/**
* Licensed to the Apache Software Foundation (ASF) under one or more
* contributor license agreements. See the NOTICE file distributed with
* this work for additional information regarding copyright ownership.
* The ASF licenses this file to You under the Apache License, Version 2.0
* (the "License"); you may not use this file except in compliance with
* the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.mahout.classifier.rbm.model;
import java.io.IOException;
import java.util.Random;
import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.FSDataInputStream;
import org.apache.hadoop.fs.FSDataOutputStream;
import org.apache.hadoop.fs.FileSystem;
import org.apache.hadoop.fs.Path;
import org.apache.mahout.classifier.rbm.layer.Layer;
import org.apache.mahout.classifier.rbm.layer.SoftmaxLayer;
import org.apache.mahout.common.ClassUtils;
import org.apache.mahout.common.distance.DistanceMeasure;
import org.apache.mahout.common.distance.EuclideanDistanceMeasure;
import org.apache.mahout.math.DenseMatrix;
import org.apache.mahout.math.Matrix;
import org.apache.mahout.math.MatrixWritable;
import org.apache.mahout.math.Vector;
import com.google.common.io.Closeables;
/**
* The Class LabeledSimpleRBM which is a simple rbm, but also has a softmax layer in addition to
* the visible layer. This softmax layer can be seen as labels.
*/
public class LabeledSimpleRBM extends SimpleRBM {
/** The softmax layer. */
SoftmaxLayer softmaxLayer;
/** The weight label matrix. */
protected Matrix weightLabelMatrix;
/**
* Instantiates a new labeled simple rbm.
*
* @param visibleLayer the visible layer
* @param hiddenLayer the hidden layer
* @param labelLayer the label layer
*/
public LabeledSimpleRBM(Layer visibleLayer, Layer hiddenLayer, SoftmaxLayer labelLayer) {
super(visibleLayer, hiddenLayer);
this.softmaxLayer = labelLayer;
// initialize the random number generator
Random rand = new Random();
this.weightLabelMatrix = new DenseMatrix(softmaxLayer.getNeuronCount(), hiddenLayer.getNeuronCount());
//small random values chosen from a zero-mean Gaussian with
//a standard deviation of about 0.01
for (int i = 0; i < weightLabelMatrix.columnSize(); i++) {
for (int j = 0; j < weightLabelMatrix.rowSize(); j++) {
weightLabelMatrix.set(j, i, rand.nextGaussian()/100);
}
}
}
/**
* Gets the softmax layer.
*
* @return the softmax layer
*/
public SoftmaxLayer getSoftmaxLayer() {
return softmaxLayer;
}
/* (non-Javadoc)
* @see org.apache.mahout.classifier.rbm.model.SimpleRBM#exciteHiddenLayer(double, boolean)
*/
@Override
public void exciteHiddenLayer(double inputFactor, boolean addInput) {
Matrix activations = visibleLayer.getTransposedActivations();
Matrix softMaxActivations = softmaxLayer.getTransposedActivations();
Vector input = activations.times(weightMatrix).times(inputFactor).viewRow(0).plus(
softMaxActivations.times(weightLabelMatrix).times(inputFactor).viewRow(0) );
if(addInput)
hiddenLayer.addInputs(input);
else
hiddenLayer.setInputs(input);
hiddenLayer.exciteNeurons();
}
/* (non-Javadoc)
* @see org.apache.mahout.classifier.rbm.model.SimpleRBM#exciteVisibleLayer(double, boolean)
*/
@Override
public void exciteVisibleLayer(double inputFactor, boolean addInput) {
super.exciteVisibleLayer(inputFactor, addInput);
if(addInput)
softmaxLayer.addInputs(weightLabelMatrix.times(getHiddenLayer().getActivations()).times(inputFactor));
else
softmaxLayer.setInputs(weightLabelMatrix.times(getHiddenLayer().getActivations()).times(inputFactor));
softmaxLayer.exciteNeurons();
}
/**
* Sets the weight label matrix.
*
* @param weightLabelMatrix the new weight label matrix
*/
public void setWeightLabelMatrix(Matrix weightLabelMatrix) {
this.weightLabelMatrix = weightLabelMatrix;
}
/**
* Gets the weight label matrix.
*
* @return the weight label matrix
*/
public Matrix getWeightLabelMatrix() {
return weightLabelMatrix;
}
/* (non-Javadoc)
* @see org.apache.mahout.classifier.rbm.model.SimpleRBM#serialize(org.apache.hadoop.fs.Path, org.apache.hadoop.conf.Configuration)
*/
@Override
public void serialize(Path output, Configuration conf) throws IOException {
FileSystem fs = output.getFileSystem(conf);
FSDataOutputStream out = fs.create(new Path(output, conf.get("rbmnr")), true);
try {
out.writeChars(this.getClass().getName()+" ");
out.writeChars(visibleLayer.getClass().getName()+" ");
out.writeChars(hiddenLayer.getClass().getName()+" ");
MatrixWritable.writeMatrix(out, weightMatrix);
MatrixWritable.writeMatrix(out, weightLabelMatrix);
} finally {
Closeables.closeQuietly(out);
}
}
/**
* Materialize.
*
* @param output the output path
* @param conf the hadoop config
* @return the labeled rbm
* @throws IOException Signals that an I/O exception has occurred.
*/
public static LabeledSimpleRBM materialize(Path output,Configuration conf) throws IOException {
FileSystem fs = output.getFileSystem(conf);
Matrix weightMatrix;
Matrix weightLabelMatrix;
String visLayerType = "";
String hidLayerType = "";
FSDataInputStream in = fs.open(output);
try {
char chr;
while((chr=in.readChar())!=' ');
while((chr=in.readChar())!=' ')
visLayerType += chr;
while((chr=in.readChar())!=' ')
hidLayerType += chr;
weightMatrix = MatrixWritable.readMatrix(in);
weightLabelMatrix =MatrixWritable.readMatrix(in);
} finally {
Closeables.closeQuietly(in);
}
Layer vl = ClassUtils.instantiateAs(visLayerType, Layer.class,new Class[]{int.class},new Object[]{weightMatrix.rowSize()});
Layer hl = ClassUtils.instantiateAs(hidLayerType, Layer.class,new Class[]{int.class},new Object[]{weightMatrix.columnSize()});
LabeledSimpleRBM rbm = new LabeledSimpleRBM(vl, hl, new SoftmaxLayer(weightLabelMatrix.rowSize()));
rbm.setWeightMatrix(weightMatrix);
rbm.setWeightLabelMatrix(weightLabelMatrix);
return rbm;
}
/**
* Gets the label.
*
* @return the label
*/
public Integer getLabel() {
return softmaxLayer.getActiveNeuron();
}
/* (non-Javadoc)
* @see org.apache.mahout.classifier.rbm.model.RBMModel#updateVisibleLayer()
*/
@Override
public void updateVisibleLayer() {
visibleLayer.updateNeurons();
softmaxLayer.updateNeurons();
}
/* (non-Javadoc)
* @see org.apache.mahout.classifier.rbm.model.SimpleRBM#clone()
*/
@Override
public LabeledSimpleRBM clone() {
LabeledSimpleRBM rbm = new LabeledSimpleRBM(visibleLayer.clone(), hiddenLayer.clone(), softmaxLayer.clone());
rbm.weightMatrix = weightMatrix.clone();
rbm.weightLabelMatrix = weightLabelMatrix.clone();
return rbm;
}
@Override
public double getReconstructionError() {
Vector input = this.visibleLayer.getActivations().clone();
Vector label = this.softmaxLayer.getActivations().clone();
exciteHiddenLayer(1,false);
updateHiddenLayer();
exciteVisibleLayer(1,false);
DistanceMeasure dm = new EuclideanDistanceMeasure();
return dm.distance(input, visibleLayer.getExcitations())+dm.distance(label, softmaxLayer.getExcitations());
}
}