The A2CConfig struct. Configuration for A2CSolver class. More...

#include <a2c_config.h>

Public Member Functions
std::ostream &	print (std::ostream &out) const
	print

void	load_from_json (const std::string &filename)
	Load the configuration from the given json file.

Public Attributes
real_t	gamma {0.99}
	Discount factor.

real_t	lambda {0.1}
	GAE lambda.

real_t	beta {0.0}
	Coefficient for accounting for entropy contribution.

real_t	policy_loss_weight { 1.0}
	policy_loss_weight. How much weight to give on the policy loss when forming the global loss

real_t	value_loss_weight {1.0}

bool	clip_policy_grad {false}
	Flag indicating whether to clip the policy grad.

bool	clip_critic_grad {false}
	Flag indicating whether to clip the critic grad.

real_t	max_grad_norm_policy {1.0}
	The value to clip the gradient for the policy.

real_t	max_grad_norm_critic {1.0}
	The value to clip the gradient for the actor.

uint_t	n_episodes {100}
	Number of training episodes.

uint_t	max_itrs_per_episode {100}
	Number of iterations per episode.

uint_t	buffer_size {100}

bool	normalize_advantages {true}

DeviceType	device_type {DeviceType::CPU}

std::string	save_model_path {""}

Detailed Description

The A2CConfig struct. Configuration for A2CSolver class.

Member Function Documentation

◆ load_from_json()

void cuberl::rl::algos::pg::A2CConfig::load_from_json ( const std::string & filename )

Load the configuration from the given json file.

◆ print()

std::ostream & cuberl::rl::algos::pg::A2CConfig::print ( std::ostream & out ) const

print

Parameters

out

Returns

Member Data Documentation

◆ beta

real_t cuberl::rl::algos::pg::A2CConfig::beta {0.0}

Coefficient for accounting for entropy contribution.

◆ buffer_size

uint_t cuberl::rl::algos::pg::A2CConfig::buffer_size {100}

◆ clip_critic_grad

bool cuberl::rl::algos::pg::A2CConfig::clip_critic_grad {false}

Flag indicating whether to clip the critic grad.

◆ clip_policy_grad

bool cuberl::rl::algos::pg::A2CConfig::clip_policy_grad {false}

Flag indicating whether to clip the policy grad.

◆ device_type

DeviceType cuberl::rl::algos::pg::A2CConfig::device_type {DeviceType::CPU}

◆ gamma

real_t cuberl::rl::algos::pg::A2CConfig::gamma {0.99}

Discount factor.

◆ lambda

real_t cuberl::rl::algos::pg::A2CConfig::lambda {0.1}

GAE lambda.

◆ max_grad_norm_critic

real_t cuberl::rl::algos::pg::A2CConfig::max_grad_norm_critic {1.0}

The value to clip the gradient for the actor.

◆ max_grad_norm_policy

real_t cuberl::rl::algos::pg::A2CConfig::max_grad_norm_policy {1.0}

The value to clip the gradient for the policy.

◆ max_itrs_per_episode

uint_t cuberl::rl::algos::pg::A2CConfig::max_itrs_per_episode {100}

Number of iterations per episode.

◆ n_episodes

uint_t cuberl::rl::algos::pg::A2CConfig::n_episodes {100}

Number of training episodes.

◆ normalize_advantages

bool cuberl::rl::algos::pg::A2CConfig::normalize_advantages {true}

◆ policy_loss_weight

real_t cuberl::rl::algos::pg::A2CConfig::policy_loss_weight { 1.0}

policy_loss_weight. How much weight to give on the policy loss when forming the global loss

◆ save_model_path

std::string cuberl::rl::algos::pg::A2CConfig::save_model_path {""}

◆ value_loss_weight

real_t cuberl::rl::algos::pg::A2CConfig::value_loss_weight {1.0}

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

libs/cuberl/include/cuberl/rl/algorithms/pg/a2c_config.h

Public Member Functions

Public Attributes