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

add some test deep reinforcement learning

Browse Source
This commit is contained in:
fanasina
2024-07-18 02:10:30 +02:00
parent 491d0c283f
commit e64cef1688
6 changed files with 484 additions and 34 deletions
Download Patch File Download Diff File Expand all files Collapse all files
deepQlearn_0/src/deepQlearning/learn_to_drive.c
+17
View File
@@ -352,6 +352,16 @@ if(/*(qlStatus->nb_episodes %125 == 0) &&*/ pprint->printed){
}
}
char *fileNameDateScore(char * pre, char* post,size_t score){
char *filename=malloc(256);
time_t t = time(NULL);
struct tm tm = *localtime(&t);
sprintf(filename,"%s%d%02d%02d_%02dh%02dm%02ds_%ld%s",pre, tm.tm_year + 1900, tm.tm_mon + 1, tm.tm_mday, tm.tm_hour, tm.tm_min, tm.tm_sec,score,post);
return filename;
}
void learn_to_drive(struct RL_agent * rlAgent){
int action;
@@ -387,7 +397,14 @@ void learn_to_drive(struct RL_agent * rlAgent){
//push_back_list_TYPE_L_INT(qlStatus->list_main_cumul, car_status->cumulative_reward);
// printf(" cumul : %ld ", car_status->cumulative_reward);
if(car_status->cumulative_reward > qlStatus->progress_best_cumul->end_list->value){
push_back_list_TYPE_L_INT(qlStatus->progress_best_cumul, car_status->cumulative_reward);
char *file = fileNameDateScore(".ff_main_",".txt",car_status->cumulative_reward);
EXPORT_TO_FILE_TENSOR_ATTRIBUTE_IN_NNEURONS(TYPE_FLOAT, rlAgent->networks->main_net ,weight_in, file);
free(file);
file = fileNameDateScore(".ff_target_",".txt",car_status->cumulative_reward);
EXPORT_TO_FILE_TENSOR_ATTRIBUTE_IN_NNEURONS(TYPE_FLOAT, rlAgent->networks->target_net ,weight_in, file);
free(file);
}
break;
}
deepQlearn_0/src/deepQlearning/learn_to_drive.h
+1
View File
@@ -14,6 +14,7 @@
#include "neuron_t/neuron_t.h"
#include "neuron_t/nneuron_t_file.h"
#include "list_t/list_t.h"
deepQlearn_0/src/deepQlearning/vehicle.c
+6 -3
View File
@@ -342,9 +342,12 @@ void print2D_blocks_indexOne_withPoint(struct blocks *blk, float scale_x, float
if(is_in_blocks(blk_point, coord))
printf("\033[0;31m"); // red
int in = is_in_blocks(blk,coord);
if(in)
printf("%d",in);
else
if(in){
if(in>9){
int div=in%10;
printf("%d",div);
}else printf("%d",in);
}else
printf("."); //printf(" ");
printf("\033[0;37m"); // white
}
deepQlearn_0/test/is_good.c
+385 -25
View File
@@ -457,7 +457,7 @@ TEST(first_learn_vehicle_rev50_8){
int randomRange = 500;
size_t nb_prod_thread = 2;
size_t nb_calc_thread = 4;
float learning_rate = 0.001;
float learning_rate =0.00001 /*0.001*/;
struct networks_qlearning *nnetworks = create_nework_qlearning(
pconf,
randomize, minR, maxR, randomRange,
@@ -465,7 +465,10 @@ TEST(first_learn_vehicle_rev50_8){
learning_rate
);
struct status_qlearning *qlstatus = create_status_qlearning ();
EXTRACT_FILE_TO_TENSOR_ATTRIBUTE_NNEURONS(TYPE_FLOAT, nnetworks->main_net, weight_in, ".ff_main_20240717_01h42m16s_5300.txt");
EXTRACT_FILE_TO_TENSOR_ATTRIBUTE_NNEURONS(TYPE_FLOAT, nnetworks->target_net, weight_in, ".ff_target_20240717_01h42m16s_5300.txt");
struct status_qlearning *qlstatus = create_status_qlearning ();
struct delay_params *dly = create_delay_params (
500/*size_t delay_between_episodes*/,
50/*size_t delay_between_games*/
@@ -475,7 +478,7 @@ TEST(first_learn_vehicle_rev50_8){
0.95/*float gamma*/,
learning_rate,
0 /* (not used!)float discount_factor*/,
0.99 /*float exploration_factor*/,
0.0001 /* 0.99*/ /*float exploration_factor*/,
20/*long int nb_training_before_update_weight_in_target*/,
10000/*size_t number_episodes*/
);
@@ -496,7 +499,8 @@ TEST(first_learn_vehicle_rev50_8){
pprint /*struct print_params * pprint*/,
qlparams/*struct qlearning_params *qlearnParams*/
);
char c;
scanf("%c",&c);
learn_to_drive(rlAgent);
free_RL_agent(rlAgent);
@@ -515,9 +519,26 @@ TEST(first_learn_vehicle_50__9){
size_t dim= 2;
struct blocks * path = create_blocks(nb_block, dim);
#if 1
copy_coordinate(path->lower_bound_block[0], (float[]){0,0});
copy_coordinate(path->upper_bound_block[0], (float[]){150,250});
copy_coordinate(path->lower_bound_block[1], (float[]){150,0});
copy_coordinate(path->upper_bound_block[1], (float[]){250,150});
copy_coordinate(path->lower_bound_block[2], (float[]){250,80});
copy_coordinate(path->upper_bound_block[2], (float[]){360,200});
copy_coordinate(path->lower_bound_block[3], (float[]){360,70});
copy_coordinate(path->upper_bound_block[3], (float[]){600,170});
copy_coordinate(path->lower_bound_block[4], (float[]){600,90});
copy_coordinate(path->upper_bound_block[4], (float[]){760,300});
copy_coordinate(path->lower_bound_block[5], (float[]){300,300});
copy_coordinate(path->upper_bound_block[5], (float[]){760,350});
copy_coordinate(path->lower_bound_block[6], (float[]){0,250});
copy_coordinate(path->upper_bound_block[6], (float[]){410,300});
/*
copy_coordinate(path->lower_bound_block[4], (float[]){0,0});
copy_coordinate(path->upper_bound_block[4], (float[]){150,250});
copy_coordinate(path->lower_bound_block[3], (float[]){150,40});
@@ -534,22 +555,6 @@ TEST(first_learn_vehicle_50__9){
copy_coordinate(path->upper_bound_block[5], (float[]){410,300});
/*
copy_coordinate(path->lower_bound_block[0], (float[]){0,0});
copy_coordinate(path->upper_bound_block[0], (float[]){150,250});
copy_coordinate(path->lower_bound_block[1], (float[]){150,0});
copy_coordinate(path->upper_bound_block[1], (float[]){250,150});
copy_coordinate(path->lower_bound_block[2], (float[]){250,80});
copy_coordinate(path->upper_bound_block[2], (float[]){360,200});
copy_coordinate(path->lower_bound_block[3], (float[]){360,70});
copy_coordinate(path->upper_bound_block[3], (float[]){600,170});
copy_coordinate(path->lower_bound_block[4], (float[]){600,90});
copy_coordinate(path->upper_bound_block[4], (float[]){760,300});
copy_coordinate(path->lower_bound_block[5], (float[]){300,300});
copy_coordinate(path->upper_bound_block[5], (float[]){760,350});
copy_coordinate(path->lower_bound_block[6], (float[]){0,250});
copy_coordinate(path->upper_bound_block[6], (float[]){410,300});
copy_coordinate(path->lower_bound_block[0], (float[]){0,0});
copy_coordinate(path->upper_bound_block[0], (float[]){100,250});
copy_coordinate(path->lower_bound_block[1], (float[]){100,0});
@@ -616,15 +621,22 @@ TEST(first_learn_vehicle_50__9){
int randomRange = 500;
size_t nb_prod_thread = 2;
size_t nb_calc_thread = 4;
float learning_rate = 0.001;
float learning_rate = 0.00001 /* 0.001*/;
struct networks_qlearning *nnetworks = create_nework_qlearning(
pconf,
randomize, minR, maxR, randomRange,
nb_prod_thread, nb_calc_thread,
learning_rate
);
/*
EXTRACT_FILE_TO_TENSOR_ATTRIBUTE_NNEURONS(TYPE_FLOAT, nnetworks->main_net, weight_in, ".ff_main_20240717_01h42m16s_5300.txt");
EXTRACT_FILE_TO_TENSOR_ATTRIBUTE_NNEURONS(TYPE_FLOAT, nnetworks->target_net, weight_in, ".ff_target_20240717_01h42m16s_5300.txt");
*/
struct status_qlearning *qlstatus = create_status_qlearning ();
EXTRACT_FILE_TO_TENSOR_ATTRIBUTE_NNEURONS(TYPE_FLOAT, nnetworks->main_net, weight_in, ".ff_main_20240717_09h11m09s_1700.txt");
EXTRACT_FILE_TO_TENSOR_ATTRIBUTE_NNEURONS(TYPE_FLOAT, nnetworks->target_net, weight_in, ".ff_target_20240717_09h11m09s_1700.txt");
struct status_qlearning *qlstatus = create_status_qlearning ();
struct delay_params *dly = create_delay_params (
500/*size_t delay_between_episodes*/,
50/*size_t delay_between_games*/
@@ -634,7 +646,7 @@ TEST(first_learn_vehicle_50__9){
0.95/*float gamma*/,
learning_rate,
0 /* (not used!)float discount_factor*/,
0.99 /*float exploration_factor*/,
0.0001/*0.99*/ /*float exploration_factor*/,
20/*long int nb_training_before_update_weight_in_target*/,
10000/*size_t number_episodes*/
);
@@ -656,6 +668,7 @@ TEST(first_learn_vehicle_50__9){
qlparams/*struct qlearning_params *qlearnParams*/
);
learn_to_drive(rlAgent);
free_RL_agent(rlAgent);
@@ -668,6 +681,353 @@ TEST(first_learn_vehicle_50__9){
#if 1
TEST(first_learn_vehicle_50__10){
size_t nb_block = 7;
size_t dim= 2;
struct blocks * path = create_blocks(nb_block, dim);
#if 1
copy_coordinate(path->lower_bound_block[0], (float[]){0,0});
copy_coordinate(path->upper_bound_block[0], (float[]){100,250});
copy_coordinate(path->lower_bound_block[1], (float[]){100,0});
copy_coordinate(path->upper_bound_block[1], (float[]){250,80});
copy_coordinate(path->lower_bound_block[2], (float[]){250,0});
copy_coordinate(path->upper_bound_block[2], (float[]){360,140});
copy_coordinate(path->lower_bound_block[3], (float[]){360,70});
copy_coordinate(path->upper_bound_block[3], (float[]){600,140});
copy_coordinate(path->lower_bound_block[4], (float[]){600,90});
copy_coordinate(path->upper_bound_block[4], (float[]){720,300});
copy_coordinate(path->lower_bound_block[5], (float[]){300,300});
copy_coordinate(path->upper_bound_block[5], (float[]){720,350});
copy_coordinate(path->lower_bound_block[6], (float[]){0,250});
copy_coordinate(path->upper_bound_block[6], (float[]){410,300});
/*
copy_coordinate(path->lower_bound_block[4], (float[]){0,0});
copy_coordinate(path->upper_bound_block[4], (float[]){150,250});
copy_coordinate(path->lower_bound_block[3], (float[]){150,40});
copy_coordinate(path->upper_bound_block[3], (float[]){250,150});
copy_coordinate(path->lower_bound_block[2], (float[]){250,80});
copy_coordinate(path->upper_bound_block[2], (float[]){360,200});
copy_coordinate(path->lower_bound_block[1], (float[]){360,70});
copy_coordinate(path->upper_bound_block[1], (float[]){600,150});
copy_coordinate(path->lower_bound_block[0], (float[]){600,90});
copy_coordinate(path->upper_bound_block[0], (float[]){760,300});
copy_coordinate(path->lower_bound_block[6], (float[]){260,300});
copy_coordinate(path->upper_bound_block[6], (float[]){760,360});
copy_coordinate(path->lower_bound_block[5], (float[]){0,250});
copy_coordinate(path->upper_bound_block[5], (float[]){410,300});
copy_coordinate(path->lower_bound_block[0], (float[]){0,0});
copy_coordinate(path->upper_bound_block[0], (float[]){150,250});
copy_coordinate(path->lower_bound_block[1], (float[]){150,0});
copy_coordinate(path->upper_bound_block[1], (float[]){250,150});
copy_coordinate(path->lower_bound_block[2], (float[]){250,80});
copy_coordinate(path->upper_bound_block[2], (float[]){360,200});
copy_coordinate(path->lower_bound_block[3], (float[]){360,70});
copy_coordinate(path->upper_bound_block[3], (float[]){600,170});
copy_coordinate(path->lower_bound_block[4], (float[]){600,90});
copy_coordinate(path->upper_bound_block[4], (float[]){760,300});
copy_coordinate(path->lower_bound_block[5], (float[]){300,300});
copy_coordinate(path->upper_bound_block[5], (float[]){760,350});
copy_coordinate(path->lower_bound_block[6], (float[]){0,250});
copy_coordinate(path->upper_bound_block[6], (float[]){410,300});
copy_coordinate(path->lower_bound_block[0], (float[]){0,300});
copy_coordinate(path->upper_bound_block[0], (float[]){400,700});
copy_coordinate(path->lower_bound_block[1], (float[]){100,0});
copy_coordinate(path->upper_bound_block[1], (float[]){1000,300});
copy_coordinate(path->lower_bound_block[2], (float[]){1000,50});
copy_coordinate(path->upper_bound_block[2], (float[]){1400,500});
copy_coordinate(path->lower_bound_block[3], (float[]){1400,200});
copy_coordinate(path->upper_bound_block[3], (float[]){1800,700});
copy_coordinate(path->lower_bound_block[4], (float[]){1100,700});
copy_coordinate(path->upper_bound_block[4], (float[]){1700,1000});
copy_coordinate(path->lower_bound_block[5], (float[]){800,600});
copy_coordinate(path->upper_bound_block[5], (float[]){1100,975});
copy_coordinate(path->lower_bound_block[6], (float[]){100,700});
copy_coordinate(path->upper_bound_block[6], (float[]){800,975});
*/
#else
copy_coordinate(path->lower_bound_block[0], (float[]){0,3});
copy_coordinate(path->upper_bound_block[0], (float[]){4,7});
copy_coordinate(path->lower_bound_block[1], (float[]){1,0});
copy_coordinate(path->upper_bound_block[1], (float[]){10,3});
copy_coordinate(path->lower_bound_block[2], (float[]){10,0.5});
copy_coordinate(path->upper_bound_block[2], (float[]){14,5});
copy_coordinate(path->lower_bound_block[3], (float[]){14,2});
copy_coordinate(path->upper_bound_block[3], (float[]){18,7});
copy_coordinate(path->lower_bound_block[4], (float[]){11,7});
copy_coordinate(path->upper_bound_block[4], (float[]){17,10});
copy_coordinate(path->lower_bound_block[5], (float[]){8,6});
copy_coordinate(path->upper_bound_block[5], (float[]){11,9.75});
copy_coordinate(path->lower_bound_block[6], (float[]){1,7});
copy_coordinate(path->upper_bound_block[6], (float[]){8,9.75});
#endif
update_bounds_limits_blocks(path);
struct vehicle *car = create_vehicle(path);
config_layers *pconf = create_config_layers_from_OneD(4,(size_t[]){3,24,24,3}); /* 3 input , 3 target; 2 hidden layer with 24 neurons each */
//config_layers *pconf = create_config_layers_from_OneD(4,(size_t[]){3,14,14,3}); /* 3 input , 3 target; 2 hidden layer with 24 neurons each */
bool randomize=true;
float minR = -0.5, maxR = 0.5;
int randomRange = 500;
size_t nb_prod_thread = 2;
size_t nb_calc_thread = 4;
float learning_rate = 0.00001 /* 0.001*/;
struct networks_qlearning *nnetworks = create_nework_qlearning(
pconf,
randomize, minR, maxR, randomRange,
nb_prod_thread, nb_calc_thread,
learning_rate
);
/*
EXTRACT_FILE_TO_TENSOR_ATTRIBUTE_NNEURONS(TYPE_FLOAT, nnetworks->main_net, weight_in, ".ff_main_20240717_01h42m16s_5300.txt");
EXTRACT_FILE_TO_TENSOR_ATTRIBUTE_NNEURONS(TYPE_FLOAT, nnetworks->target_net, weight_in, ".ff_target_20240717_01h42m16s_5300.txt");
*/
EXTRACT_FILE_TO_TENSOR_ATTRIBUTE_NNEURONS(TYPE_FLOAT, nnetworks->main_net, weight_in, ".ff_main_20240717_09h11m09s_1700.txt");
EXTRACT_FILE_TO_TENSOR_ATTRIBUTE_NNEURONS(TYPE_FLOAT, nnetworks->target_net, weight_in, ".ff_target_20240717_09h11m09s_1700.txt");
struct status_qlearning *qlstatus = create_status_qlearning ();
struct delay_params *dly = create_delay_params (
500/*size_t delay_between_episodes*/,
50/*size_t delay_between_games*/
);
struct qlearning_params *qlparams = create_qlearning_params (
0.95/*float gamma*/,
learning_rate,
0 /* (not used!)float discount_factor*/,
0.0001/*0.99*/ /*float exploration_factor*/,
20/*long int nb_training_before_update_weight_in_target*/,
10000/*size_t number_episodes*/
);
/* UPDATE_ATTRIBUTE_NEURONE_IN_ALL_LAYERS(TYPE_FLOAT, nnetworks->main_net, d_f_act , df );
UPDATE_ATTRIBUTE_NEURONE_IN_ALL_LAYERS(TYPE_FLOAT, nnetworks->main_net, f_act, f );
UPDATE_ATTRIBUTE_NEURONE_IN_ALL_LAYERS(TYPE_FLOAT, nnetworks->target_net, d_f_act , df );
UPDATE_ATTRIBUTE_NEURONE_IN_ALL_LAYERS(TYPE_FLOAT, nnetworks->target_net, f_act , f );
*/
struct print_params *pprint = create_print_params(
12/*float scale_x*/,12 /*float scale_y*/,
dly/*struct delay_params * dly_p*/
);
struct RL_agent *rlAgent = create_RL_agent (
nnetworks /*struct networks_qlearning * networks*/,
car /*struct vehicle * car*/,
qlstatus /*struct status_qlearning * status*/,
pprint /*struct print_params * pprint*/,
qlparams/*struct qlearning_params *qlearnParams*/
);
learn_to_drive(rlAgent);
free_RL_agent(rlAgent);
}
#endif
#if 1
TEST(first_learn_vehicle_50__11){
size_t nb_block = 10;
size_t dim= 2;
struct blocks * path = create_blocks(nb_block, dim);
#if 1
copy_coordinate(path->lower_bound_block[9], (float[]){0,0});
copy_coordinate(path->upper_bound_block[9], (float[]){100,250});
copy_coordinate(path->lower_bound_block[0], (float[]){100,0});
copy_coordinate(path->upper_bound_block[0], (float[]){250,80});
copy_coordinate(path->lower_bound_block[1], (float[]){250,0});
copy_coordinate(path->upper_bound_block[1], (float[]){360,140});
copy_coordinate(path->lower_bound_block[2], (float[]){360,70});
copy_coordinate(path->upper_bound_block[2], (float[]){600,140});
copy_coordinate(path->lower_bound_block[3], (float[]){600,90});
copy_coordinate(path->upper_bound_block[3], (float[]){720,300});
copy_coordinate(path->lower_bound_block[4], (float[]){300,300});
copy_coordinate(path->upper_bound_block[4], (float[]){720,350});
copy_coordinate(path->lower_bound_block[5], (float[]){300,150});
copy_coordinate(path->upper_bound_block[5], (float[]){410,300});
copy_coordinate(path->lower_bound_block[6], (float[]){120,150});
copy_coordinate(path->upper_bound_block[6], (float[]){300,210});
copy_coordinate(path->lower_bound_block[7], (float[]){120,210});
copy_coordinate(path->upper_bound_block[7], (float[]){270,350});
copy_coordinate(path->lower_bound_block[8], (float[]){0,250});
copy_coordinate(path->upper_bound_block[8], (float[]){120,350});
/*
copy_coordinate(path->lower_bound_block[4], (float[]){0,0});
copy_coordinate(path->upper_bound_block[4], (float[]){150,250});
copy_coordinate(path->lower_bound_block[3], (float[]){150,40});
copy_coordinate(path->upper_bound_block[3], (float[]){250,150});
copy_coordinate(path->lower_bound_block[2], (float[]){250,80});
copy_coordinate(path->upper_bound_block[2], (float[]){360,200});
copy_coordinate(path->lower_bound_block[1], (float[]){360,70});
copy_coordinate(path->upper_bound_block[1], (float[]){600,150});
copy_coordinate(path->lower_bound_block[0], (float[]){600,90});
copy_coordinate(path->upper_bound_block[0], (float[]){760,300});
copy_coordinate(path->lower_bound_block[6], (float[]){260,300});
copy_coordinate(path->upper_bound_block[6], (float[]){760,360});
copy_coordinate(path->lower_bound_block[5], (float[]){0,250});
copy_coordinate(path->upper_bound_block[5], (float[]){410,300});
copy_coordinate(path->lower_bound_block[0], (float[]){0,0});
copy_coordinate(path->upper_bound_block[0], (float[]){150,250});
copy_coordinate(path->lower_bound_block[1], (float[]){150,0});
copy_coordinate(path->upper_bound_block[1], (float[]){250,150});
copy_coordinate(path->lower_bound_block[2], (float[]){250,80});
copy_coordinate(path->upper_bound_block[2], (float[]){360,200});
copy_coordinate(path->lower_bound_block[3], (float[]){360,70});
copy_coordinate(path->upper_bound_block[3], (float[]){600,170});
copy_coordinate(path->lower_bound_block[4], (float[]){600,90});
copy_coordinate(path->upper_bound_block[4], (float[]){760,300});
copy_coordinate(path->lower_bound_block[5], (float[]){300,300});
copy_coordinate(path->upper_bound_block[5], (float[]){760,350});
copy_coordinate(path->lower_bound_block[6], (float[]){0,250});
copy_coordinate(path->upper_bound_block[6], (float[]){410,300});
copy_coordinate(path->lower_bound_block[0], (float[]){0,300});
copy_coordinate(path->upper_bound_block[0], (float[]){400,700});
copy_coordinate(path->lower_bound_block[1], (float[]){100,0});
copy_coordinate(path->upper_bound_block[1], (float[]){1000,300});
copy_coordinate(path->lower_bound_block[2], (float[]){1000,50});
copy_coordinate(path->upper_bound_block[2], (float[]){1400,500});
copy_coordinate(path->lower_bound_block[3], (float[]){1400,200});
copy_coordinate(path->upper_bound_block[3], (float[]){1800,700});
copy_coordinate(path->lower_bound_block[4], (float[]){1100,700});
copy_coordinate(path->upper_bound_block[4], (float[]){1700,1000});
copy_coordinate(path->lower_bound_block[5], (float[]){800,600});
copy_coordinate(path->upper_bound_block[5], (float[]){1100,975});
copy_coordinate(path->lower_bound_block[6], (float[]){100,700});
copy_coordinate(path->upper_bound_block[6], (float[]){800,975});
*/
#else
copy_coordinate(path->lower_bound_block[0], (float[]){0,3});
copy_coordinate(path->upper_bound_block[0], (float[]){4,7});
copy_coordinate(path->lower_bound_block[1], (float[]){1,0});
copy_coordinate(path->upper_bound_block[1], (float[]){10,3});
copy_coordinate(path->lower_bound_block[2], (float[]){10,0.5});
copy_coordinate(path->upper_bound_block[2], (float[]){14,5});
copy_coordinate(path->lower_bound_block[3], (float[]){14,2});
copy_coordinate(path->upper_bound_block[3], (float[]){18,7});
copy_coordinate(path->lower_bound_block[4], (float[]){11,7});
copy_coordinate(path->upper_bound_block[4], (float[]){17,10});
copy_coordinate(path->lower_bound_block[5], (float[]){8,6});
copy_coordinate(path->upper_bound_block[5], (float[]){11,9.75});
copy_coordinate(path->lower_bound_block[6], (float[]){1,7});
copy_coordinate(path->upper_bound_block[6], (float[]){8,9.75});
#endif
update_bounds_limits_blocks(path);
struct vehicle *car = create_vehicle(path);
config_layers *pconf = create_config_layers_from_OneD(4,(size_t[]){3,24,24,3}); /* 3 input , 3 target; 2 hidden layer with 24 neurons each */
//config_layers *pconf = create_config_layers_from_OneD(4,(size_t[]){3,14,14,3}); /* 3 input , 3 target; 2 hidden layer with 24 neurons each */
bool randomize=true;
float minR = -0.5, maxR = 0.5;
int randomRange = 500;
size_t nb_prod_thread = 2;
size_t nb_calc_thread = 4;
float learning_rate = 0.00001 /* 0.001*/;
struct networks_qlearning *nnetworks = create_nework_qlearning(
pconf,
randomize, minR, maxR, randomRange,
nb_prod_thread, nb_calc_thread,
learning_rate
);
//print_vehicle_n_path(car, 12, 12);
EXTRACT_FILE_TO_TENSOR_ATTRIBUTE_NNEURONS(TYPE_FLOAT, nnetworks->main_net, weight_in, ".ff_main_20240717_09h11m09s_1700.txt");
EXTRACT_FILE_TO_TENSOR_ATTRIBUTE_NNEURONS(TYPE_FLOAT, nnetworks->target_net, weight_in, ".ff_target_20240717_09h11m09s_1700.txt");
struct status_qlearning *qlstatus = create_status_qlearning ();
struct delay_params *dly = create_delay_params (
500/*size_t delay_between_episodes*/,
50/*size_t delay_between_games*/
);
struct qlearning_params *qlparams = create_qlearning_params (
0.95/*float gamma*/,
learning_rate,
0 /* (not used!)float discount_factor*/,
0.0001/*0.99*/ /*float exploration_factor*/,
20/*long int nb_training_before_update_weight_in_target*/,
10000/*size_t number_episodes*/
);
/* UPDATE_ATTRIBUTE_NEURONE_IN_ALL_LAYERS(TYPE_FLOAT, nnetworks->main_net, d_f_act , df );
UPDATE_ATTRIBUTE_NEURONE_IN_ALL_LAYERS(TYPE_FLOAT, nnetworks->main_net, f_act, f );
UPDATE_ATTRIBUTE_NEURONE_IN_ALL_LAYERS(TYPE_FLOAT, nnetworks->target_net, d_f_act , df );
UPDATE_ATTRIBUTE_NEURONE_IN_ALL_LAYERS(TYPE_FLOAT, nnetworks->target_net, f_act , f );
*/
struct print_params *pprint = create_print_params(
12/*float scale_x*/,12 /*float scale_y*/,
dly/*struct delay_params * dly_p*/
);
struct RL_agent *rlAgent = create_RL_agent (
nnetworks /*struct networks_qlearning * networks*/,
car /*struct vehicle * car*/,
qlstatus /*struct status_qlearning * status*/,
pprint /*struct print_params * pprint*/,
qlparams/*struct qlearning_params *qlearnParams*/
);
learn_to_drive(rlAgent);
free_RL_agent(rlAgent);
}
#endif
#if 1
TEST(first_learn_vehicle){
size_t nb_block = 7;