std::vector<Creature> Creature::Crossover(Creature mate){
std::vector<Creature> children;
std::vector<Brain> childrens_brain;
children.resize(2);
childrens_brain.resize(2);
std::uniform_real_distribution<float> int_dist(0.0f,1.0f);
double should_crossover = int_dist(rng_.mt_rng_);
SettingsManager::Instance()->GetCrossover();
if(SettingsManager::Instance()->GetCrossover() >= should_crossover) {
// TODO: if crossover should be done, else return the incomming creatures
childrens_brain = brain_.Crossover(mate.GetBrain());
children[0].brain_ = childrens_brain[0];
children[1].brain_ = childrens_brain[1];
}
else {
children[0].brain_ = brain_;
children[1].brain_ = mate.GetBrain();
}
return children;
}
// An HSA version of C++AMP program
int main ()
{
// define inputs and output
const int vecSize = 2048;
int table_a[vecSize];
int table_b[vecSize];
int table_c[vecSize];
int *p_a = &table_a[0];
int *p_b = &table_b[0];
int *p_c = &table_c[0];
// initialize test data
std::random_device rd;
std::uniform_int_distribution<int32_t> int_dist;
for (int i = 0; i < vecSize; ++i) {
table_a[i] = int_dist(rd);
table_b[i] = int_dist(rd);
}
// launch kernel
std::cout << "async launch the 1st kernel\n";
Concurrency::extent<1> e(vecSize);
Concurrency::completion_future fut = Concurrency::async_parallel_for_each(
e,
[=](Concurrency::index<1> idx) restrict(amp) {
p_c[idx[0]] = p_a[idx[0]] + p_b[idx[0]];
});
// An example which shows how to use accelerator_view::create_marker()
// and how to use tick function
bool test() {
bool ret = true;
// define inputs and output
const int vecSize = 2048;
hc::array_view<int, 1> table_a(vecSize);
hc::array_view<int, 1> table_b(vecSize);
hc::array_view<int, 1> table_c(vecSize);
// initialize test data
std::random_device rd;
std::uniform_int_distribution<int32_t> int_dist;
for (int i = 0; i < vecSize; ++i) {
table_a[i] = int_dist(rd);
table_b[i] = int_dist(rd);
}
// launch kernel
hc::extent<1> e(vecSize);
hc::completion_future fut = hc::parallel_for_each(
e,
[=](hc::index<1> idx) __HC__ {
for (int i = 0; i < LOOP_COUNT; ++i)
table_c(idx) = table_a(idx) + table_b(idx);
});
// An HSA version of C++AMP program
int main ()
{
// define inputs and output
const int vecSize = 2048;
int table_a[vecSize];
int table_b[vecSize];
int table_c[vecSize];
int *p_a = &table_a[0];
int *p_b = &table_b[0];
int *p_c = &table_c[0];
// initialize test data
std::random_device rd;
std::uniform_int_distribution<int32_t> int_dist;
for (int i = 0; i < vecSize; ++i) {
table_a[i] = int_dist(rd);
table_b[i] = int_dist(rd);
}
// launch kernel
Concurrency::extent<1> e(vecSize);
Concurrency::completion_future fut = Concurrency::async_parallel_for_each(
e.tile<256>(),
[=](Concurrency::tiled_index<256> idx) restrict(amp) {
int fidx = idx.global[0];
for (int i = 0; i < 1024 * 1024; ++i)
p_c[fidx] = p_a[fidx] + p_b[fidx];
});
uint LogDiscreteSampler::sample(mt19937 & prng) {
if (has_finite) {
double random01_log_sample_scaled = log(generate_canonical<double,std::numeric_limits<double>::digits>(prng)) + cum_probs.back();
return search(random01_log_sample_scaled);
} else {
std::uniform_int_distribution<int> int_dist(0, cum_probs.size() - 1);
return int_dist(prng);
}
}
SchemeObject* random(SchemeObject* args, SchemeObjectCreator* creator) {
SchemeObject* top = args->car();
std::default_random_engine generator;
if (top->is_flonum()) {
double value = top->flonum_value();
std::uniform_real_distribution<double> float_dist(0.0, value);
double result = float_dist(generator);
return creator->make_flonum(result);
} else if (top->is_fixnum()) {
long value = top->fixnum_value();
std::uniform_int_distribution<long> int_dist(0, value);
long result = int_dist(generator);
return creator->make_fixnum(result);
}
}
int gen_int() { return int_dist(gen); }
void generate (SNPMatrix& A,
std::vector<double>& y,
const snp_int_t& num_rhs,
SNPSet& true_snps) {
typedef typename SNPSet::value_type snp_type;
/* psuedo-random generator engine for various phases in this function */
boost::random::mt19937 prng;
prng.seed (int_params[RAND_SEED_INDEX]);
/* parameters that we will refer to over and over again */
snp_int_t M = int_params[M_INDEX]; /* not making it const for BLAS */
snp_int_t N = int_params[N_INDEX]; /* not making it const for BLAS */
const double cutoff_for_0 = dbl_params[CUTOFF_FOR_0_INDEX];
const double cutoff_for_1 = dbl_params[CUTOFF_FOR_1_INDEX];
double signal_strength = dbl_params[SIGNAL_STRENGTH_INDEX];
/*************************************************************************/
/* generate epistatic and marginal SNPs */
const snp_int_t num_mar_snps = floor (dbl_params [PERCENT_MAR_INDEX] *
static_cast<double>(int_params [NUM_TRUE_SNPS_INDEX]));
const snp_int_t num_epi_snps =
int_params [NUM_TRUE_SNPS_INDEX]-num_mar_snps;
/* initialize a uniform random number generator in the range [0,N) */
boost::uniform_int<snp_int_t> int_dist(0, (N-1));
for (snp_int_t i=0;i<num_mar_snps;++i)
true_snps.insert(snp_type(int_dist(prng)));
/* generate some epistatic snp pairs */
generate_epistatic_t<snp_type,
boost::uniform_int<snp_int_t>,
boost::random::mt19937> gen_epistatic (int_dist,prng);
for (snp_int_t i=0; i<num_epi_snps; ++i) {
snp_type epi_snp = gen_epistatic();
while (true_snps.end()!=true_snps.find(epi_snp)) epi_snp=gen_epistatic();
true_snps.insert(epi_snp);
}
/*************************************************************************/
/* generate A */
boost::uniform_real<double> real_dist (0.0 /*lb*/, 1.0 /*ub*/);
for (snp_int_t col=0; col<N; ++col) for (snp_int_t row=0; row<M; ++row) {
const double next_rnd = real_dist(prng);
const unsigned char next_snp = ((cutoff_for_0>next_rnd)? 0:
((cutoff_for_0<=next_rnd && cutoff_for_1>next_rnd)) ? 1:2);
A.set (row,col,next_snp);
}
/*************************************************************************/
/* generate y */
boost::normal_distribution<double> nrml_dist (0.0/*mean*/, 1.0/*sigma*/);
std::vector<double> multiplier (M);
int truncated_M=static_cast<int> (M);
/* Initialize y to zero for good measure */
y.resize (M*num_rhs, 0.0);
/* Marginal: Set the right positions of x to be signal strength */
typename SNPSet::iterator first = true_snps.begin ();
while (first != true_snps.end()) {
/* Generate the column based on a linear model */
if (first->marginal()) A.interact (first->first(), multiplier.begin());
else A.interact (first->first(), first->second(), multiplier.begin());
daxpy_ (&truncated_M,
&signal_strength,
&(multiplier[0]),
&ONE_STEP,
&(y[0]),
&ONE_STEP);
++first;
}
/* Copy this "y" into all the other "y"'s */
double* y_src = &(y[0]);
for (snp_int_t i=1; i<num_rhs; ++i) {
double* y_dst = y_src + (i*truncated_M);
dcopy_ (&truncated_M, y_src, &ONE_STEP, y_dst, &ONE_STEP);
}
/* Now, for each "y", add a different gaussian noise */
for (size_t i=0; i<y.size(); ++i) y[i]+=nrml_dist(prng);
}
IntegralType operator()(const IntegralType& ULIMIT) {
Distribution int_dist (0, ULIMIT-1);
return int_dist (prng);
}
