Eigen::VectorXd gaussianProposal(uint id, const Eigen::VectorXd& sample, double sigma)
{
// Random number generators
static std::random_device rd;
static std::mt19937 generator(rd());
static std::normal_distribution<> rand; // Standard normal
// Vary each paramater according to a Gaussian distribution
Eigen::VectorXd proposal(sample.rows());
for (int i = 0; i < proposal.rows(); i++)
proposal(i) = sample(i) + rand(generator) * sigma;
return proposal;
}
std::vector<double> Metropolis::propose_jump(const std::vector<double>& coordinates) const {
assert(coordinates.size() == this->jump_sigma.size());
std::vector<double> proposal(coordinates.size());
for (size_t i=0; i<coordinates.size(); i++) {
proposal[i] = gRandom->Gaus(coordinates[i], this->jump_sigma[i]);
}
return proposal;
}
int main(int argc, char** argv) {
QApplication app(argc, argv);
std::vector<Eigen::Matrix<double, 2, 1> > data;
NormalDistribution<2> dist(Eigen::Matrix<double, 2, 1>(1.0, 1.0),
(Eigen::Matrix<double, 2, 2>() << 2, 0, 0, 2).finished());
NormalDistribution<2> proposal(Eigen::Matrix<double, 2, 1>(10.0, 10.0),
Eigen::Matrix<double, 2, 2>::Identity() * 0.5);
MetropolisHastingsSampler::getSamples<double, double, 2>(dist, proposal, data,
10000);
ScatterPlot<2> plot("NormalDistribution2vMH", data);
plot.show();
return app.exec();
}
void mcmc(tree *tr, analdef *adef)
{
int i=0;
tr->startLH = tr->likelihood;
printBothOpen("start minimalistic search with LH %f\n", tr->likelihood);
printBothOpen("tr LH %f, startLH %f\n", tr->likelihood, tr->startLH);
int insert_id;
int j;
int maxradius = 30;
int accepted_spr = 0, accepted_nni = 0, accepted_bl = 0, accepted_model = 0, accepted_gamma = 0, inserts = 0;
int rejected_spr = 0, rejected_nni = 0, rejected_bl = 0, rejected_model = 0, rejected_gamma = 0;
int num_moves = 10000;
boolean proposalAccepted;
boolean proposalSuccess;
prop which_proposal;
double testr;
double acceptance;
srand (440);
double totalTime = 0.0, proposalTime = 0.0, blTime = 0.0, printTime = 0.0;
double t_start = gettime();
double t;
//allocate states
double bl_prior_exp_lambda = 0.1;
double bl_sliding_window_w = 0.005;
double gm_sliding_window_w = 0.75;
double rt_sliding_window_w = 0.5;
state *curstate = state_init(tr, adef, maxradius, bl_sliding_window_w, rt_sliding_window_w, gm_sliding_window_w, bl_prior_exp_lambda);
printStateFileHeader(curstate);
set_start_bl(curstate);
printf("start bl_prior: %f\n",curstate->bl_prior);
set_start_prior(curstate);
curstate->hastings = 1;//needs to be set by the proposal when necessary
/* Set the starting LH with a full traversal */
evaluateGeneric(tr, tr->start, TRUE);
tr->startLH = tr->likelihood;
printBothOpen("Starting with tr LH %f, startLH %f\n", j, tr->likelihood, tr->startLH);
/* Set reasonable model parameters */
evaluateGeneric(curstate->tr, curstate->tr->start, FALSE); // just for validation
printBothOpen("tr LH before modOpt %f\n",curstate->tr->likelihood);
printSubsRates(curstate->tr, curstate->model, curstate->numSubsRates);
/* optimize the model with Brents method for reasonable starting points */
modOpt(curstate->tr, curstate->adef, 5.0); /* not by proposal, just using std raxml machinery... */
evaluateGeneric(curstate->tr, curstate->tr->start, FALSE); // just for validation
printBothOpen("tr LH after modOpt %f\n",curstate->tr->likelihood);
printSubsRates(curstate->tr, curstate->model, curstate->numSubsRates);
recordSubsRates(curstate->tr, curstate->model, curstate->numSubsRates, curstate->curSubsRates);
int first = 1;
/* beginning of the MCMC chain */
for(j=0; j<num_moves; j++)
{
//printBothOpen("iter %d, tr LH %f, startLH %f\n",j, tr->likelihood, tr->startLH);
//printRecomTree(tr, TRUE, "startiter");
proposalAccepted = FALSE;
t = gettime();
/*
evaluateGeneric(tr, tr->start); // just for validation
printBothOpen("before proposal, iter %d tr LH %f, startLH %f\n", j, tr->likelihood, tr->startLH);
*/
which_proposal = proposal(curstate);
if (first == 1)
{
first = 0;
curstate->curprior = curstate->newprior;
}
//printBothOpen("proposal done, iter %d tr LH %f, startLH %f\n", j, tr->likelihood, tr->startLH);
assert(which_proposal == SPR || which_proposal == stNNI ||
which_proposal == UPDATE_ALL_BL ||
which_proposal == UPDATE_MODEL || which_proposal == UPDATE_GAMMA);
proposalTime += gettime() - t;
/* decide upon acceptance */
testr = (double)rand()/(double)RAND_MAX;
//should look something like
acceptance = fmin(1,(curstate->hastings) *
(exp(curstate->newprior-curstate->curprior)) * (exp(curstate->tr->likelihood-curstate->tr->startLH)));
/*
//printRecomTree(tr, FALSE, "after proposal");
printBothOpen("after proposal, iter %d tr LH %f, startLH %f\n", j, tr->likelihood, tr->startLH);
*/
if(testr < acceptance)
{
proposalAccepted = TRUE;
switch(which_proposal)
{
case SPR:
//printRecomTree(tr, TRUE, "after accepted");
// printBothOpen("SPR new topology , iter %d tr LH %f, startLH %f\n", j, tr->likelihood, tr->startLH);
accepted_spr++;
break;
case stNNI:
printBothOpen("NNI new topology , iter %d tr LH %f, startLH %f\n", j, tr->likelihood, tr->startLH);
accepted_nni++;
break;
case UPDATE_ALL_BL:
// printBothOpen("BL new , iter %d tr LH %f, startLH %f\n", j, tr->likelihood, tr->startLH);
accepted_bl++;
break;
case UPDATE_MODEL:
// printBothOpen("Model new, iter %d tr LH %f, startLH %f\n", j, tr->likelihood, tr->startLH);
accepted_model++;
break;
case UPDATE_GAMMA:
// printBothOpen("Gamma new, iter %d tr LH %f, startLH %f\n", j, tr->likelihood, tr->startLH);
accepted_gamma++;
break;
default:
assert(0);
}
curstate->tr->startLH = curstate->tr->likelihood; //new LH
curstate->curprior = curstate->newprior;
}
else
{
//printBothOpen("rejected , iter %d tr LH %f, startLH %f, %i \n", j, tr->likelihood, tr->startLH, which_proposal);
resetState(which_proposal,curstate);
switch(which_proposal)
{
case SPR:
rejected_spr++;
break;
case stNNI:
rejected_nni++;
break;
case UPDATE_ALL_BL:
rejected_bl++;
break;
case UPDATE_MODEL:
rejected_model++;
break;
case UPDATE_GAMMA:
rejected_gamma++;
break;
default:
assert(0);
}
evaluateGeneric(tr, tr->start, FALSE);
// just for validation
if(fabs(curstate->tr->startLH - tr->likelihood) > 1.0E-10)
{
printBothOpen("WARNING: LH diff %.10f\n", curstate->tr->startLH - tr->likelihood);
}
//printRecomTree(tr, TRUE, "after reset");
//printBothOpen("after reset, iter %d tr LH %f, startLH %f\n", j, tr->likelihood, tr->startLH);
assert(fabs(curstate->tr->startLH - tr->likelihood) < 1.0E-10);
}
inserts++;
/* need to print status */
if (j % 50 == 0)
{
t = gettime();
printBothOpen("sampled at iter %d, tr LH %f, startLH %f, prior %f, incr %f\n",j, tr->likelihood, tr->startLH, curstate->curprior, tr->likelihood - tr->startLH);
boolean printBranchLengths = TRUE;
/*printSimpleTree(tr, printBranchLengths, adef);*/
//TODO: print some parameters to a file
printStateFile(j,curstate);
printTime += gettime() - t;
}
}
t = gettime();
treeEvaluate(tr, 1);
blTime += gettime() - t;
printBothOpen("accepted SPR %d, accepted stNNI %d, accepted BL %d, accepted model %d, accepted gamma %d, num moves tried %d, SPRs with max radius %d\n",
accepted_spr, accepted_nni, accepted_bl, accepted_model, accepted_gamma, num_moves, maxradius);
printBothOpen("rejected SPR %d, rejected stNNI %d, rejected BL %d, rejected model %d, rejected gamma %d\n",
rejected_spr, rejected_nni, rejected_bl, rejected_model, rejected_gamma);
printBothOpen("ratio SPR %f, ratio stNNI %f, ratio BL %f, ratio model %f, ratio gamma %f\n",
accepted_spr/(double)(rejected_spr+accepted_spr), accepted_nni/(double)(rejected_nni+accepted_nni), accepted_bl/(double)(rejected_bl+accepted_bl),
accepted_model/(double)(rejected_model+accepted_model), accepted_gamma/(double)(rejected_gamma+accepted_gamma));
printBothOpen("total %f, BL %f, printing %f, proposal %f\n", gettime()- t_start, blTime, printTime, proposalTime);
assert(inserts == num_moves);
state_free(curstate);
}