ppo_config.h

Go to the documentation of this file.

#ifndef PPO_CONFIG_H
#define PPO_CONFIG_H
 
 
#include "cuberl/base/cuberl_types.h"
//#include "rlenvs/rlenvs_consts.h"
#include "cuberl/utils/train_enum_type.h"
 
#include <ostream>
#include <string>
 
namespace cuberl {
namespace rl {
namespace algos {
namespace pg {
    
//using namespace rlenvscpp::consts;
    
struct PPOConfig
{
 
    real_t gamma{0.99};
 
    real_t epsilon{0.01};
    
    bool clip_policy_grad{false};
    
    bool clip_critic_grad{false};
 
    real_t max_grad_norm_policy{1.0};
    
    real_t max_grad_norm_critic{1.0};
    
    uint_t n_episodes{100};
 
    uint_t max_itrs_per_episode{100};
 
    uint_t buffer_size{100};
 
    uint_t max_passes_over_batch{4};
 
 
    real_t clip_epsilon {0.5};
 
    bool normalize_advantages{true};
 
    DeviceType device_type{DeviceType::CPU};
 
    std::string save_model_path{""};
    
    std::ostream& print(std::ostream& out)const;
    
    void load_from_json(const std::string& filename);
};
 
    
inline
std::ostream& operator<<(std::ostream& out, const PPOConfig& opts){
    return opts.print(out);
}
 
} // pg
} // algos
} // rl
} // cuberl
 
#endif