void agent_init(const char* task_spec)
{
/*Struct to hold the parsed task spec*/
taskspec_t *ts=(taskspec_t*)malloc(sizeof(taskspec_t));
int decode_result = decode_taskspec( ts, task_spec );
if(decode_result!=0){
printf("Could not decode task spec, code: %d for task spec: %s\n",decode_result,task_spec);
exit(1);
}
/* Lots of assertions to make sure that we can handle this problem. */
assert(getNumIntObs(ts)==1);
assert(getNumDoubleObs(ts)==0);
assert(isIntObsMax_special(ts,0)==0);
assert(isIntObsMin_special(ts,0)==0);
numStates=getIntObsMax(ts,0)+1;
assert(getNumIntAct(ts)==1);
assert(getNumDoubleAct(ts)==0);
assert(isIntActMax_special(ts,0)==0);
assert(isIntActMin_special(ts,0)==0);
numActions=getIntActMax(ts,0)+1;
free_taskspec_struct(ts); /* Make the taskspec struct a "blank slate" */
free(ts); /* Free the structure itself */
/*Here is where you might allocate storage for parameters (value function or policy, last action, last observation, etc)*/
/*Here you would parse the task spec if you felt like it*/
/*Allocate memory for a one-dimensional integer action using utility functions from RLStruct_util*/
allocateRLStruct(&this_action,1,0,0);
allocateRLStruct(&last_action,1,0,0);
/* That is equivalent to:
this_action.numInts = 1;
this_action.intArray = (int*)calloc(1,sizeof(int));
this_action.numDoubles = 0;
this_action.doubleArray = 0;
this_action.numChars = 0;
this_action.charArray = 0;
*/
/*Allocate memory for a one-dimensional integer observation using utility functions from RLStruct_util*/
last_observation=allocateRLStructPointer(1,0,0);
/*Later we will parse this from the task spec, but for now*/
value_function=(double *)calloc(numActions*numStates,sizeof(double));
}
/*****************************
RL-Glue Methods
*******************************/
const char* env_init(){
char *task_spec_string="VERSION RL-Glue-3.0 PROBLEMTYPE episodic "
"DISCOUNTFACTOR 1 OBSERVATIONS INTS (0 107) "
"ACTIONS INTS (0 3) REWARDS (-100.0 10.0) "
"EXTRA SampleMinesEnvironment(C/C++) by Brian Tanner.";
the_world.numRows = 6;
the_world.numCols = 18;
/* Allocate the observation variable */
allocateRLStruct(&this_observation,1,0,0);
/* That is equivalent to:
this_observation.numInts = 1;
this_observation.intArray = (int*)calloc(1,sizeof(int));
this_observation.numDoubles = 0;
this_observation.doubleArray = 0;
this_observation.numChars = 0;
this_observation.charArray = 0;
*/
/* Setup the reward_observation variable */
this_reward_observation.observation=&this_observation;
this_reward_observation.reward=0;
this_reward_observation.terminal=0;
return task_spec_string;
}
void agent_init(const char* task_spec)
{
srand(time(0));
/*Struct to hold the parsed task spec*/
taskspec_t *ts = (taskspec_t*)malloc(sizeof(taskspec_t));
int decode_result = decode_taskspec(ts, task_spec);
if (decode_result != 0){
printf("Could not decode task spec, code: %d for task spec: %s\n", decode_result, task_spec);
exit(1);
}
possibleActions = getIntActMax(ts,0)-1;
allocateRLStruct(¤tAction, 1, 0, 0);
//load();
}
