org.deeplearning4j.nn.conf.memory

Class NetworkMemoryReport

java.lang.Object

org.deeplearning4j.nn.conf.memory.MemoryReport

org.deeplearning4j.nn.conf.memory.NetworkMemoryReport

public class NetworkMemoryReport
extends MemoryReport

Network memory reports is a class that is used to store/represent the memory requirements of a MultiLayerNetwork or ComputationGraph, composed of multiple layers and/or vertices.

Author:

Alex Black

Field Summary

Fields inherited from class org.deeplearning4j.nn.conf.memory.MemoryReport

CACHE_MODE_ALL_ZEROS

Constructor Summary

Constructors

Constructor and Description
NetworkMemoryReport(Map<String,MemoryReport> layerAndVertexReports, Class<?> modelClass, String modelName, InputType... networkInputTypes)

Method Summary

Modifier and Type	Method and Description
long	getMemoryBytes(MemoryType memoryType, int minibatchSize, MemoryUseMode memoryUseMode, CacheMode cacheMode, org.nd4j.linalg.api.buffer.DataBuffer.Type dataType) Get the memory estimate (in bytes) for the specified type of memory
String	getName() Name of the object that the memory report was generated for
Class<?>	getReportClass()
long	getTotalMemoryBytes(int minibatchSize, MemoryUseMode memoryUseMode, CacheMode cacheMode, org.nd4j.linalg.api.buffer.DataBuffer.Type dataType) Get the total memory use in bytes for the given configuration
String	toString()

Methods inherited from class org.deeplearning4j.nn.conf.memory.MemoryReport

cacheModeMapFor, fromJson, fromYaml, getBytesPerElement, getMemoryBytes, getTotalMemoryBytes, toJson, toYaml

Methods inherited from class java.lang.Object

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

Constructor Detail

NetworkMemoryReport

public NetworkMemoryReport(@NonNull
                           Map<String,MemoryReport> layerAndVertexReports,
                           @NonNull
                           Class<?> modelClass,
                           String modelName,
                           @NonNull
                           InputType... networkInputTypes)

Method Detail

getReportClass

public Class<?> getReportClass()

Specified by:

getReportClass in class MemoryReport

Returns:

Class that the memory report was generated for

getName

public String getName()

Description copied from class: MemoryReport

Name of the object that the memory report was generated for

Specified by:

getName in class MemoryReport

Returns:

Name of the object

getTotalMemoryBytes

public long getTotalMemoryBytes(int minibatchSize,
                                @NonNull
                                MemoryUseMode memoryUseMode,
                                @NonNull
                                CacheMode cacheMode,
                                @NonNull
                                org.nd4j.linalg.api.buffer.DataBuffer.Type dataType)

Description copied from class: MemoryReport

Get the total memory use in bytes for the given configuration

Specified by:

getTotalMemoryBytes in class MemoryReport

Parameters:

minibatchSize - Mini batch size to estimate the memory for

memoryUseMode - The memory use mode (training or inference)

cacheMode - The CacheMode to use

dataType - Nd4j datatype

Returns:

The estimated total memory consumption in bytes

getMemoryBytes

public long getMemoryBytes(MemoryType memoryType,
                           int minibatchSize,
                           MemoryUseMode memoryUseMode,
                           CacheMode cacheMode,
                           org.nd4j.linalg.api.buffer.DataBuffer.Type dataType)

Description copied from class: MemoryReport

Get the memory estimate (in bytes) for the specified type of memory

Specified by:

getMemoryBytes in class MemoryReport

Parameters:

memoryType - Type of memory to get the estimate for invites

minibatchSize - Mini batch size to estimate the memory for

memoryUseMode - The memory use mode (training or inference)

cacheMode - The CacheMode to use

dataType - Nd4j datatype

Returns:

Estimated memory use for the given memory type

toString

public String toString()

Specified by:

toString in class MemoryReport