int main(int argc, char* argv[])
{
std::string filepath("CARTPOLENN");
unsigned int nbrepi = 100;
float gamma_ = 0.5f;
float EOE = 5.0f; //in seconds...
SimulatorRKCARTPOLE env_(EOE);
float lr_ = 5e-2f;//unused...
if( argc>1)
{
lr_ = atof(argv[1]);
}
float eps_ = 0.1f;
int dimActionSpace_ = 1;
int dimStateSpace_ = 4;
QFANN<float> fa_( lr_, eps_, gamma_, dimActionSpace_, filepath);
QLEARNINGXPReplay instance(nbrepi, gamma_, (Environment<float>*)(&env_), (FA<float>*)&fa_);
instance.run(nbrepi);
fa_.save(filepath);
}
DbException::DbException(const DbException &that)
: __DB_STD(exception)()
, what_(dupString(that.what_))
, err_(that.err_)
, env_(0)
{
}
DbException &DbException::operator = (const DbException &that)
{
if (this != &that) {
err_ = that.err_;
delete [] what_;
what_ = dupString(that.what_);
}
return (*this);
}
void DbException::describe(const char *prefix, const char *description)
{
char *msgbuf, *p, *end;
msgbuf = new char[MAX_DESCRIPTION_LENGTH];
p = msgbuf;
end = msgbuf + MAX_DESCRIPTION_LENGTH - 1;
if (prefix != NULL) {
strncpy(p, prefix, (p < end) ? end - p: 0);
p += strlen(prefix);
strncpy(p, ": ", (p < end) ? end - p: 0);
p += 2;
}
if (description != NULL) {
strncpy(p, description, (p < end) ? end - p: 0);
p += strlen(description);
if (err_ != 0) {
strncpy(p, ": ", (p < end) ? end - p: 0);
p += 2;
}
}
if (err_ != 0) {
strncpy(p, db_strerror(err_), (p < end) ? end - p: 0);
p += strlen(db_strerror(err_));
}
/*
* If the result was too long, the buffer will not be null-terminated,
* so we need to fix that here before duplicating it.
*/
if (p >= end)
*end = '\0';
what_ = dupString(msgbuf);
delete [] msgbuf;
}
int main(int argc, char* argv[])
{
std::string filepath("CARTPOLENNACTORCRITIC");
std::string filepathRSFA(filepath+".FA.txt");
std::string filepathRSPA(filepath+".PA.txt");
unsigned int nbrthread = 4;
unsigned int nbrepi = 100;
float gamma_ = 0.99f;
float EOE = 5.0f; //in seconds...
SimulatorRKCARTPOLE env_(EOE);
/*
2x10
float lrPA_ = 1e-2f;
float lrFA_ = 1e-2f;
*/
/*
4x10
*/
/*
float lrPA_ = 1e-2f;
float lrFA_ = 1e-2f;
*/
/*
4x50
*/
float lrPA_ = 1e-4f;
float lrFA_ = 1e-4f;
/**/
float eps_ = 0.9f;
int dimActionSpace_ = 1;
int dimStateSpace_ = 4;
Topology topoFA;
unsigned int nbrneuronsFA = 50;
unsigned int nbrlayerFA = 4;
unsigned int nbrinputFA = dimActionSpace_+dimStateSpace_;
unsigned int nbroutputFA = 1;
topoFA.push_back(nbrinputFA,NTNONE); //input layer
//topoFA.push_back(nbrinputFA,NTSIGMOID); //input layer
//for(int i=nbrlayerFA;i--;) topoFA.push_back(nbrneuronsFA, NTSIGMOID);
for(int i=nbrlayerFA;i--;) topoFA.push_back(nbrneuronsFA, NTTANH);
//topoFA.push_back(nbroutputFA, NTNONE); //linear output
//topoFA.push_back(nbroutputFA, NTSIGMOID); //linear output
topoFA.push_back(nbroutputFA, NTTANH); //linear output
QFANN<float> fa_( lrFA_, eps_, gamma_, dimActionSpace_, topoFA, filepathRSFA);
Topology topoPA;
unsigned int nbrneuronsPA = 50;
unsigned int nbrlayerPA = 4;
unsigned int nbrinputPA = dimStateSpace_;
unsigned int nbroutputPA = dimActionSpace_;
topoPA.push_back(nbrinputPA,NTNONE); //input layer
//topoPA.push_back(nbrinputPA,NTSIGMOID); //input layer
//for(int i=nbrlayerPA;i--;) topoPA.push_back(nbrneuronsPA, NTSIGMOID);
for(int i=nbrlayerPA;i--;) topoPA.push_back(nbrneuronsPA, NTTANH);
//it would be difficult to get to higher values with a nonlinearity that would reduice the range of possibility, maybe...
//topoPA.push_back(nbroutputPA, NTNONE); //linear output
//topoPA.push_back(nbroutputPA, NTSIGMOID); //linear output
topoPA.push_back(nbroutputPA, NTTANH); //linear output
QPANN<float> pa_( lrPA_, eps_, gamma_, dimActionSpace_, topoPA, filepathRSPA);
//QLEARNINGXPReplay instance(nbrepi, gamma_, (Environment<float>*)(&env_), (FA<float>*)&fa_);
//QLEARNINGXPReplayActorCritic instance(nbrepi, gamma_, (Environment<float>*)(&env_), (FA<float>*)&fa_, (PA<float>*)&pa_);
float momentumUpdate = 1e-4f;
int freqUpdate = 1;
DDPGA3C instance(nbrepi, gamma_, (Environment<float>*)(&env_), (FA<float>*)&fa_, (PA<float>*)&pa_, momentumUpdate, freqUpdate);
instance.run(nbrepi,nbrthread);
fa_.save(filepath+"FA");
pa_.save(filepath+"PA");
}
