hex.genmodel.easy

Class EasyPredictModelWrapper

java.lang.Object

hex.genmodel.easy.EasyPredictModelWrapper

All Implemented Interfaces:

java.io.Serializable

public class EasyPredictModelWrapper
extends java.lang.Object
implements java.io.Serializable

An easy-to-use prediction wrapper for generated models. Instantiate as follows. The following two are equivalent. EasyPredictModelWrapper model = new EasyPredictModelWrapper(rawModel); EasyPredictModelWrapper model = new EasyPredictModelWrapper( new EasyPredictModelWrapper.Config() .setModel(rawModel) .setConvertUnknownCategoricalLevelsToNa(false)); Note that for any given model, you must use the exact one correct predict method below based on the model category. By default, unknown categorical levels result in a thrown PredictUnknownCategoricalLevelException. The API was designed with this default to make the simplest possible setup inform the user if there are concerns with the data quality. An alternate behavior is to automatically convert unknown categorical levels to N/A. To do this, use setConvertUnknownCategoricalLevelsToNa(true) instead. If you choose to convert unknown categorical levels to N/A, you can see how many times this is happening with the following methods: getTotalUnknownCategoricalLevelsSeen() getUnknownCategoricalLevelsSeenPerColumn()

See the top-of-tree master version of this file here on github.

See Also:

Serialized Form

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	EasyPredictModelWrapper.Config Configuration builder for instantiating a Wrapper.

Constructor Summary

Constructors

Constructor and Description
EasyPredictModelWrapper(EasyPredictModelWrapper.Config config) Create a wrapper for a generated model.
EasyPredictModelWrapper(GenModel model) Create a wrapper for a generated model.

Method Summary

Methods

Modifier and Type	Method and Description
java.lang.String	getHeader() Some autoencoder thing, I'm not sure what this does.
ModelCategory	getModelCategory() Get the category (type) of model.
java.lang.String[]	getResponseDomainValues() Get the array of levels for the response column.
long	getTotalUnknownCategoricalLevelsSeen() Get the total number unknown categorical levels seen.
java.util.concurrent.ConcurrentHashMap<java.lang.String,java.util.concurrent.atomic.AtomicLong>	getUnknownCategoricalLevelsSeenPerColumn() Get unknown categorical level counts.
AbstractPrediction	predict(RowData data)
AbstractPrediction	predict(RowData data, ModelCategory mc) Make a prediction on a new data point.
AutoEncoderModelPrediction	predictAutoEncoder(RowData data) Make a prediction on a new data point using an AutoEncoder model.
BinomialModelPrediction	predictBinomial(RowData data) Make a prediction on a new data point using a Binomial model.
ClusteringModelPrediction	predictClustering(RowData data) Make a prediction on a new data point using a Clustering model.
DimReductionModelPrediction	predictDimReduction(RowData data) Make a prediction on a new data point using a Dimension Reduction model (PCA, GLRM)
KLimeModelPrediction	predictKLime(RowData data) Make a prediction on a new data point using a k-LIME model.
MultinomialModelPrediction	predictMultinomial(RowData data) Make a prediction on a new data point using a Multinomial model.
RegressionModelPrediction	predictRegression(RowData data) Make a prediction on a new data point using a Regression model.
Word2VecPrediction	predictWord2Vec(RowData data) Lookup word embeddings for a given word (or set of words).
SortedClassProbability[]	sortByDescendingClassProbability(BinomialModelPrediction p) A helper function to return an array of binomial class probabilities for a prediction in sorted order.
SortedClassProbability[]	sortByDescendingClassProbability(MultinomialModelPrediction p) A helper function to return an array of multinomial class probabilities for a prediction in sorted order.

Methods inherited from class java.lang.Object

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

Constructor Detail

EasyPredictModelWrapper

public EasyPredictModelWrapper(EasyPredictModelWrapper.Config config)

Create a wrapper for a generated model.

Parameters:

config - The wrapper configuration

EasyPredictModelWrapper

public EasyPredictModelWrapper(GenModel model)

Create a wrapper for a generated model.

Parameters:

model - The generated model

Method Detail

getTotalUnknownCategoricalLevelsSeen

public long getTotalUnknownCategoricalLevelsSeen()

Get the total number unknown categorical levels seen. A single prediction may contribute more than one to the count. The count is only updated when setConvertUnknownCategoricalLevelsToNa is set to true.

Returns:

A long value.

getUnknownCategoricalLevelsSeenPerColumn

public java.util.concurrent.ConcurrentHashMap<java.lang.String,java.util.concurrent.atomic.AtomicLong> getUnknownCategoricalLevelsSeenPerColumn()

Get unknown categorical level counts. A single prediction may contribute to more than one count. Counts are only updated when setConvertUnknownCategoricalLevelsToNa is set to true.

Returns:

A hash map with a per-column count of unknown categorical levels seen when making predictions.

predict

public AbstractPrediction predict(RowData data,
                         ModelCategory mc)
                           throws PredictException

Make a prediction on a new data point. The type of prediction returned depends on the model type. The caller needs to know what type of prediction to expect. This call is convenient for generically automating model deployment. For specific applications (where the kind of model is known and doesn't change), it is recommended to call specific prediction calls like predictBinomial() directly.

Parameters:

data - A new data point.

Returns:

The prediction.

Throws:

PredictException

predict

public AbstractPrediction predict(RowData data)
                           throws PredictException

Throws:

PredictException

predictAutoEncoder

public AutoEncoderModelPrediction predictAutoEncoder(RowData data)
                                              throws PredictException

Make a prediction on a new data point using an AutoEncoder model.

Parameters:

data - A new data point.

Returns:

The prediction.

Throws:

PredictException

predictDimReduction

public DimReductionModelPrediction predictDimReduction(RowData data)
                                                throws PredictException

Make a prediction on a new data point using a Dimension Reduction model (PCA, GLRM)

Parameters:

data - A new data point.

Returns:

The prediction.

Throws:

PredictException

predictWord2Vec

public Word2VecPrediction predictWord2Vec(RowData data)
                                   throws PredictException

Lookup word embeddings for a given word (or set of words).

Parameters:

data - RawData structure, every key with a String value will be translated to an embedding

Returns:

The prediction

Throws:

PredictException - if model is not a WordEmbedding model

predictBinomial

public BinomialModelPrediction predictBinomial(RowData data)
                                        throws PredictException

Make a prediction on a new data point using a Binomial model.

Parameters:

data - A new data point.

Returns:

The prediction.

Throws:

PredictException

predictMultinomial

public MultinomialModelPrediction predictMultinomial(RowData data)
                                              throws PredictException

Make a prediction on a new data point using a Multinomial model.

Parameters:

data - A new data point.

Returns:

The prediction.

Throws:

PredictException

sortByDescendingClassProbability

public SortedClassProbability[] sortByDescendingClassProbability(BinomialModelPrediction p)

A helper function to return an array of binomial class probabilities for a prediction in sorted order. The returned array has the most probable class in position 0.

Parameters:

p - The prediction.

Returns:

An array with sorted class probabilities.

sortByDescendingClassProbability

public SortedClassProbability[] sortByDescendingClassProbability(MultinomialModelPrediction p)

A helper function to return an array of multinomial class probabilities for a prediction in sorted order. The returned array has the most probable class in position 0.

Parameters:

p - The prediction.

Returns:

An array with sorted class probabilities.

predictClustering

public ClusteringModelPrediction predictClustering(RowData data)
                                            throws PredictException

Make a prediction on a new data point using a Clustering model.

Parameters:

data - A new data point.

Returns:

The prediction.

Throws:

PredictException

predictRegression

public RegressionModelPrediction predictRegression(RowData data)
                                            throws PredictException

Make a prediction on a new data point using a Regression model.

Parameters:

data - A new data point.

Returns:

The prediction.

Throws:

PredictException

predictKLime

public KLimeModelPrediction predictKLime(RowData data)
                                  throws PredictException

Make a prediction on a new data point using a k-LIME model.

Parameters:

data - A new data point.

Returns:

The prediction.

Throws:

PredictException

getModelCategory

public ModelCategory getModelCategory()

Get the category (type) of model.

Returns:

The category.

getResponseDomainValues

public java.lang.String[] getResponseDomainValues()

Get the array of levels for the response column. "Domain" just means list of level names for a categorical (aka factor, enum) column. If the response column is numerical and not categorical, this will return null.

Returns:

The array.

getHeader

public java.lang.String getHeader()

Some autoencoder thing, I'm not sure what this does.

Returns:

CSV header for autoencoder.