The Brain uses the instance of MyCaffe (e.g. the open project) to run new actions and train the network. More...

Inheritance diagram for MyCaffe.trainers.dqn.noisy.simple.Brain< T >:

Public Member Functions
	Brain (MyCaffeControl< T > mycaffe, PropertySet properties, CryptoRandom random, Phase phase)
	The constructor. More...

void	Dispose ()
	Release all resources used by the Brain. More...

GetDataArgs	getDataArgs (Phase phase, int nAction)
	Returns the GetDataArgs used to retrieve new data from the envrionment implemented by derived parent trainer. More...

SimpleDatum	Preprocess (StateBase s, bool bUseRawInput, out bool bDifferent, bool bReset=false)
	Preprocesses the data. More...

int	act (SimpleDatum sd, SimpleDatum sdClip, int nActionCount)
	Returns the action from running the model. The action returned is either randomly selected (when using Exploration), or calculated via a forward pass (when using Exploitation). More...

bool	GetModelUpdated ()
	Get whether or not the model has been udpated or not. More...

void	UpdateTargetModel ()
	The UpdateTargetModel transfers the trained layers from the active Net to the target Net. More...

void	Train (int nIteration, MemoryCollection rgSamples, int nActionCount)
	Train the model at the current iteration. More...

void	OnOverlay (OverlayArgs e)
	The OnOverlay callback is called just before displaying the gym image, thus allowing for an overlay to be applied to the image. More...

void	SaveWeights (string strFile)
	Save the weight and bias values to file. More...

void	LoadWeights (string strFile)
	Load the weight and bias values from file. More...

Properties
int	BatchSize `[get]`
	Returns the batch size defined by the model. More...

Log	Log `[get]`
	Returns the output log. More...

CancelEvent	Cancel `[get]`
	Returns the Cancel event used to cancel all MyCaffe tasks. More...

Detailed Description

The Brain uses the instance of MyCaffe (e.g. the open project) to run new actions and train the network.

Template Parameters

T	Specifies the base type, which should be the same base type used for MyCaffe. This type is either double or float.

Definition at line 399 of file TrainerNoisyDqn.cs.

Constructor & Destructor Documentation

◆ Brain()

MyCaffe.trainers.dqn.noisy.simple.Brain< T >.Brain	(	MyCaffeControl< T >	mycaffe,
		PropertySet	properties,
		CryptoRandom	random,
		Phase	phase
	)

The constructor.

Parameters

mycaffe	Specifies the instance of MyCaffe assoiated with the open project - when using more than one Brain, this is the master project.
properties	Specifies the properties passed into the trainer.
random	Specifies the random number generator used.
phase	Specifies the phase under which to run.

Definition at line 439 of file TrainerNoisyDqn.cs.

Member Function Documentation

◆ act()

int MyCaffe.trainers.dqn.noisy.simple.Brain< T >.act	(	SimpleDatum	sd,
		SimpleDatum	sdClip,
		int	nActionCount
	)

Returns the action from running the model. The action returned is either randomly selected (when using Exploration), or calculated via a forward pass (when using Exploitation).

Parameters

sd	Specifies the data to run the model on.
sdClip	Specifies the clip data (if any exits).
nActionCount	Returns the number of actions in the action set.

Returns: The action value is returned.

Definition at line 628 of file TrainerNoisyDqn.cs.

◆ Dispose()

void MyCaffe.trainers.dqn.noisy.simple.Brain< T >.Dispose ( )

Release all resources used by the Brain.

Definition at line 509 of file TrainerNoisyDqn.cs.

◆ getDataArgs()

GetDataArgs MyCaffe.trainers.dqn.noisy.simple.Brain< T >.getDataArgs	(	Phase	phase,
		int	nAction
	)

Returns the GetDataArgs used to retrieve new data from the envrionment implemented by derived parent trainer.

Parameters

phase	Specifies the phase under which to get the data.
nAction	Specifies the action to run, or -1 to reset the environment.

Returns: A new GetDataArgs is returned.

Definition at line 554 of file TrainerNoisyDqn.cs.

◆ GetModelUpdated()

bool MyCaffe.trainers.dqn.noisy.simple.Brain< T >.GetModelUpdated ( )

Get whether or not the model has been udpated or not.

Returns: If the model has been updated from the last call to this function, true is returned, otherwise false is returned.

Definition at line 645 of file TrainerNoisyDqn.cs.

◆ LoadWeights()

void MyCaffe.trainers.dqn.noisy.simple.Brain< T >.LoadWeights ( string strFile )

Load the weight and bias values from file.

Parameters

strFile Specifies the file to load the weights from.

Definition at line 1138 of file TrainerNoisyDqn.cs.

◆ OnOverlay()

void MyCaffe.trainers.dqn.noisy.simple.Brain< T >.OnOverlay ( OverlayArgs e )

The OnOverlay callback is called just before displaying the gym image, thus allowing for an overlay to be applied to the image.

Parameters

e	Specifies the arguments to the callback which contains the original display image.

Implements MyCaffe.trainers.IxTrainerGetDataCallback.

Definition at line 987 of file TrainerNoisyDqn.cs.

◆ Preprocess()

SimpleDatum MyCaffe.trainers.dqn.noisy.simple.Brain< T >.Preprocess	(	StateBase	s,
		bool	bUseRawInput,
		out bool	bDifferent,
		bool	bReset = `false`
	)

Preprocesses the data.

Parameters

s	Specifies the state and data to use.
bUseRawInput	Specifies whether or not to use the raw data true, or a difference of the current and previous data false (default = false).
bDifferent	Returns whether or not the current state data is different from the previous - note this is only set when NOT using raw input, otherwise true is always returned.
bReset	Optionally, specifies to reset the last sd to null.

Returns: The preprocessed data is returned.

Definition at line 592 of file TrainerNoisyDqn.cs.

◆ SaveWeights()

void MyCaffe.trainers.dqn.noisy.simple.Brain< T >.SaveWeights ( string strFile )

Save the weight and bias values to file.

Parameters

strFile Specifies the file to save the weights to.

Definition at line 1089 of file TrainerNoisyDqn.cs.

◆ Train()

void MyCaffe.trainers.dqn.noisy.simple.Brain< T >.Train	(	int	nIteration,
		MemoryCollection	rgSamples,
		int	nActionCount
	)

Train the model at the current iteration.

Parameters

nIteration	Specifies the current iteration.
rgSamples	Contains the samples to train the model with along with the priorities associated with the samples.
nActionCount	Specifies the number of actions in the action set.

Definition at line 670 of file TrainerNoisyDqn.cs.

◆ UpdateTargetModel()

void MyCaffe.trainers.dqn.noisy.simple.Brain< T >.UpdateTargetModel ( )

The UpdateTargetModel transfers the trained layers from the active Net to the target Net.

Definition at line 655 of file TrainerNoisyDqn.cs.

Property Documentation

◆ BatchSize

int MyCaffe.trainers.dqn.noisy.simple.Brain< T >.BatchSize

get

Returns the batch size defined by the model.

Definition at line 563 of file TrainerNoisyDqn.cs.

◆ Cancel

CancelEvent MyCaffe.trainers.dqn.noisy.simple.Brain< T >.Cancel

get

Returns the Cancel event used to cancel all MyCaffe tasks.

Definition at line 579 of file TrainerNoisyDqn.cs.

◆ Log

Log MyCaffe.trainers.dqn.noisy.simple.Brain< T >.Log

get

Returns the output log.

Definition at line 571 of file TrainerNoisyDqn.cs.

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

C:/Data/Data/SS_Projects/Intelligence/GitHub/MyCaffe/MyCaffe.trainers/dqn.noisy.simple/TrainerNoisyDqn.cs

Public Member Functions

Properties

Detailed Description

Constructor & Destructor Documentation

◆ Brain()

Member Function Documentation

◆ act()

◆ Dispose()

◆ getDataArgs()

◆ GetModelUpdated()

◆ LoadWeights()

◆ OnOverlay()

◆ Preprocess()

◆ SaveWeights()

◆ Train()

◆ UpdateTargetModel()

Property Documentation

◆ BatchSize

◆ Cancel

◆ Log