fanasina/y_project
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[update] learn to drive: decrease learning rate when very good rewards

Browse Source
This commit is contained in:
fanasina
2026-02-11 00:28:28 +01:00
parent a4012693cd
commit d8ae2729df
1 changed files with 3 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
deepQlearn_0/src/deepQlearning/learn_to_drive.c
+3 -1
View File
@@ -276,6 +276,7 @@ void free_RL_agent(struct RL_agent *rlAgent){
} }
#define ACCEPTABLE_REWARD 1000 #define ACCEPTABLE_REWARD 1000
#define VERY_GOOD_REWARD 10000
#define UPDATE_PARAMS 1 #define UPDATE_PARAMS 1
#define UPDATE_EXPLOR_FAC 1 #define UPDATE_EXPLOR_FAC 1
@@ -322,7 +323,8 @@ void train_qlearning(struct RL_agent * rlAgent,
#if UPDATE_PARAMS #if UPDATE_PARAMS
if((car_status->cumulative_reward > ACCEPTABLE_REWARD) || (rlAgent->status->nb_episodes % 100 == 0) ){ if((car_status->cumulative_reward > ACCEPTABLE_REWARD) || (rlAgent->status->nb_episodes % 100 == 0) ){
float new_value = ( (net_main->learning_rate < qlParams->minimum_threshold_learning_rate /*0.0001*/) ? net_main->learning_rate :(net_main->learning_rate ) * qlParams->factor_update_learning_rate /*0.995*/ ); float new_value = ( (net_main->learning_rate < qlParams->minimum_threshold_learning_rate /*0.0001*/) ? net_main->learning_rate :(net_main->learning_rate ) * qlParams->factor_update_learning_rate /*0.995*/ );
UPDATE_ATTRIBUTE_NEURONE_IN_ALL_LAYERS(TYPE_FLOAT, net_main, learning_rate, new_value); if((car_status->cumulative_reward > VERY_GOOD_REWARD) ) new_value = (net_main->learning_rate ) * qlParams->factor_update_learning_rate /*0.995*/ ;
UPDATE_ATTRIBUTE_NEURONE_IN_ALL_LAYERS(TYPE_FLOAT, net_main, learning_rate, new_value);
qlParams->learning_rate = new_value; qlParams->learning_rate = new_value;
#if UPDATE_EXPLOR_FAC #if UPDATE_EXPLOR_FAC
qlParams->exploration_factor = (qlParams->exploration_factor < qlParams->minimum_threshold_exploration_factor) ? qlParams->exploration_factor : qlParams->exploration_factor * qlParams->factor_update_exploration_factor ; qlParams->exploration_factor = (qlParams->exploration_factor < qlParams->minimum_threshold_exploration_factor) ? qlParams->exploration_factor : qlParams->exploration_factor * qlParams->factor_update_exploration_factor ;