map.

getExcuse

public Optional<Excuse<T>> getExcuse(Example<T> example)

Deep learning models don't do excuses. Use an Explainer.

This method always returns Optional.empty().

Specified by:

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

Parameters:

example - The input example.

Returns:

Optional.empty().

getOutputName

public String getOutputName()

Gets the name of the output operation.

Returns:

The output operation name.

exportModel

public void exportModel(String path)
                 throws IOException

Exports this model as a SavedModelBundle, writing to the supplied directory.

Parameters:

path - The directory to export to.

Throws:

IOException - If it failed to write to the directory.

close

public void close()

Specified by:

close in interface AutoCloseable