cpid Namespace Reference

The TorchCraftAI training library. More...

Namespaces
	detail
	distributed
	netutils

Classes
class	AsyncBatcher
class	BaseSampler
	A sampler takes the output of the model, and outputs an action accordingly. More...
struct	BatchedPGReplayBufferFrame
class	BatchedPGTrainer
	Off policy policy gradient with a critic. More...
class	BlobPublisher
	Publisher for ZeroMQ PUB-SUB pattern. More...
class	BlobStorage
	A simple interface for key-value data blob stores. More...
class	BlobStorageDisk
class	BlobStorageRedis
	Blob storage in Redis. More...
class	BlobSubscriber
	Subscriber for ZeroMQ PUB-SUB pattern. More...
class	CentralCpid2kTrainer
	A trainer that sends episodes to one or more central instances. More...
class	CentralTrainer
	A trainer that sends episodes to one or more central instances. More...
class	Checkpointer
class	ContinuousDeterministicSampler
	This sampler expects as input an unordered_map<string, Variant> containing an entry policyKey, which is a tensor of size [b, n]. More...
class	ContinuousGaussianSampler
	This sampler expects as input an unordered_map<string, Variant>, containing an entry policyKey, which is a tensor of size [b, n]. More...
class	Cpid2kGlobalState
	Encapsulates information about the participating peers in a cpid2k job. More...
class	Cpid2kHeartBeater
	Periodically sends out heartbeats to a Redis instance. More...
class	Cpid2kMetrics
	Helper class to aggregate metrics locally, and send them reguarly as events in the redis database as key 'prefix:metricEvents'. More...
class	Cpid2kWorker
	Helper class for job coordination via a central Redis instance. More...
struct	Cpid2kWorkerInfo
class	DiscreteMaxSampler
	This sampler expects as input an unordered_map<string, Variant>, containing an entry QKey, which is a tensor of size [b, n]. More...
struct	EpisodeData
struct	EpisodeTuple
class	EpsGreedySampler
	This sampler expects as input an unordered_map<string, Variant> containing an entry QKey, which is a tensor of size [b, n]. More...
class	ESTrainer
class	Evaluator
struct	HandleGuard
class	MetricsContext
class	MultinomialSampler
	This sampler expects as input an unordered_map<string, Variant>, which contains an entry policyKey, which is a tensor of size [b, n]. More...
struct	OnlineZORBReplayBufferFrame
	State, Action taken, reward Taking in an additional action taken allows you to not just take the max action, but use your own inference strategy, for example, if some actions are invalid. More...
class	OnlineZORBTrainer
	This is a OnlineZORBTrainer that works with multiple actions per frame. More...
struct	pairhash
class	permanent_lock
	This exactly a unique lock that doesn't unlock on delete. More...
class	priority_lock
	This is exactly an unique_lock without automatic lock, except that the lock functions accepts a priority. More...
class	priority_mutex
	This class implements a mutex that offers some control over the priority of the waiting threads. More...
class	RedisClient
	Simple, synchronous C++ wrapper for the Hiredis Redis client. More...
class	RedisReply
	Wrapper class for redisReply from Hiredis. More...
class	RedisStore
	c10d::Store for rendez-vous backed by Redis. More...
class	ReplayBuffer
	Stores an unordered_map[GameUID] = unordered_map[int, Episode] All the public functions here should be autolocking and therefore relatively thread safe. More...
struct	ReplayBufferFrame
	Stub base class for replay buffer frames. More...
class	ReqRepClient
	A request-reply client backed by ZeroMQ. More...
class	ReqRepServer
	A request-reply server backed by ZeroMQ. More...
struct	RewardBufferFrame
class	SubBatchAsyncBatcher
	A batcher that can operate on (already) batched data Should be used when features have a variable batch dimension, for instance the number of units controlled. More...
class	Trainer
	The Trainer should be shared amongst multiple different nodes, and attached to a single Module. More...
class	ZeroMQBufferedConsumer
	A buffered consumer that sends data via ZeroMQ. More...
class	ZeroMQBufferedProducer
	A buffered producer that obtains data via ZeroMQ. More...
class	ZOBackpropModel
	See trainer for output format of these models. More...

Typedefs
using	EpisodeServer = ZeroMQBufferedProducer< EpisodeData >
using	EpisodeClient = ZeroMQBufferedConsumer< EpisodeData >
using	GameUID = std::string
using	EpisodeKey = std::string
using	EpisodeHandle = Trainer::EpisodeHandle

Functions
	DEFINE_FLAG_OPERATORS (BlobPublisher::DataFlags)
float	kaiming_normal_ (torch::Tensor tensor, double gain)
ag::Optimizer	selectOptimizer (std::shared_ptr< torch::nn::Module > module)
std::map< std::string, std::string >	optimizerFlags ()
GameUID	genGameUID ()
std::ostream &	operator<< (std::ostream &os, EpisodeHandle const &handle)

Variables
const constexpr auto	kDefaultEpisodeKey = ""
const std::string	kValueKey = "V"
const std::string	kQKey = "Q"
const std::string	kPiKey = "Pi"
const std::string	kSigmaKey = "std"
const std::string	kActionQKey = "actionQ"
const std::string	kActionKey = "action"
const std::string	kPActionKey = "pAction"

Detailed Description

The TorchCraftAI training library.

Typedef Documentation

using cpid::EpisodeClient = typedef ZeroMQBufferedConsumer<EpisodeData>

using cpid::EpisodeHandle = typedef Trainer::EpisodeHandle

using cpid::EpisodeKey = typedef std::string

using cpid::EpisodeServer = typedef ZeroMQBufferedProducer<EpisodeData>

using cpid::GameUID = typedef std::string

Function Documentation

cpid::DEFINE_FLAG_OPERATORS

(

BlobPublisher::DataFlags

)

GameUID cpid::genGameUID

(

)

float cpid::kaiming_normal_	(	torch::Tensor	tensor,
		double	gain
	)

std::ostream& cpid::operator<<	(	std::ostream &	os,
		EpisodeHandle const &	handle
	)

std::map< std::string, std::string > cpid::optimizerFlags

(

)

ag::Optimizer cpid::selectOptimizer

(

std::shared_ptr< torch::nn::Module >

module

)

Variable Documentation

const std::string cpid::kActionKey = "action"

inline

const std::string cpid::kActionQKey = "actionQ"

inline

const constexpr auto cpid::kDefaultEpisodeKey = ""

const std::string cpid::kPActionKey = "pAction"

inline

const std::string cpid::kPiKey = "Pi"

inline

const std::string cpid::kQKey = "Q"

inline

const std::string cpid::kSigmaKey = "std"

inline

const std::string cpid::kValueKey = "V"

inline