rl_param_container Class Reference

#include <reinforcement_learning.h>

Inherits soar_module::object_container< T >.

Public Types
enum	apoptosis_choices
enum	decay_choices
enum	learning_choices

Public Member Functions
	rl_param_container (agent *new_agent)
Public Member Functions inherited from soar_module::object_container< T >
void	for_each (accumulator< T * > &f)
T *	get (const char *name)
	object_container (agent *new_agent)
virtual	~object_container ()

Static Public Member Functions
static const std::vector < std::pair< std::string, param_accessor< double > * > > &	get_documentation_params ()

Data Fields
rl_apoptosis_param *	apoptosis
soar_module::decimal_param *	apoptosis_decay
rl_apoptosis_thresh_param *	apoptosis_thresh
soar_module::boolean_param *	chunk_stop
soar_module::constant_param < decay_choices > *	decay_mode
soar_module::decimal_param *	discount_rate
soar_module::decimal_param *	et_decay_rate
soar_module::decimal_param *	et_tolerance
soar_module::boolean_param *	hrl_discount
rl_learning_param *	learning
soar_module::constant_param < learning_choices > *	learning_policy
soar_module::decimal_param *	learning_rate
soar_module::boolean_param *	meta
soar_module::decimal_param *	meta_learning_rate
soar_module::boolean_param *	temporal_discount
soar_module::boolean_param *	temporal_extension
soar_module::string_param *	update_log_path

Additional Inherited Members
Protected Member Functions inherited from soar_module::object_container< T >
void	add (T *new_object)
Protected Attributes inherited from soar_module::object_container< T >
agent *	my_agent
std::map< std::string, T * > *	objects

Detailed Description

Definition at line 74 of file reinforcement_learning.h.

Member Enumeration Documentation

enum rl_param_container::apoptosis_choices

Enumerator:

apoptosis_none
apoptosis_chunks
apoptosis_rl

Definition at line 86 of file reinforcement_learning.h.

{ apoptosis_none, apoptosis_chunks, apoptosis_rl };

enum rl_param_container::decay_choices

Enumerator:

normal_decay
exponential_decay
logarithmic_decay
delta_bar_delta_decay

Definition at line 84 of file reinforcement_learning.h.

{ normal_decay, exponential_decay, logarithmic_decay, delta_bar_delta_decay };

enum rl_param_container::learning_choices

Enumerator:

sarsa
q

Definition at line 77 of file reinforcement_learning.h.

{ sarsa, q };

Constructor & Destructor Documentation

rl_param_container::rl_param_container

(

agent *

new_agent

)

Definition at line 58 of file reinforcement_learning.cpp.

References soar_module::object_container< T >::add(), soar_module::constant_param< T >::add_mapping(), apoptosis, apoptosis_chunks, apoptosis_decay, apoptosis_none, apoptosis_rl, apoptosis_thresh, chunk_stop, decay_mode, delta_bar_delta_decay, discount_rate, et_decay_rate, et_tolerance, exponential_decay, hrl_discount, learning, learning_policy, learning_rate, logarithmic_decay, meta, meta_learning_rate, soar_module::object_container< T >::my_agent, normal_decay, soar_module::off, soar_module::on, q, sarsa, temporal_discount, temporal_extension, and update_log_path.

                                                        : soar_module::param_container( new_agent )
{
    // learning
    learning = new rl_learning_param( "learning", soar_module::off, new soar_module::f_predicate<soar_module::boolean>(), new_agent );
    add( learning );

    // meta-learning-rate
    meta_learning_rate = new soar_module::decimal_param( "meta-learning-rate", 0.1, new soar_module::btw_predicate<double>( 0, 1, true ), new soar_module::f_predicate<double>() );
    add( meta_learning_rate );

    // update-log-path
    update_log_path = new soar_module::string_param( "update-log-path", "", new soar_module::predicate<const char *>(), new soar_module::f_predicate<const char *>() );
    add( update_log_path );

 // discount-rate
 discount_rate = new soar_module::decimal_param( "discount-rate", 0.9, new soar_module::btw_predicate<double>( 0, 1, true ), new soar_module::f_predicate<double>() );
 add( discount_rate );

 // learning-rate
 learning_rate = new soar_module::decimal_param( "learning-rate", 0.3, new soar_module::btw_predicate<double>( 0, 1, true ), new soar_module::f_predicate<double>() );
 add( learning_rate );

 // learning-policy
 learning_policy = new soar_module::constant_param<learning_choices>( "learning-policy", sarsa, new soar_module::f_predicate<learning_choices>() );
 learning_policy->add_mapping( sarsa, "sarsa" );
 learning_policy->add_mapping( q, "q-learning" );
 add( learning_policy );

    // decay-mode
    decay_mode = new soar_module::constant_param<decay_choices>( "decay-mode", normal_decay, new soar_module::f_predicate<decay_choices>() );
    decay_mode->add_mapping( normal_decay, "normal" );
    decay_mode->add_mapping( exponential_decay, "exp" );
    decay_mode->add_mapping( logarithmic_decay, "log" );
    decay_mode->add_mapping( delta_bar_delta_decay, "delta-bar-delta" );
    add( decay_mode );

 // eligibility-trace-decay-rate
 et_decay_rate = new soar_module::decimal_param( "eligibility-trace-decay-rate", 0, new soar_module::btw_predicate<double>( 0, 1, true ), new soar_module::f_predicate<double>() );
 add( et_decay_rate );

 // eligibility-trace-tolerance
 et_tolerance = new soar_module::decimal_param( "eligibility-trace-tolerance", 0.001, new soar_module::gt_predicate<double>( 0, false ), new soar_module::f_predicate<double>() );
 add( et_tolerance );

 // temporal-extension
 temporal_extension = new soar_module::boolean_param( "temporal-extension", soar_module::on, new soar_module::f_predicate<soar_module::boolean>() );
 add( temporal_extension );

 // hrl-discount
 hrl_discount = new soar_module::boolean_param( "hrl-discount", soar_module::off, new soar_module::f_predicate<soar_module::boolean>() );
 add( hrl_discount );

 // temporal-discount
 temporal_discount = new soar_module::boolean_param( "temporal-discount", soar_module::on, new soar_module::f_predicate<soar_module::boolean>() );
 add( temporal_discount );

 // chunk-stop
 chunk_stop = new soar_module::boolean_param( "chunk-stop", soar_module::on, new soar_module::f_predicate<soar_module::boolean>() );
 add( chunk_stop );

 // meta
 meta = new soar_module::boolean_param( "meta", soar_module::off, new soar_module::f_predicate<soar_module::boolean>() );
 add( meta );

 // apoptosis
 apoptosis = new rl_apoptosis_param( "apoptosis", apoptosis_none, new soar_module::f_predicate<apoptosis_choices>(), my_agent );
 apoptosis->add_mapping( apoptosis_none, "none" );
 apoptosis->add_mapping( apoptosis_chunks, "chunks" );
 apoptosis->add_mapping( apoptosis_rl, "rl-chunks" );
 add( apoptosis );

 // apoptosis-decay
 apoptosis_decay = new soar_module::decimal_param( "apoptosis-decay", 0.5, new soar_module::btw_predicate<double>( 0, 1, true ), new rl_apoptosis_predicate<double>( my_agent ) );
 add( apoptosis_decay );

 // apoptosis-thresh
 apoptosis_thresh = new rl_apoptosis_thresh_param( "apoptosis-thresh", -2.0, new soar_module::gt_predicate<double>( 0, false ), new rl_apoptosis_predicate<double>( my_agent ) );
 add( apoptosis_thresh );
};

Member Function Documentation

const std::vector< std::pair< std::string, param_accessor< double > * > > & rl_param_container::get_documentation_params ( )

static

Definition at line 45 of file reinforcement_learning.cpp.

Referenced by rl_perform_update(), and rl_rule_meta().

                                                                                                             {
    static std::vector<std::pair<std::string, param_accessor<double> *> > documentation_params;
    static bool initted = false;
    if (!initted) {
        initted = true;
        // Is it okay to use new here, because this is a static variable anyway,
        // so it's not going to happen more than once and shouldn't ever be cleaned up?
        documentation_params.push_back(std::make_pair("rl-updates", new rl_updates_accessor()));
        documentation_params.push_back(std::make_pair("delta-bar-delta-h", new rl_dbd_h_accessor()));
    }
    return documentation_params;
}

Field Documentation

rl_apoptosis_param* rl_param_container::apoptosis

Definition at line 104 of file reinforcement_learning.h.

Referenced by destroy_soar_agent(), do_one_top_level_phase(), excise_production(), reinitialize_soar(), retract_instantiation(), and rl_param_container().

soar_module::decimal_param* rl_param_container::apoptosis_decay

Definition at line 105 of file reinforcement_learning.h.

Referenced by rl_param_container(), and rl_apoptosis_param::set_value().

rl_apoptosis_thresh_param* rl_param_container::apoptosis_thresh

Definition at line 106 of file reinforcement_learning.h.

Referenced by rl_param_container(), and rl_apoptosis_param::set_value().

soar_module::boolean_param* rl_param_container::chunk_stop

Definition at line 100 of file reinforcement_learning.h.

Referenced by add_production_to_rete(), and rl_param_container().

soar_module::constant_param<decay_choices>* rl_param_container::decay_mode

Definition at line 93 of file reinforcement_learning.h.

Referenced by rl_param_container(), and rl_perform_update().

soar_module::decimal_param* rl_param_container::discount_rate

Definition at line 89 of file reinforcement_learning.h.

Referenced by rl_param_container(), rl_perform_update(), and rl_tabulate_reward_value_for_goal().

soar_module::decimal_param* rl_param_container::et_decay_rate

Definition at line 94 of file reinforcement_learning.h.

Referenced by rl_param_container(), and rl_perform_update().

soar_module::decimal_param* rl_param_container::et_tolerance

Definition at line 95 of file reinforcement_learning.h.

Referenced by rl_param_container(), and rl_perform_update().

soar_module::boolean_param* rl_param_container::hrl_discount

Definition at line 97 of file reinforcement_learning.h.

Referenced by rl_param_container(), and rl_tabulate_reward_value_for_goal().

rl_learning_param* rl_param_container::learning

Definition at line 88 of file reinforcement_learning.h.

Referenced by rl_enabled(), and rl_param_container().

soar_module::constant_param<learning_choices>* rl_param_container::learning_policy

Definition at line 92 of file reinforcement_learning.h.

Referenced by exploration_choose_according_to_policy(), and rl_param_container().

soar_module::decimal_param* rl_param_container::learning_rate

Definition at line 90 of file reinforcement_learning.h.

Referenced by rl_param_container(), and rl_perform_update().

soar_module::boolean_param* rl_param_container::meta

Definition at line 101 of file reinforcement_learning.h.

Referenced by rl_param_container(), rl_perform_update(), and rl_rule_meta().

soar_module::decimal_param* rl_param_container::meta_learning_rate

Definition at line 91 of file reinforcement_learning.h.

Referenced by rl_param_container(), and rl_perform_update().

soar_module::boolean_param* rl_param_container::temporal_discount

Definition at line 98 of file reinforcement_learning.h.

Referenced by rl_param_container(), rl_perform_update(), and rl_tabulate_reward_value_for_goal().

soar_module::boolean_param* rl_param_container::temporal_extension

Definition at line 96 of file reinforcement_learning.h.

Referenced by rl_param_container(), rl_perform_update(), and rl_store_data().

soar_module::string_param* rl_param_container::update_log_path

Definition at line 102 of file reinforcement_learning.h.

Referenced by rl_param_container(), and rl_perform_update().

---

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

• src/reinforcement_learning.h
• src/reinforcement_learning.cpp