org.apache.mahout.classifier.sgd
Class GradientMachine

java.lang.Object
  extended by org.apache.mahout.classifier.AbstractVectorClassifier
      extended by org.apache.mahout.classifier.sgd.GradientMachine
All Implemented Interfaces:
Closeable, org.apache.hadoop.io.Writable, OnlineLearner

public class GradientMachine
extends AbstractVectorClassifier
implements OnlineLearner, org.apache.hadoop.io.Writable

Online gradient machine learner that tries to minimize the label ranking hinge loss. Implements a gradient machine with one sigmpod hidden layer. It tries to minimize the ranking loss of some given set of labels, so this can be used for multi-class, multi-label or auto-encoding of sparse data (e.g. text).


Field Summary
static int WRITABLE_VERSION
           
 
Fields inherited from class org.apache.mahout.classifier.AbstractVectorClassifier
MIN_LOG_LIKELIHOOD
 
Constructor Summary
GradientMachine(int numFeatures, int numHidden, int numOutput)
           
 
Method Summary
 Vector classify(Vector instance)
          Compute and return a vector containing n-1 scores, where n is equal to numCategories(), given an input vector instance.
 Vector classifyNoLink(Vector instance)
          Compute and return a vector of scores before applying the inverse link function.
 double classifyScalar(Vector instance)
          Classifies a vector in the special case of a binary classifier where AbstractVectorClassifier.classify(Vector) would return a vector with only one element.
 void close()
          Prepares the classifier for classification and deallocates any temporary data structures.
 GradientMachine copy()
           
 void copyFrom(GradientMachine other)
           
 DenseVector hiddenToOutput(Vector hiddenActivation)
          Feeds forward from hidden to output
 void initWeights(Random gen)
          Initialize weights.
 DenseVector inputToHidden(Vector input)
          Feeds forward from input to hidden unit..
 GradientMachine learningRate(double learningRate)
          Chainable configuration option.
 int numCategories()
          Returns the number of categories that a target variable can be assigned to.
 int numFeatures()
           
 int numHidden()
           
 void readFields(DataInput in)
           
 GradientMachine regularization(double regularization)
          Chainable configuration option.
 GradientMachine sparsity(double sparsity)
          Chainable configuration option.
 GradientMachine sparsityLearningRate(double sparsityLearningRate)
          Chainable configuration option.
 void train(int actual, Vector instance)
          Updates the model using a particular target variable value and a feature vector.
 void train(long trackingKey, int actual, Vector instance)
          Updates the model using a particular target variable value and a feature vector.
 void train(long trackingKey, String groupKey, int actual, Vector instance)
          Updates the model using a particular target variable value and a feature vector.
 void updateRanking(Vector hiddenActivation, Collection<Integer> goodLabels, int numTrials, Random gen)
          Updates using ranking loss.
 void write(DataOutput out)
           
 
Methods inherited from class org.apache.mahout.classifier.AbstractVectorClassifier
classify, classifyFull, classifyFull, classifyFull, classifyScalar, logLikelihood
 
Methods inherited from class java.lang.Object
clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait
 

Field Detail

WRITABLE_VERSION

public static final int WRITABLE_VERSION
See Also:
Constant Field Values
Constructor Detail

GradientMachine

public GradientMachine(int numFeatures,
                       int numHidden,
                       int numOutput)
Method Detail

initWeights

public void initWeights(Random gen)
Initialize weights.

Parameters:
gen - random number generator.

learningRate

public GradientMachine learningRate(double learningRate)
Chainable configuration option.

Parameters:
learningRate - New value of initial learning rate.
Returns:
This, so other configurations can be chained.

regularization

public GradientMachine regularization(double regularization)
Chainable configuration option.

Parameters:
regularization - A positive value that controls the weight vector size.
Returns:
This, so other configurations can be chained.

sparsity

public GradientMachine sparsity(double sparsity)
Chainable configuration option.

Parameters:
sparsity - A value between zero and one that controls the fraction of hidden units that are activated on average.
Returns:
This, so other configurations can be chained.

sparsityLearningRate

public GradientMachine sparsityLearningRate(double sparsityLearningRate)
Chainable configuration option.

Parameters:
sparsityLearningRate - New value of initial learning rate for sparsity.
Returns:
This, so other configurations can be chained.

copyFrom

public void copyFrom(GradientMachine other)

numCategories

public int numCategories()
Description copied from class: AbstractVectorClassifier
Returns the number of categories that a target variable can be assigned to. A vector classifier will encode it's output as an integer from 0 to numCategories()-1 (inclusive).

Specified by:
numCategories in class AbstractVectorClassifier
Returns:
The number of categories.

numFeatures

public int numFeatures()

numHidden

public int numHidden()

inputToHidden

public DenseVector inputToHidden(Vector input)
Feeds forward from input to hidden unit..

Returns:
Hidden unit activations.

hiddenToOutput

public DenseVector hiddenToOutput(Vector hiddenActivation)
Feeds forward from hidden to output

Returns:
Output unit activations.

updateRanking

public void updateRanking(Vector hiddenActivation,
                          Collection<Integer> goodLabels,
                          int numTrials,
                          Random gen)
Updates using ranking loss.

Parameters:
hiddenActivation - the hidden unit's activation
goodLabels - the labels you want ranked above others.
numTrials - how many times you want to search for the highest scoring bad label.
gen - Random number generator.

classify

public Vector classify(Vector instance)
Description copied from class: AbstractVectorClassifier
Compute and return a vector containing n-1 scores, where n is equal to numCategories(), given an input vector instance. Higher scores indicate that the input vector is more likely to belong to that category. The categories are denoted by the integers 0 through n-1 (inclusive), and the scores in the returned vector correspond to categories 1 through n-1 (leaving out category 0). It is assumed that the score for category 0 is one minus the sum of the scores in the returned vector.

Specified by:
classify in class AbstractVectorClassifier
Parameters:
instance - A feature vector to be classified.
Returns:
A vector of probabilities in 1 of n-1 encoding.

classifyNoLink

public Vector classifyNoLink(Vector instance)
Description copied from class: AbstractVectorClassifier
Compute and return a vector of scores before applying the inverse link function. For logistic regression and other generalized linear models, this is just the linear part of the classification.

The implementation of this method provided by AbstractVectorClassifier throws an UnsupportedOperationException. Your subclass must explicitly override this method to support this operation.

Overrides:
classifyNoLink in class AbstractVectorClassifier
Parameters:
instance - A feature vector to be classified.
Returns:
A vector of scores. If transformed by the link function, these will become probabilities.

classifyScalar

public double classifyScalar(Vector instance)
Description copied from class: AbstractVectorClassifier
Classifies a vector in the special case of a binary classifier where AbstractVectorClassifier.classify(Vector) would return a vector with only one element. As such, using this method can avoid the allocation of a vector.

Specified by:
classifyScalar in class AbstractVectorClassifier
Parameters:
instance - The feature vector to be classified.
Returns:
The score for category 1.
See Also:
AbstractVectorClassifier.classify(Vector)

copy

public GradientMachine copy()

write

public void write(DataOutput out)
           throws IOException
Specified by:
write in interface org.apache.hadoop.io.Writable
Throws:
IOException

readFields

public void readFields(DataInput in)
                throws IOException
Specified by:
readFields in interface org.apache.hadoop.io.Writable
Throws:
IOException

close

public void close()
Description copied from interface: OnlineLearner
Prepares the classifier for classification and deallocates any temporary data structures. An online classifier should be able to accept more training after being closed, but closing the classifier may make classification more efficient.

Specified by:
close in interface Closeable
Specified by:
close in interface OnlineLearner

train

public void train(long trackingKey,
                  String groupKey,
                  int actual,
                  Vector instance)
Description copied from interface: OnlineLearner
Updates the model using a particular target variable value and a feature vector.

There may an assumption that if multiple passes through the training data are necessary that the tracking key for a record will be the same for each pass and that there will be a relatively large number of distinct tracking keys and that the low-order bits of the tracking keys will not correlate with any of the input variables. This tracking key is used to assign training examples to different test/training splits.

Examples of useful tracking keys include id-numbers for the training records derived from a database id for the base table from the which the record is derived, or the offset of the original data record in a data file.

Specified by:
train in interface OnlineLearner
Parameters:
trackingKey - The tracking key for this training example.
groupKey - An optional value that allows examples to be grouped in the computation of the update to the model.
actual - The value of the target variable. This value should be in the half-open interval [0..n) where n is the number of target categories.
instance - The feature vector for this example.

train

public void train(long trackingKey,
                  int actual,
                  Vector instance)
Description copied from interface: OnlineLearner
Updates the model using a particular target variable value and a feature vector.

There may an assumption that if multiple passes through the training data are necessary that the tracking key for a record will be the same for each pass and that there will be a relatively large number of distinct tracking keys and that the low-order bits of the tracking keys will not correlate with any of the input variables. This tracking key is used to assign training examples to different test/training splits.

Examples of useful tracking keys include id-numbers for the training records derived from a database id for the base table from the which the record is derived, or the offset of the original data record in a data file.

Specified by:
train in interface OnlineLearner
Parameters:
trackingKey - The tracking key for this training example.
actual - The value of the target variable. This value should be in the half-open interval [0..n) where n is the number of target categories.
instance - The feature vector for this example.

train

public void train(int actual,
                  Vector instance)
Description copied from interface: OnlineLearner
Updates the model using a particular target variable value and a feature vector.

There may an assumption that if multiple passes through the training data are necessary, then the training examples will be presented in the same order. This is because the order of training examples may be used to assign records to different data splits for evaluation by cross-validation. Without the order invariance, records might be assigned to training and test splits and error estimates could be seriously affected.

If re-ordering is necessary, then using the alternative API which allows a tracking key to be added to the training example can be used.

Specified by:
train in interface OnlineLearner
Parameters:
actual - The value of the target variable. This value should be in the half-open interval [0..n) where n is the number of target categories.
instance - The feature vector for this example.


Copyright © 2008–2014 The Apache Software Foundation. All rights reserved.