Connects Layer's together into a direct acrylic graph (DAG) specified by a NetParameter More...

Inheritance diagram for MyCaffe.common.Net< T >:

Public Member Functions
	Net (CudaDnn< T > cuda, Log log, NetParameter p, CancelEvent evtCancel, IXDatabaseBase db, Phase phaseOverride=Phase.NONE, AutoResetEvent evtTrainingCompleted=null, Net< T > sharedNet=null, onGetWorkspace getws=null, onSetWorkspace setws=null)
	The Net constructor. More...

void	Dispose ()
	Releases all resources (GPU and Host) used by the Net. More...

void	Init (NetParameter p, Phase phaseOverride=Phase.NONE, AutoResetEvent evtTrainingCompleted=null)
	Initialize a network with a NetParameter. More...

Phase	SetPhase (Phase phase)
	Change the phase of the network. More...

void	RestorePhase ()
	Restore the network phase to its original state. More...

void	SetEnablePassthrough (bool bEnable)
	Enables/disables passthrough on each layer of the net. More...

NetParameter	FilterNet (NetParameter param)
	Removes layers that the user specified should be excluded given the current phase, level and stage. More...

bool	StateMeetsRule (NetState state, NetStateRule rule, string strLayer)
	Returns whether NetState state meets NetStateRule rule. More...

double	ForwardFromTo (int nStart=0, int nEnd=int.MaxValue)
	The FromTo variant of forward and backward operate on the (topological) ordering by which the net is specified. For general DAG netowrks, note that (1) computing from one layer to another might entail extra computation on unrelated branches, and (2) computation starting in the middle may be incorrect if all the layers of a fan-in are not included. More...

BlobCollection< T >	Forward ()
	Run forward with the input Blob's already fed separately. More...

BlobCollection< T >	Forward (out double dfLoss)
	Run forward with the input Blob's already fed separately. More...

BlobCollection< T >	Forward (BlobCollection< T > colBottom, out double dfLoss, bool bReshape=false)
	Run forward using a set of bottom blobs and return the result. More...

void	Backward (int nStart=int.MaxValue, int nEnd=0)
	The network backward should take no input and output, since it solely computes the gradient w.r.t. the parameters, and the data has already been provided during the forward pass. More...

void	ShareTrainedLayersWith (Net< T > srcNet, bool bEnableLog=false)
	For an already initialized net, implicitly compies (i.e., using no additional memory) the pre-trained layers from another Net. More...

void	CopyInternalBlobsTo (Net< T > dstNet)
	Copy the internal blobs from one net to another. More...

void	CopyTrainedLayersTo (Net< T > dstNet)
	Copies the trained layer of this Net to another Net. More...

void	CopyTrainedLayersTo (Net< T > dstNet, DictionaryEx< string, string > rgLayerNames, bool bTranspose)
	Copies the trained layers of this Net to another Net. More...

void	Reshape ()
	Reshape all layers from the bottom to the top. More...

void	CopyTrainedLayersFrom (NetParameter param)
	For an already initialized net, CopyTrainedLayersFrom copies the already trained layers from another net parameter instance. More...

NetParameter	ToProto (bool bIncludeBlobs)
	Writes the net to a proto. More...

void	Update ()
	Updates the network weights based on the diff values computed. More...

void	ClearParamDiffs ()
	Zero out the diffs of all netw parameters. This should be run before Backward. More...

void	ShareWeights ()
	Shares weight data of owner blobs with shared blobs. More...

bool	ForwardBackward (BlobCollection< T > colBottom, out double dfLocalLoss, TRAIN_STEP step=TRAIN_STEP.NONE)
	Runs a Forward pass followed by a Backward pass. More...

List< int >	top_ids (int layer_id)
	Returns the ids of the top blobs of layer i. More...

List< int >	bottom_ids (int layer_id)
	Returns the ids of the bottom blobs of layer i. More...

BlobCollection< T >	layer_blobs (string strLayerName)
	Returns the collection of Blobs internal to a Layer. More...

void	SetLearnedParameters (BlobCollection< T > col)
	Sets the learned parameters. More...

Blob< T >	param_by_name (string strName, bool bThrowExceptionOnError=true)
	Returns a parameter given its name. More...

bool	has_blob (string strBlobName)
	Returns whether or not the Net contains a given Blob. More...

Blob< T >	blob_by_name (string strName, bool bThrowExceptionOnError=true)
	Returns a blob given its name. More...

int	blob_index_by_name (string strName)
	Returns the index of a blob given its name. More...

bool	has_layer (string strLayer)
	Returns whether or not the Net has a given Layer by its name. More...

Layer< T >	layer_by_name (string strLayer, bool bThrowExceptionOnError=true)
	Returns a Layer given its name. More...

int	layer_index_by_name (string strLayer)
	Returns a Layer's index given its name. More...

void	set_debug_info (bool bVal)
	Sets the debug information flag. More...

NetParameter	InsertSplits (NetParameter param)
	Create a new NetParameter and insert splits into it based on a given NetParameter. More...

void	LoadWeights (byte[] rgWeights, IXPersist< T > persist, List< string > inputWtInfo=null, List< string > targetWtInfo=null, string strSkipBlobType=null)
	Loads new weights into the Net. More...

byte[]	SaveWeights (IXPersist< T > persist, bool bSaveDiff=false)
	Save the weights to a byte array. More...

Layer< T >	FindLayerOwningBlob (Blob< T > b)
	Finds the Layer that owns a given Blob. More...

int	FindLayerIndexOwningBlob (Blob< T > b)
	Finds the index of the Layer that owns a given Blob. More...

Blob< T >	FindBlob (string strName)
	Finds a Blob in the Net by name. More...

BlobCollection< T >	FindBottomBlobsOfLayer (string strLayer)
	Returns the collection of bottom blobs for a given layer. More...

BlobCollection< T >	FindTopBlobsOfLayer (string strLayer)
	Returns the collection of top blobs for a given layer. More...

Blob< T >	FindLossBottomBlob ()
	Find the bottom blob of the Loss layer if it exists, otherwise null is returned. More...

DebugInformation< T >	GetDebugInformation (bool bDetectNans)
	Returns the DebugInformation for the Net. More...

string	GetDataSource ()
	Returns the data source used by the network. More...

bool	ReInitializeParameters (WEIGHT_TARGET target, params string[] rgstrLayers)
	Re-initializes the blobs and each of the specified layers by re-running the filler (if any) specified by the layer. When the 'rgstr' parameter is null or otherwise empty, the blobs of all layers are re-initialized. More...

Layer< T >	FindLayer (LayerParameter.LayerType? type, string strName)
	Find the layer with the matching type, name and or both. More...

List< Layer< T > >	FindLayers (LayerParameter.LayerType type)
	Find the layers with the matching type. More...

Layer< T >	FindLayer (string strType, string strName)
	Find the layer with the matching type, name and or both. More...

Layer< T >	FindLastLayer (LayerParameter.LayerType type)
	Find the last layer with the matching type. More...

Protected Member Functions
virtual void	Dispose (bool bDisposing)
	Releases all resources (GPU and Host) used by the Net. More...

void	AppendTop (NetParameter param, int layer_id, int top_id, List< string > available_blobs, DictionaryEx< string, int > blob_name_to_idx)
	Append a new input or top blob to the net. More...

int	AppendBottom (NetParameter param, int layer_id, int bottom_id, List< string > available_blobs, DictionaryEx< string, int > blob_name_to_idx)
	Append a new bottom blob to the net. More...

void	AppendParam (NetParameter param, int layer_id, int param_id)
	Append a new parameter blob to the net. More...

void	InputDebugInfo (int input_id)
	Helper for displaying debug info in Forward about input blobs. More...

void	ForwardDebugInfo (int layer_id)
	Helper for displaying debug info in Forward. More...

void	BackwardDebugInfo (int layer_id)
	Helper for displaying debug info in Backward. More...

void	UpdateDebugInfo (int param_id)
	Helper for displaying debug info in Update. More...

Properties
bool	EnableBreakOnFirstNaN `[getset]`
	Enable/disable break the first NaN functionality where training stops immediately upon detecting a NaN in one of the Layer blobs. More...

bool	EnableDetailedNanDetection `[getset]`
	Enable/disable whether or not detailed nans are detected - this will make debugging slower and is only recommended when running on a TCC enabled driver (as opposed to an WDM driver one used with the monitor). More...

bool	EnableLayerDebugging `[getset]`
	Enable/disable layer debugging which causes each layer to check for NAN/INF on each forward/backward pass and throw an exception when found. More...

string	ActiveLabelCounts `[get]`
	Returns the active label counts observed during training. More...

string	LabelQueryHitPercents `[get]`
	Return the label query hit percentages for the active datasource. More...

string	LabelQueryEpochs `[get]`
	Return the label query epochs for the active datasource. More...

string	BoostQueryHitPercents `[get]`
	Return the boost query hit percentages for the active datasource. More...

NetParameter	net_param `[get]`
	Returns the net parameter. More...

string	name `[get]`
	Returns the network name. More...

List< string >	layer_names `[get]`
	Returns the layer names. More...

List< string >	blob_names `[get]`
	Returns the blob names. More...

BlobCollection< T >	blobs `[get]`
	Returns the blobs. More...

List< Layer< T > >	layers `[get]`
	Returns the layers. More...

Phase	phase `[get]`
	Returns the network phase: TRAIN or TEST More...

List< BlobCollection< T > >	bottom_vecs `[get]`
	Returns the bottom vecs for each layer – usually you won't need this unless you do per-layer checks such as gradients. More...

List< BlobCollection< T > >	top_vecs `[get]`
	Returns the top vecs for each layer – usually y ou won't need this unless you do per-layer checks such as gradients. More...

List< List< bool > >	bottom_need_backward `[get]`
	Returns the collection of lists that tell whether or not the bottom of each layer needs a backward pass or not. More...

List< double >	blob_loss_weights `[get]`
	Returns the collection of blob loss weights. More...

List< bool >	layer_need_backward `[get]`
	Returns a collection of items that tell whether each layer nees a backward pass or not. More...

BlobCollection< T >	parameters `[get]`
	Returns the parameters. More...

BlobCollection< T >	learnable_parameters `[get]`
	Returns the learnable parameters. More...

List< double?>	params_lr `[get]`
	Returns the learnable parameter learning rate multipliers. More...

List< double?>	params_weight_decay `[get]`
	Returns the learnable parameter decay multipliers. More...

DictionaryEx< string, int >	param_names_index `[get]`
	Returns the dictionary look for parameter names to their indexes. More...

List< int >	param_owners `[get]`
	Returns the list of parameter owner indexes. More...

List< string >	param_display_names `[get]`
	Returns the list of parameter display names. More...

int	num_inputs `[get]`
	Returns the number of inputs. More...

int	num_outputs `[get]`
	Returns the number of outputs. More...

BlobCollection< T >	input_blobs `[get]`
	Returns the collection of input Blobs. More...

BlobCollection< T >	output_blobs `[get]`
	Returns the collection of output Blobs. More...

List< int >	output_blob_indices `[get]`
	Returns a list of the output Blob indexes. More...

List< int >	input_blob_indices `[get]`
	Returns a list of the input Blob indexes. More...

CudaDnn< T >	Cuda `[get]`
	Returns the instance of CudaDnn used by this network. More...

Events
EventHandler< WorkspaceArgs >	OnGetWorkspace
	Specifies the OnGetWorkspace event that fires when the getWorkspace() function is called by a layer to get a shareable workspace to conserve GPU memory. More...

EventHandler< WorkspaceArgs >	OnSetWorkspace
	Specifies the OnSetWorkspace event that fires when the setWorkspace() function is called by a layer to get a shareable workspace to conserve GPU memory. More...

EventHandler< GetIterationArgs >	OnGetIteration
	Specifies the OnGetIteration event that fires when a layer needs to get the current iteration from the solver. More...

Detailed Description

Connects Layer's together into a direct acrylic graph (DAG) specified by a NetParameter

Template Parameters

T	Specifies the base type float or double. Using float is recommended to conserve GPU memory.

Definition at line 22 of file Net.cs.

Constructor & Destructor Documentation

◆ Net()

MyCaffe.common.Net< T >.Net	(	CudaDnn< T >	cuda,
		Log	log,
		NetParameter	p,
		CancelEvent	evtCancel,
		IXDatabaseBase	db,
		Phase	phaseOverride = `Phase.NONE`,
		AutoResetEvent	evtTrainingCompleted = `null`,
		Net< T >	sharedNet = `null`,
		onGetWorkspace	getws = `null`,
		onSetWorkspace	setws = `null`
	)

The Net constructor.

Parameters

cuda	Specifies the CudaDnn connection to Cuda.
log	Specifies the Log for output.
p	Specifies the NetParameter used to initialize the Net.
evtCancel	Specifies the CancelEvent used to cancel operations run by the Net.
db	Specifies the in-memory MyCaffeDatabase.
phaseOverride	Optionally, specifies an override of the Phase for which the Net is used.
evtTrainingCompleted	Optionally, specifies an auto reset event that is set after training has completed.
sharedNet	Specifies another Net that shares the GPU memory created by this Net.
getws	Optionally, specifies the handler for getting the workspace.
setws	Optionally, specifies the handler for setting the workspace.

Definition at line 161 of file Net.cs.

Member Function Documentation

◆ AppendBottom()

int MyCaffe.common.Net< T >.AppendBottom	(	NetParameter	param,
		int	layer_id,
		int	bottom_id,
		List< string >	available_blobs,
		DictionaryEx< string, int >	blob_name_to_idx
	)

protected

Append a new bottom blob to the net.

Parameters

param	Specifies the NetParameter used.
layer_id	Specifies the Layer index associated with the Blob.
bottom_id	Specifies the Blob index of the (bottom) Blob.
available_blobs	Specifies the list of available Blob names.
blob_name_to_idx	Specifies the Blob name to index dictionary lookup.

Definition at line 1272 of file Net.cs.

◆ AppendParam()

void MyCaffe.common.Net< T >.AppendParam	(	NetParameter	param,
		int	layer_id,
		int	param_id
	)

protected

Append a new parameter blob to the net.

Parameters

param	Specifies the NetParameter used.
layer_id	Specifies the Layer index associated with the Blob.
param_id	Specifies the Blob index of the (parameter) Blob.

Definition at line 1303 of file Net.cs.

◆ AppendTop()

void MyCaffe.common.Net< T >.AppendTop	(	NetParameter	param,
		int	layer_id,
		int	top_id,
		List< string >	available_blobs,
		DictionaryEx< string, int >	blob_name_to_idx
	)

protected

Append a new input or top blob to the net.

Parameters

param	Specifies the NetParameter used.
layer_id	Specifies the Layer index associated with the Blob.
top_id	Specifies the Blob index of the (top) Blob.
available_blobs	Specifies the list of available Blob names.
blob_name_to_idx	Specifies the Blob name to index dictionary lookup.

Definition at line 1154 of file Net.cs.

◆ Backward()

void MyCaffe.common.Net< T >.Backward	(	int	nStart = `int.MaxValue`,
		int	nEnd = `0`
	)

The network backward should take no input and output, since it solely computes the gradient w.r.t. the parameters, and the data has already been provided during the forward pass.

Parameters

nStart	Specifies the Layer index where the Backward operation is to start.
nEnd	Specifies the Layer index of the last Layer that the Backward operation is run.

Definition at line 1499 of file Net.cs.

◆ BackwardDebugInfo()

void MyCaffe.common.Net< T >.BackwardDebugInfo ( int layer_id )

protected

Helper for displaying debug info in Backward.

Parameters

layer_id Specifies the Layer index associated with the Blob.

Definition at line 1587 of file Net.cs.

◆ blob_by_name()

Blob< T > MyCaffe.common.Net< T >.blob_by_name	(	string	strName,
		bool	bThrowExceptionOnError = `true`
	)

Returns a blob given its name.

Parameters

strName	Specifies the Blob's name.
bThrowExceptionOnError	Optionally, specifies to throw an exception when the layer is not found. The default = true.

Returns: The Blob with the given name is returned, or null if not found.

Definition at line 2245 of file Net.cs.

◆ blob_index_by_name()

int MyCaffe.common.Net< T >.blob_index_by_name ( string strName )

Returns the index of a blob given its name.

Parameters

strName Specifies the name of the blob to look for.

Returns: The index of the blob within the 'blobs' is returned.

Definition at line 2268 of file Net.cs.

◆ bottom_ids()

List< int > MyCaffe.common.Net< T >.bottom_ids ( int layer_id )

Returns the ids of the bottom blobs of layer i.

Parameters

layer_id Specifies the index of the Layer.

Returns

Definition at line 2050 of file Net.cs.

◆ ClearParamDiffs()

void MyCaffe.common.Net< T >.ClearParamDiffs ( )

Zero out the diffs of all netw parameters. This should be run before Backward.

Definition at line 1907 of file Net.cs.

◆ CopyInternalBlobsTo()

void MyCaffe.common.Net< T >.CopyInternalBlobsTo ( Net< T > dstNet )

Copy the internal blobs from one net to another.

Parameters

dstNet Specifies the destination net.

Definition at line 1699 of file Net.cs.

◆ CopyTrainedLayersFrom()

void MyCaffe.common.Net< T >.CopyTrainedLayersFrom ( NetParameter param )

For an already initialized net, CopyTrainedLayersFrom copies the already trained layers from another net parameter instance.

Parameters

param Specifies the NetParameter to use.

Definition at line 1814 of file Net.cs.

◆ CopyTrainedLayersTo() [1/2]

void MyCaffe.common.Net< T >.CopyTrainedLayersTo ( Net< T > dstNet )

Copies the trained layer of this Net to another Net.

Parameters

dstNet Specifies the Net where the trained layers are to be copied.

Definition at line 1714 of file Net.cs.

◆ CopyTrainedLayersTo() [2/2]

void MyCaffe.common.Net< T >.CopyTrainedLayersTo	(	Net< T >	dstNet,
		DictionaryEx< string, string >	rgLayerNames,
		bool	bTranspose
	)

Copies the trained layers of this Net to another Net.

Parameters

dstNet	Specifies the destination Net where the trained layers are to be copied.
rgLayerNames	Specifies the layer name dictionary lookup, where only the weights of layer names within the Dictionary lookup are copied.
bTranspose	Specifies whether or not to copy the weights and transpose the copy.

Definition at line 1754 of file Net.cs.

◆ Dispose() [1/2]

void MyCaffe.common.Net< T >.Dispose ( )

Releases all resources (GPU and Host) used by the Net.

Definition at line 258 of file Net.cs.

◆ Dispose() [2/2]

virtual void MyCaffe.common.Net< T >.Dispose ( bool bDisposing )

protectedvirtual

Releases all resources (GPU and Host) used by the Net.

Parameters

bDisposing Set to true when called from Dispose().

Definition at line 184 of file Net.cs.

◆ FilterNet()

NetParameter MyCaffe.common.Net< T >.FilterNet ( NetParameter param )

Removes layers that the user specified should be excluded given the current phase, level and stage.

Parameters

param Specifies the NetParameter to filter.

Returns: The newly filtered NetParmeter is returned.

Definition at line 1020 of file Net.cs.

◆ FindBlob()

Blob< T > MyCaffe.common.Net< T >.FindBlob ( string strName )

Finds a Blob in the Net by name.

Parameters

strName Specifies the Blob name.

Returns: If found, the Blob is returned, otherwise null is returned.

Definition at line 2592 of file Net.cs.

◆ FindBottomBlobsOfLayer()

BlobCollection< T > MyCaffe.common.Net< T >.FindBottomBlobsOfLayer ( string strLayer )

Returns the collection of bottom blobs for a given layer.

Parameters

strLayer Specifies the layer name.

Returns: The collection of bottom blobs is returned.

Definition at line 2623 of file Net.cs.

◆ FindLastLayer()

Layer< T > MyCaffe.common.Net< T >.FindLastLayer ( LayerParameter.LayerType type )

Find the last layer with the matching type.

Parameters

type	Specifies the layer type.

Returns: The layer found (if any) is returned, otherwise null is returned.

Definition at line 2806 of file Net.cs.

◆ FindLayer() [1/2]

Layer< T > MyCaffe.common.Net< T >.FindLayer	(	LayerParameter.LayerType?	type,
		string	strName
	)

Find the layer with the matching type, name and or both.

Parameters

type	Specifies the layer type.
strName	Specifies the layer name.

Returns: The layer found (if any) is returned, otherwise null is returned.

Definition at line 2748 of file Net.cs.

◆ FindLayer() [2/2]

Layer< T > MyCaffe.common.Net< T >.FindLayer	(	string	strType,
		string	strName
	)

Find the layer with the matching type, name and or both.

Parameters

strType	Specifies the layer type.
strName	Specifies the layer name.

Returns: The layer found (if any) is returned, otherwise null is returned.

Definition at line 2795 of file Net.cs.

◆ FindLayerIndexOwningBlob()

int MyCaffe.common.Net< T >.FindLayerIndexOwningBlob ( Blob< T > b )

Finds the index of the Layer that owns a given Blob.

Parameters

b	Specifies the Blob to search for.

Returns: If found, the Layer index is returned, otherwise -1 is returned.

Definition at line 2576 of file Net.cs.

◆ FindLayerOwningBlob()

Layer< T > MyCaffe.common.Net< T >.FindLayerOwningBlob ( Blob< T > b )

Finds the Layer that owns a given Blob.

Parameters

b	Specifies the Blob to search for.

Returns: If found, the Layer is returned, otherwise null is returned.

Definition at line 2560 of file Net.cs.

◆ FindLayers()

List< Layer< T > > MyCaffe.common.Net< T >.FindLayers ( LayerParameter.LayerType type )

Find the layers with the matching type.

Parameters

type	Specifies the type of layer to find.

Returns: The list of layers is returned.

Definition at line 2776 of file Net.cs.

◆ FindLossBottomBlob()

Blob< T > MyCaffe.common.Net< T >.FindLossBottomBlob ( )

Find the bottom blob of the Loss layer if it exists, otherwise null is returned.

Returns: Return the blob of the loss layer or null if not found.

Definition at line 2664 of file Net.cs.

◆ FindTopBlobsOfLayer()

BlobCollection< T > MyCaffe.common.Net< T >.FindTopBlobsOfLayer ( string strLayer )

Returns the collection of top blobs for a given layer.

Parameters

strLayer Specifies the layer name.

Returns: The collection of top blobs is returned.

Definition at line 2644 of file Net.cs.

◆ Forward() [1/3]

BlobCollection< T > MyCaffe.common.Net< T >.Forward ( )

Run forward with the input Blob's already fed separately.

You can get the input blobs using input_blobs().

Returns: The collection of output Blobs is returned.

Definition at line 1445 of file Net.cs.

◆ Forward() [2/3]

BlobCollection< T > MyCaffe.common.Net< T >.Forward	(	BlobCollection< T >	colBottom,
		out double	dfLoss,
		bool	bReshape = `false`
	)

Run forward using a set of bottom blobs and return the result.

Parameters

colBottom
dfLoss
bReshape	Optionally, specifies to reshape the input to the size of the colBottom blobs supplied.

Returns

Definition at line 1472 of file Net.cs.

◆ Forward() [3/3]

BlobCollection< T > MyCaffe.common.Net< T >.Forward ( out double dfLoss )

Run forward with the input Blob's already fed separately.

You can get the input blobs using input_blobs().

Parameters

dfLoss Returns the loss of the operation.

Returns: The collection of output Blobs is returned.

Definition at line 1459 of file Net.cs.

◆ ForwardBackward()

bool MyCaffe.common.Net< T >.ForwardBackward	(	BlobCollection< T >	colBottom,
		out double	dfLocalLoss,
		TRAIN_STEP	step = `TRAIN_STEP.NONE`
	)

Runs a Forward pass followed by a Backward pass.

Parameters

colBottom	Optionally, specifies input data passed to the Forward pass.
dfLocalLoss	Returns the local loss of the Forward pass.
step	Optionally, specifies to step forward or backward.

Returns: If EnableBreakOnFirstNaN == true and a NaN is detected, this function returns false, otherwise true is returned.

Definition at line 1943 of file Net.cs.

◆ ForwardDebugInfo()

void MyCaffe.common.Net< T >.ForwardDebugInfo ( int layer_id )

protected

Helper for displaying debug info in Forward.

Parameters

layer_id Specifies the Layer index associated with the Blob.

Definition at line 1558 of file Net.cs.

◆ ForwardFromTo()

double MyCaffe.common.Net< T >.ForwardFromTo	(	int	nStart = `0`,
		int	nEnd = `int.MaxValue`
	)

The FromTo variant of forward and backward operate on the (topological) ordering by which the net is specified. For general DAG netowrks, note that (1) computing from one layer to another might entail extra computation on unrelated branches, and (2) computation starting in the middle may be incorrect if all the layers of a fan-in are not included.

Parameters

nStart	Optionally, specifies the index of the first Layer where the Forward operation is to start.
nEnd	Optionally, specifies the index of the last Layer to run the Forward operation on.

Returns: The loss is returned.

Definition at line 1402 of file Net.cs.

◆ GetDataSource()

string MyCaffe.common.Net< T >.GetDataSource ( )

Returns the data source used by the network.

Returns: The data source name is returned.

Definition at line 2711 of file Net.cs.

◆ GetDebugInformation()

DebugInformation< T > MyCaffe.common.Net< T >.GetDebugInformation ( bool bDetectNans )

Returns the DebugInformation for the Net.

Parameters

bDetectNans Specifies whether or not to detect Nan's in the data.

Returns: The DebugInformation for the Net is returned.

Definition at line 2692 of file Net.cs.

◆ has_blob()

bool MyCaffe.common.Net< T >.has_blob ( string strBlobName )

Returns whether or not the Net contains a given Blob.

Parameters

strBlobName Specifies the Blob name.

Returns: If the Net has the Blob, true is returned, otherwise false is returned.

Definition at line 2234 of file Net.cs.

◆ has_layer()

bool MyCaffe.common.Net< T >.has_layer ( string strLayer )

Returns whether or not the Net has a given Layer by its name.

Parameters

strLayer Specifies the Layer name.

Returns: If the Net contains the Layer, true is returned, otherwise false is returned.

Definition at line 2281 of file Net.cs.

◆ Init()

void MyCaffe.common.Net< T >.Init	(	NetParameter	p,
		Phase	phaseOverride = `Phase.NONE`,
		AutoResetEvent	evtTrainingCompleted = `null`
	)

Initialize a network with a NetParameter.

Parameters

p	Specifies the NetParameter.
phaseOverride	Optionally, specifies a Phase override for which the Net will run under.
evtTrainingCompleted	Optionally, specifies an auto reset event that is set upon the completion of training.

Definition at line 287 of file Net.cs.

◆ InputDebugInfo()

void MyCaffe.common.Net< T >.InputDebugInfo ( int input_id )

protected

Helper for displaying debug info in Forward about input blobs.

Parameters

input_id Specifies the index of the input Blob within input_blobs().

Definition at line 1544 of file Net.cs.

◆ InsertSplits()

NetParameter MyCaffe.common.Net< T >.InsertSplits ( NetParameter param )

Create a new NetParameter and insert splits into it based on a given NetParameter.

Splits are used when a given input (top) is used by more than one Layer. For example a DataLayer 'label' top may used by both an AccuracyLayer and SoftmaxLossLayer. In such a case a split is created that allows the 'label' top to be sent to both.

Parameters

param Specifies the original NetParameter.

Returns: A new NetParameter containing split layers is returned.

Definition at line 2347 of file Net.cs.

◆ layer_blobs()

BlobCollection< T > MyCaffe.common.Net< T >.layer_blobs ( string strLayerName )

Returns the collection of Blobs internal to a Layer.

Parameters

strLayerName Specifies the name of the Layer.

Returns: The Layer's internal Blobs are returned in a collection.

Definition at line 2094 of file Net.cs.

◆ layer_by_name()

Layer< T > MyCaffe.common.Net< T >.layer_by_name	(	string	strLayer,
		bool	bThrowExceptionOnError = `true`
	)

Returns a Layer given its name.

Parameters

strLayer	Specifies the Layer name.
bThrowExceptionOnError	Optionally, specifies to throw an exception when the layer is not found. The default = true.

Returns: The Layer with the given name is returned, or null if not found.

Definition at line 2292 of file Net.cs.

◆ layer_index_by_name()

int MyCaffe.common.Net< T >.layer_index_by_name ( string strLayer )

Returns a Layer's index given its name.

Parameters

strLayer Specifies the Layer name.

Returns: The index of the Layer with the given name is returned, or -1 if not found.

Definition at line 2315 of file Net.cs.

◆ LoadWeights()

void MyCaffe.common.Net< T >.LoadWeights	(	byte[]	rgWeights,
		IXPersist< T >	persist,
		List< string >	inputWtInfo = `null`,
		List< string >	targetWtInfo = `null`,
		string	strSkipBlobType = `null`
	)

Loads new weights into the Net.

Parameters

rgWeights	Specifies the weights themselves.
persist	Specifies an interface to the persistance object used to load the weights.
inputWtInfo	Optionally, specifies the input blobs to import. Note, when set, the targetWtInfo must also be specified. When null, this parameter is ignored.
targetWtInfo	Optionally, specifies the target blobs to import into. Note, when set, the inputWtInfo must also be specified. When null, this parameter is ignored.
strSkipBlobType	Optionally, specifies a blob type where weights are NOT loaded. See Blob.BLOB_TYPE for the types of Blobs.

Definition at line 2510 of file Net.cs.

◆ param_by_name()

Blob< T > MyCaffe.common.Net< T >.param_by_name	(	string	strName,
		bool	bThrowExceptionOnError = `true`
	)

Returns a parameter given its name.

Parameters

strName	Specifies the Parameter's name.
bThrowExceptionOnError	Optionally, specifies to throw an exception when the layer is not found. The default = true.

Returns: The Blob with the given name is returned, or null if not found.

Definition at line 2167 of file Net.cs.

◆ ReInitializeParameters()

bool MyCaffe.common.Net< T >.ReInitializeParameters	(	WEIGHT_TARGET	target,
		params string[]	rgstrLayers
	)

Re-initializes the blobs and each of the specified layers by re-running the filler (if any) specified by the layer.
When the 'rgstr' parameter is null or otherwise empty, the blobs of all layers are re-initialized.

Parameters

target	Specifies the weights to target (e.g. weights, bias or both).
rgstrLayers	Specifies the layers to reinitialize, when null or empty, all layers are re-initialized

Returns: If a layer is specified and found, true is returned, otherwise false is returned.

Definition at line 2729 of file Net.cs.

◆ Reshape()

void MyCaffe.common.Net< T >.Reshape ( )

Reshape all layers from the bottom to the top.

This is useful to propagate changes to layer sizes without running a forward pass, e.g. to compute output feature size.

Definition at line 1800 of file Net.cs.

◆ RestorePhase()

void MyCaffe.common.Net< T >.RestorePhase ( )

Restore the network phase to its original state.

Definition at line 831 of file Net.cs.

◆ SaveWeights()

byte[] MyCaffe.common.Net< T >.SaveWeights	(	IXPersist< T >	persist,
		bool	bSaveDiff = `false`
	)

Save the weights to a byte array.

Parameters

persist	Specifies an interface to the persistance object used to save the weights.
bSaveDiff	Specifies to save the diff values.

Returns: The byte array containing the weights is returned.

Definition at line 2541 of file Net.cs.

◆ set_debug_info()

void MyCaffe.common.Net< T >.set_debug_info ( bool bVal )

Sets the debug information flag.

When set, extra debug information is output during each Forward and Backward pass, which will slow down training.

Parameters

bVal	Specifies whether to enable/disable debug information.

Definition at line 2330 of file Net.cs.

◆ SetEnablePassthrough()

void MyCaffe.common.Net< T >.SetEnablePassthrough ( bool bEnable )

Enables/disables passthrough on each layer of the net.

If enabled by a given layer, the Bottom inputs are copied directly to the Top outputs during the forward pass and the forward pass returns. This is used by the BatchDataLayer.

Parameters

bEnable Specifies whether or not to enable passthrough.

Definition at line 996 of file Net.cs.

◆ SetLearnedParameters()

void MyCaffe.common.Net< T >.SetLearnedParameters ( BlobCollection< T > col )

Sets the learned parameters.

Parameters

col	Specifies a collection of Blobs containing the learned parameters.

Definition at line 2108 of file Net.cs.

◆ SetPhase()

Phase MyCaffe.common.Net< T >.SetPhase ( Phase phase )

Change the phase of the network.

Parameters

phase Specifies the new phase.

Returns: The previous phase is returned.

Definition at line 814 of file Net.cs.

◆ ShareTrainedLayersWith()

void MyCaffe.common.Net< T >.ShareTrainedLayersWith	(	Net< T >	srcNet,
		bool	bEnableLog = `false`
	)

For an already initialized net, implicitly compies (i.e., using no additional memory) the pre-trained layers from another Net.

Parameters

srcNet	Specifies the source Net whos blobs are shared with the calling Net.
bEnableLog	Optionally, specifies to enable the output log (default = false).

Definition at line 1653 of file Net.cs.

◆ ShareWeights()

void MyCaffe.common.Net< T >.ShareWeights ( )

Shares weight data of owner blobs with shared blobs.

Note: this is called by Net::Init, and thus should normally not be called manually.

Definition at line 1923 of file Net.cs.

◆ StateMeetsRule()

bool MyCaffe.common.Net< T >.StateMeetsRule	(	NetState	state,
		NetStateRule	rule,
		string	strLayer
	)

Returns whether NetState state meets NetStateRule rule.

Parameters

state	Specifies the NetState to test.
rule	Specifies the NetStateRul to test against the NetState.
strLayer	Specifies the name of the Layer for which the test is taking place.

Returns: If the NetState of the Layer meets the NetStateRule, true is returned, otherwise false is returned.

Definition at line 1062 of file Net.cs.

◆ top_ids()

List< int > MyCaffe.common.Net< T >.top_ids ( int layer_id )

Returns the ids of the top blobs of layer i.

Parameters

layer_id Specifies the index of the Layer.

Returns

Definition at line 2038 of file Net.cs.

◆ ToProto()

NetParameter MyCaffe.common.Net< T >.ToProto ( bool bIncludeBlobs )

Writes the net to a proto.

Returns: A new NetParameter is returned.

Definition at line 1865 of file Net.cs.

◆ Update()

void MyCaffe.common.Net< T >.Update ( )

Updates the network weights based on the diff values computed.

Definition at line 1896 of file Net.cs.

◆ UpdateDebugInfo()

void MyCaffe.common.Net< T >.UpdateDebugInfo ( int param_id )

protected

Helper for displaying debug info in Update.

Parameters

param_id Specifies the parameter index associated with the Blob.

Definition at line 1622 of file Net.cs.

Property Documentation

◆ ActiveLabelCounts

string MyCaffe.common.Net< T >.ActiveLabelCounts

get

Returns the active label counts observed during training.

Definition at line 913 of file Net.cs.

◆ blob_loss_weights

List<double> MyCaffe.common.Net< T >.blob_loss_weights

get

Returns the collection of blob loss weights.

Definition at line 2068 of file Net.cs.

◆ blob_names

List<string> MyCaffe.common.Net< T >.blob_names

get

Returns the blob names.

Definition at line 1986 of file Net.cs.

◆ blobs

BlobCollection<T> MyCaffe.common.Net< T >.blobs

get

Returns the blobs.

Definition at line 1994 of file Net.cs.

◆ BoostQueryHitPercents

string MyCaffe.common.Net< T >.BoostQueryHitPercents

get

Return the boost query hit percentages for the active datasource.

Definition at line 973 of file Net.cs.

◆ bottom_need_backward

List<List<bool> > MyCaffe.common.Net< T >.bottom_need_backward

get

Returns the collection of lists that tell whether or not the bottom of each layer needs a backward pass or not.

Definition at line 2060 of file Net.cs.

◆ bottom_vecs

List<BlobCollection<T> > MyCaffe.common.Net< T >.bottom_vecs

get

Returns the bottom vecs for each layer – usually you won't need this unless you do per-layer checks such as gradients.

Definition at line 2019 of file Net.cs.

◆ Cuda

CudaDnn<T> MyCaffe.common.Net< T >.Cuda

get

Returns the instance of CudaDnn used by this network.

Definition at line 2820 of file Net.cs.

◆ EnableBreakOnFirstNaN

bool MyCaffe.common.Net< T >.EnableBreakOnFirstNaN

getset

Enable/disable break the first NaN functionality where training stops immediately upon detecting a NaN in one of the Layer blobs.

Definition at line 844 of file Net.cs.

◆ EnableDetailedNanDetection

bool MyCaffe.common.Net< T >.EnableDetailedNanDetection

getset

Enable/disable whether or not detailed nans are detected - this will make debugging slower and is only recommended when running on a TCC enabled driver (as opposed to an WDM driver one used with the monitor).

Definition at line 854 of file Net.cs.

◆ EnableLayerDebugging

bool MyCaffe.common.Net< T >.EnableLayerDebugging

getset

Enable/disable layer debugging which causes each layer to check for NAN/INF on each forward/backward pass and throw an exception when found.

This option dramatically slows down training and is only recommended during debugging.

Definition at line 866 of file Net.cs.

◆ input_blob_indices

List<int> MyCaffe.common.Net< T >.input_blob_indices

get

Returns a list of the input Blob indexes.

Definition at line 2224 of file Net.cs.

◆ input_blobs

BlobCollection<T> MyCaffe.common.Net< T >.input_blobs

get

Returns the collection of input Blobs.

Definition at line 2200 of file Net.cs.

◆ LabelQueryEpochs

string MyCaffe.common.Net< T >.LabelQueryEpochs

get

Return the label query epochs for the active datasource.

Definition at line 954 of file Net.cs.

◆ LabelQueryHitPercents

string MyCaffe.common.Net< T >.LabelQueryHitPercents

get

Return the label query hit percentages for the active datasource.

Definition at line 935 of file Net.cs.

◆ layer_names

List<string> MyCaffe.common.Net< T >.layer_names

get

Returns the layer names.

Definition at line 1978 of file Net.cs.

◆ layer_need_backward

List<bool> MyCaffe.common.Net< T >.layer_need_backward

get

Returns a collection of items that tell whether each layer nees a backward pass or not.

Definition at line 2076 of file Net.cs.

◆ layers

List<Layer<T> > MyCaffe.common.Net< T >.layers

get

Returns the layers.

Definition at line 2002 of file Net.cs.

◆ learnable_parameters

BlobCollection<T> MyCaffe.common.Net< T >.learnable_parameters

get

Returns the learnable parameters.

Definition at line 2116 of file Net.cs.

◆ name

string MyCaffe.common.Net< T >.name

get

Returns the network name.

Definition at line 1970 of file Net.cs.

◆ net_param

NetParameter MyCaffe.common.Net< T >.net_param

get

Returns the net parameter.

Definition at line 1856 of file Net.cs.

◆ num_inputs

int MyCaffe.common.Net< T >.num_inputs

get

Returns the number of inputs.

Definition at line 2184 of file Net.cs.

◆ num_outputs

int MyCaffe.common.Net< T >.num_outputs

get

Returns the number of outputs.

Definition at line 2192 of file Net.cs.

◆ output_blob_indices

List<int> MyCaffe.common.Net< T >.output_blob_indices

get

Returns a list of the output Blob indexes.

Definition at line 2216 of file Net.cs.

◆ output_blobs

BlobCollection<T> MyCaffe.common.Net< T >.output_blobs

get

Returns the collection of output Blobs.

Definition at line 2208 of file Net.cs.

◆ param_display_names

List<string> MyCaffe.common.Net< T >.param_display_names

get

Returns the list of parameter display names.

Definition at line 2156 of file Net.cs.

◆ param_names_index

DictionaryEx<string, int> MyCaffe.common.Net< T >.param_names_index

get

Returns the dictionary look for parameter names to their indexes.

Definition at line 2140 of file Net.cs.

◆ param_owners

List<int> MyCaffe.common.Net< T >.param_owners

get

Returns the list of parameter owner indexes.

Definition at line 2148 of file Net.cs.

◆ parameters

BlobCollection<T> MyCaffe.common.Net< T >.parameters

get

Returns the parameters.

Definition at line 2084 of file Net.cs.

◆ params_lr

List<double?> MyCaffe.common.Net< T >.params_lr

get

Returns the learnable parameter learning rate multipliers.

Definition at line 2124 of file Net.cs.

◆ params_weight_decay

List<double?> MyCaffe.common.Net< T >.params_weight_decay

get

Returns the learnable parameter decay multipliers.

Definition at line 2132 of file Net.cs.

◆ phase

Phase MyCaffe.common.Net< T >.phase

get

Returns the network phase: TRAIN or TEST

Definition at line 2010 of file Net.cs.

◆ top_vecs

List<BlobCollection<T> > MyCaffe.common.Net< T >.top_vecs

get

Returns the top vecs for each layer – usually y ou won't need this unless you do per-layer checks such as gradients.

Definition at line 2028 of file Net.cs.

Event Documentation

◆ OnGetIteration

EventHandler<GetIterationArgs> MyCaffe.common.Net< T >.OnGetIteration

Specifies the OnGetIteration event that fires when a layer needs to get the current iteration from the solver.

Definition at line 136 of file Net.cs.

◆ OnGetWorkspace

EventHandler<WorkspaceArgs> MyCaffe.common.Net< T >.OnGetWorkspace

Specifies the OnGetWorkspace event that fires when the getWorkspace() function is called by a layer to get a shareable workspace to conserve GPU memory.

Definition at line 128 of file Net.cs.

◆ OnSetWorkspace

EventHandler<WorkspaceArgs> MyCaffe.common.Net< T >.OnSetWorkspace

Specifies the OnSetWorkspace event that fires when the setWorkspace() function is called by a layer to get a shareable workspace to conserve GPU memory.

Definition at line 132 of file Net.cs.

The documentation for this class was generated from the following file:

C:/Data/Data/SS_Projects/Intelligence/GitHub/MyCaffe/MyCaffe/common/Net.cs

Public Member Functions

Protected Member Functions

Properties

Events

Detailed Description

Constructor & Destructor Documentation

◆ Net()

Member Function Documentation

◆ AppendBottom()

◆ AppendParam()

◆ AppendTop()

◆ Backward()

◆ BackwardDebugInfo()

◆ blob_by_name()

◆ blob_index_by_name()

◆ bottom_ids()

◆ ClearParamDiffs()

◆ CopyInternalBlobsTo()

◆ CopyTrainedLayersFrom()

◆ CopyTrainedLayersTo() [1/2]

◆ CopyTrainedLayersTo() [2/2]

◆ Dispose() [1/2]

◆ Dispose() [2/2]

◆ FilterNet()

◆ FindBlob()

◆ FindBottomBlobsOfLayer()

◆ FindLastLayer()

◆ FindLayer() [1/2]

◆ FindLayer() [2/2]

◆ FindLayerIndexOwningBlob()

◆ FindLayerOwningBlob()

◆ FindLayers()

◆ FindLossBottomBlob()

◆ FindTopBlobsOfLayer()

◆ Forward() [1/3]

◆ Forward() [2/3]

◆ Forward() [3/3]

◆ ForwardBackward()

◆ ForwardDebugInfo()

◆ ForwardFromTo()

◆ GetDataSource()

◆ GetDebugInformation()

◆ has_blob()

◆ has_layer()

◆ Init()

◆ InputDebugInfo()

◆ InsertSplits()

◆ layer_blobs()

◆ layer_by_name()

◆ layer_index_by_name()

◆ LoadWeights()

◆ param_by_name()

◆ ReInitializeParameters()

◆ Reshape()

◆ RestorePhase()

◆ SaveWeights()

◆ set_debug_info()

◆ SetEnablePassthrough()

◆ SetLearnedParameters()

◆ SetPhase()

◆ ShareTrainedLayersWith()

◆ ShareWeights()

◆ StateMeetsRule()

◆ top_ids()

◆ ToProto()

◆ Update()

◆ UpdateDebugInfo()

Property Documentation

◆ ActiveLabelCounts

◆ blob_loss_weights

◆ blob_names

◆ blobs

◆ BoostQueryHitPercents

◆ bottom_need_backward

◆ bottom_vecs

◆ Cuda

◆ EnableBreakOnFirstNaN

◆ EnableDetailedNanDetection

◆ EnableLayerDebugging

◆ input_blob_indices