Many resources are needed to download a project. Please understand that we have to compensate our server costs. Thank you in advance. Project price only 1 $
You can buy this project and download/modify it how often you want.
/*
* ******************************************************************************
* *
* *
* * This program and the accompanying materials are made available under the
* * terms of the Apache License, Version 2.0 which is available at
* * https://www.apache.org/licenses/LICENSE-2.0.
* *
* * See the NOTICE file distributed with this work for additional
* * information regarding copyright ownership.
* * 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.
* *
* * SPDX-License-Identifier: Apache-2.0
* *****************************************************************************
*/
package org.deeplearning4j.nn.conf.ocnn;
import lombok.*;
import org.deeplearning4j.nn.api.Layer;
import org.deeplearning4j.nn.api.ParamInitializer;
import org.deeplearning4j.nn.conf.NeuralNetConfiguration;
import org.deeplearning4j.nn.conf.layers.BaseOutputLayer;
import org.deeplearning4j.nn.conf.layers.LayerValidation;
import org.deeplearning4j.nn.layers.ocnn.OCNNParamInitializer;
import org.deeplearning4j.optimize.api.TrainingListener;
import org.nd4j.linalg.activations.IActivation;
import org.nd4j.linalg.activations.impl.ActivationIdentity;
import org.nd4j.linalg.api.buffer.DataType;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.learning.regularization.Regularization;
import org.nd4j.linalg.lossfunctions.ILossFunction;
import org.nd4j.shade.jackson.annotation.JsonCreator;
import org.nd4j.shade.jackson.annotation.JsonIgnoreProperties;
import org.nd4j.shade.jackson.annotation.JsonProperty;
import java.util.Collection;
import java.util.List;
import java.util.Map;
@Data
@NoArgsConstructor
@ToString(callSuper = true)
@EqualsAndHashCode(callSuper = true)
@JsonIgnoreProperties("lossFn")
public class OCNNOutputLayer extends BaseOutputLayer {
//embedded hidden layer size
//aka "K"
private int hiddenSize;
private double nu = 0.04;
private int windowSize = 10000;
private double initialRValue = 0.1;
private boolean configureR = true;
/**
* Psuedo code from keras: start_time = time.time() for epoch in range(100): # Train with each example
* sess.run(updates, feed_dict={X: train_X,r:rvalue}) rvalue = nnScore(train_X, w_1, w_2, g) with sess.as_default():
* rvalue = rvalue.eval() rvalue = np.percentile(rvalue,q=100*nu) print("Epoch = %d, r = %f" % (epoch + 1,rvalue))
*/
private int lastEpochSinceRUpdated = 0;
public OCNNOutputLayer(Builder builder) {
super(builder);
this.hiddenSize = builder.hiddenLayerSize;
this.nu = builder.nu;
this.activationFn = builder.activation;
this.windowSize = builder.windowSize;
this.initialRValue = builder.initialRValue;
this.configureR = builder.configureR;
}
@JsonCreator
@SuppressWarnings("unused")
public OCNNOutputLayer(@JsonProperty("hiddenSize") int hiddenSize, @JsonProperty("nu") double nu,
@JsonProperty("activation") IActivation activation, @JsonProperty("windowSize") int windowSize,
@JsonProperty("initialRValue") double initialRValue,
@JsonProperty("configureR") boolean configureR) {
this.hiddenSize = hiddenSize;
this.nu = nu;
this.activationFn = activation;
this.windowSize = windowSize;
this.initialRValue = initialRValue;
this.configureR = configureR;
}
@Override
public ILossFunction getLossFn() {
return lossFn;
}
@Override
public Layer instantiate(NeuralNetConfiguration conf, Collection trainingListeners,
int layerIndex, INDArray layerParamsView, boolean initializeParams, DataType networkDataType) {
LayerValidation.assertNInNOutSet("OCNNOutputLayer", getLayerName(), layerIndex, getNIn(), getNOut());
org.deeplearning4j.nn.layers.ocnn.OCNNOutputLayer ret =
new org.deeplearning4j.nn.layers.ocnn.OCNNOutputLayer(conf, networkDataType);
ret.setListeners(trainingListeners);
ret.setIndex(layerIndex);
ret.setParamsViewArray(layerParamsView);
Map paramTable = initializer().init(conf, layerParamsView, initializeParams);
ret.setParamTable(paramTable);
ret.setConf(conf);
ret.setActivation(activationFn);
if (lastEpochSinceRUpdated == 0 && configureR) {
paramTable.get(OCNNParamInitializer.R_KEY).putScalar(0, initialRValue);
}
return ret;
}
@Override
public long getNOut() {
//we don't change number of outputs here
return 1L;
}
@Override
public ParamInitializer initializer() {
return OCNNParamInitializer.getInstance();
}
@Override
public List getRegularizationByParam(String paramName) {
//Not applicable
return null;
}
@Getter
@Setter
@NoArgsConstructor
public static class Builder extends BaseOutputLayer.Builder {
/**
* The hidden layer size for the one class neural network. Note this would be nOut on a dense layer. NOut in
* this neural net is always set to 1 though.
*
*/
protected int hiddenLayerSize;
/**
* For nu definition see the paper
*
*/
protected double nu = 0.04;
/**
* The number of examples to use for computing the quantile for the r value update. This value should generally
* be the same as the number of examples in the dataset
*
*/
protected int windowSize = 10000;
/**
* The activation function to use with ocnn
*
*/
protected IActivation activation = new ActivationIdentity();
/**
* The initial r value to use for ocnn for definition, see the paper, note this is only active when {@link
* #configureR} is specified as true
*/
protected double initialRValue = 0.1;
/**
* Whether to use the specified {@link #initialRValue} or use the weight initialization with the neural network
* for the r value
*/
protected boolean configureR = true;
/**
* Whether to use the specified {@link #initialRValue} or use the weight initialization with the neural network
* for the r value
*
* @param configureR true if we should use the initial {@link #initialRValue}
*/
public Builder configureR(boolean configureR) {
this.setConfigureR(configureR);
return this;
}
/**
* The initial r value to use for ocnn for definition, see the paper, note this is only active when {@link
* #configureR} is specified as true
*
* @param initialRValue the int
*/
public Builder initialRValue(double initialRValue) {
this.setInitialRValue(initialRValue);
return this;
}
/**
* The number of examples to use for computing the quantile for the r value update. This value should generally
* be the same as the number of examples in the dataset
*
* @param windowSize the number of examples to use for computing the quantile of the dataset for the r value
* update
*/
public Builder windowSize(int windowSize) {
this.setWindowSize(windowSize);
return this;
}
/**
* For nu definition see the paper
*
* @param nu the nu for ocnn
*/
public Builder nu(double nu) {
this.setNu(nu);
return this;
}
/**
* The activation function to use with ocnn
*
* @param activation the activation function to sue
*/
public Builder activation(IActivation activation) {
this.setActivation(activation);
return this;
}
/**
* The hidden layer size for the one class neural network. Note this would be nOut on a dense layer. NOut in
* this neural net is always set to 1 though.
*
* @param hiddenLayerSize the hidden layer size to use with ocnn
*/
public Builder hiddenLayerSize(int hiddenLayerSize) {
this.setHiddenLayerSize(hiddenLayerSize);
return this;
}
@Override
public Builder nOut(int nOut) {
throw new UnsupportedOperationException(
"Unable to specify number of outputs with ocnn. Outputs are fixed to 1.");
}
@Override
public void setNOut(long nOut){
throw new UnsupportedOperationException(
"Unable to specify number of outputs with ocnn. Outputs are fixed to 1.");
}
@Override
@SuppressWarnings("unchecked")
public OCNNOutputLayer build() {
return new OCNNOutputLayer(this);
}
}
}
