org.nd4j.linalg.learning.GradientUpdater Maven / Gradle / Ivy
package org.nd4j.linalg.learning;
import org.nd4j.linalg.api.ndarray.INDArray;
import java.io.Serializable;
/**
* Gradient modifications:
* Calculates an update and tracks related
* information for gradient changes over time
* for handling updates.
*
* @author Adam Gibson
*/
public interface GradientUpdater extends Serializable {
/**
* For a give input size (length) array, how big is the internal state?
* Typically 0, 1 or 2x the input size, depending on the type of updater
*
* @param inputSize Length of the input array
* @return Number of elements in the internal state
*/
int stateSizeForInputSize(int inputSize);
/**
* For the internal updater state (if any): set this to use the provided array.
* Used during initialization, and when restoring the updater state (after serialization, for example)
* @param viewArray Array (that is a view of a larger array) to use for the state.
* @param gradientShape
* @param gradientOrder
* @param initialize If true: the updater must initialize the view array. If false: no change to view array contents
*/
void setStateViewArray(INDArray viewArray, int[] gradientShape, char gradientOrder, boolean initialize);
/**
* update(learningRate,momentum)
*
* @param args
*/
void update(Object... args);
/**
* Modify the gradient
* to be an update
*
* @param gradient the gradient to modify
* @param iteration
* @return the modified gradient
*/
INDArray getGradient(INDArray gradient, int iteration);
/**
* Get a GradientUpdaterAggregator. The aggregator is used (typically in distributed learning scenarios) to combine
* separate GradientUpdater instances for different networks (usually by averaging).
*
* @param addThis If true: return a GradientUpdaterAggregator with the GradientUpdater already added.
* If false: return an empty (uninitialized) GradientUpdaterAggregator
* @deprecated Use Updater view array functionality instead
*/
@Deprecated
GradientUpdaterAggregator getAggregator(boolean addThis);
}