MyCaffe.trainers.pg.simple.TrainerPG< T > Class Template Reference

The TrainerPG implements a simple Policy Gradient trainer inspired by Andrej Karpathy's blog posed referenced. More...

Inheritance diagram for MyCaffe.trainers.pg.simple.TrainerPG< T >:

Public Member Functions
	TrainerPG (MyCaffeControl< T > mycaffe, PropertySet properties, CryptoRandom random, IxTrainerCallback icallback)
	The constructor. More...
void	Dispose ()
	Releases all resources used. More...
bool	Initialize ()
	Initialize the trainer. More...
bool	Shutdown (int nWait)
	Shutdown the trainer. More...
ResultCollection	RunOne (int nDelay=1000)
	Run a single cycle on the environment after the delay. More...
byte[]	Run (int nN, PropertySet runProp, out string type)
	Run a set of iterations and return the resuts. More...
bool	Test (int nN, ITERATOR_TYPE type)
	Run the test cycle - currently this is not implemented. More...
bool	Train (int nN, ITERATOR_TYPE type, TRAIN_STEP step)
	Train the network using a modified PG training algorithm optimized for GPU use. More...

Detailed Description

The TrainerPG implements a simple Policy Gradient trainer inspired by Andrej Karpathy's blog posed referenced.

See also

1. Deep Reinforcement Learning: Pong from Pixels, by Andrej Karpathy, 2016, Github.io

2. GitHub: karpathy/pg-pong.py, by Andrej Karpathy, 2016, Github

3. CS231n Convolution Neural Networks for Visual Recognition by Karpathy, Stanford

4. MyCaffe: A Complete C# Re-Write of Caffe with Reinforcement Learning by D. Brown, 2018, arXiv

Definition at line 25 of file TrainerPG.cs.

Constructor & Destructor Documentation

TrainerPG()

MyCaffe.trainers.pg.simple.TrainerPG< T >.TrainerPG	(	MyCaffeControl< T >	mycaffe,
		PropertySet	properties,
		CryptoRandom	random,
		IxTrainerCallback	icallback
	)

The constructor.

Parameters

mycaffe	Specifies the MyCaffeControl to use for learning and prediction.
properties	Specifies the property set containing the key/value pairs of property settings.
random	Specifies a Random number generator used for random selection.
icallback	Specifies the callback for parent notifications and queries.

Definition at line 39 of file TrainerPG.cs.

Member Function Documentation

Dispose()

void MyCaffe.trainers.pg.simple.TrainerPG< T >.Dispose

(

)

Releases all resources used.

Definition at line 50 of file TrainerPG.cs.

Initialize()

bool MyCaffe.trainers.pg.simple.TrainerPG< T >.Initialize

(

)

Initialize the trainer.

Returns

Returns true.

Implements MyCaffe.trainers.IxTrainer.

Definition at line 58 of file TrainerPG.cs.

Run()

byte[] MyCaffe.trainers.pg.simple.TrainerPG< T >.Run	(	int	nN,
		PropertySet	runProp,
		out string	type
	)

Run a set of iterations and return the resuts.

Parameters

nN	Specifies the number of samples to run.
runProp	Optionally specifies properties to use when running.
type	Returns the data type contained in the byte stream.

Returns

The results of the run containing the action are returned as a byte stream.

Implements MyCaffe.trainers.IxTrainerRL.

Definition at line 116 of file TrainerPG.cs.

RunOne()

ResultCollection MyCaffe.trainers.pg.simple.TrainerPG< T >.RunOne

(

int

nDelay = 1000

)

Run a single cycle on the environment after the delay.

Parameters

nDelay

Specifies a delay to wait before running.

Returns

The results of the run containing the action are returned.

Implements MyCaffe.trainers.IxTrainerRL.

Definition at line 100 of file TrainerPG.cs.

Shutdown()

bool MyCaffe.trainers.pg.simple.TrainerPG< T >.Shutdown

(

int

nWait

)

Shutdown the trainer.

Parameters

nWait

Specifies a wait in ms. for the shutdown to complete.

Returns

Returns true.

Implements MyCaffe.trainers.IxTrainer.

Definition at line 82 of file TrainerPG.cs.

Test()

bool MyCaffe.trainers.pg.simple.TrainerPG< T >.Test	(	int	nN,
		ITERATOR_TYPE	type
	)

Run the test cycle - currently this is not implemented.

Parameters

nN	Specifies the number of iterations (based on the ITERATION_TYPE) to run, or -1 to ignore.
type	Specifies the iteration type (default = ITERATION).

Returns

A value of true is returned when handled, false otherwise.

Implements MyCaffe.trainers.IxTrainer.

Definition at line 132 of file TrainerPG.cs.

Train()

bool MyCaffe.trainers.pg.simple.TrainerPG< T >.Train	(	int	nN,
		ITERATOR_TYPE	type,
		TRAIN_STEP	step
	)

Train the network using a modified PG training algorithm optimized for GPU use.

Parameters

nN	Specifies the number of iterations (based on the ITERATION_TYPE) to run, or -1 to ignore.
type	Specifies the iteration type (default = ITERATION).
step	Specifies the stepping mode to use (when debugging).

Returns

A value of true is returned when handled, false otherwise.

Implements MyCaffe.trainers.IxTrainer.

Definition at line 158 of file TrainerPG.cs.

---

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

• C:/Data/Data/SS_Projects/Intelligence/GitHub/MyCaffe/MyCaffe.trainers/pg.simple/TrainerPG.cs