void player_exp_gain(struct player *p, s32b amount)
{
p->exp += amount;
if (p->exp < p->max_exp)
p->max_exp += amount / 10;
adjust_level(p, TRUE);
}
void player_exp_lose(struct player *p, s32b amount, bool permanent)
{
if (p->exp < amount)
amount = p->exp;
p->exp -= amount;
if (permanent)
p->max_exp -= amount;
adjust_level(p, TRUE);
}
double estimate_step_size(const mat &Uhat, const mat & gamma_grad,const mat & w, double sG,struct STATUS & status, const struct GRASTA_OPT &options)
{
const double LEVEL_FACTOR = 2;
mat DL_prev_gamma;
int newlevel = 0;
double t = 0.0;
if (fabs(status.step_scale) < 0.0000000000001) {
status.step_scale = options.STEP_SCALE*(1+options.MIN_MU)/sG;
if (!options.QUIET) PRINTF("Estimated step-scale %.2e, sigmoid :[%.2f, %.2f, %.2f]\n",
status.step_scale,options.OMEGA, options.FMAX, options.FMIN);
}
if (options.ADAPTIVE) {
DL_prev_gamma = status.last_gamma - Uhat*(trans(Uhat)*status.last_gamma);
//double grad_ip = trace(status.last_w * (trans(DL_prev_gamma) * gamma_grad) * trans(w));
double grad_ip = trace(status.last_w * (trans(status.last_gamma) * gamma_grad) * trans(w));
//double normalization = norm(DL_prev_gamma* trans(status.last_w), "fro") * norm(gamma_grad * trans(w), "fro");
double normalization = norm(status.last_gamma * trans(status.last_w),"fro") * norm(gamma_grad * trans(w),"fro");
double grad_ip_normalization = 0.0;
if (fabs(normalization) > 0.00001 )
grad_ip_normalization = grad_ip/normalization;
status.last_mu = max(status.last_mu + sigmoid(-grad_ip_normalization, options.OMEGA, options.FMAX, options.FMIN) , options.MIN_MU);
t = status.step_scale * pow((double)LEVEL_FACTOR, (double)(-status.level)) * sG; // (1+status.last_mu);
}
else{
status.last_mu = status.last_mu +1;
t = status.step_scale * sG / (status.last_mu);
}
if (t > options.MAX_STEPSIZE)
t= options.MAX_STEPSIZE;
if (options.ADAPTIVE)
adjust_level(status, options);
return t;
}
