size_t PhaseEstimator::LoadRegion(int region)
{
if (region_num_reads_[region] == 0) // Nothing to load ?
return 0;
if (region_reads_[region].size() > 0) // Region already loaded?
return 0;
ClockTimer timer;
timer.StartTimer();
region_reads_[region].reserve(region_num_reads_[region]);
int region_x = region % num_regions_x_;
int region_y = region / num_regions_x_;
int begin_x = region_x * region_size_x_;
int begin_y = region_y * region_size_y_;
int end_x = min(begin_x + region_size_x_, chip_size_x_);
int end_y = min(begin_y + region_size_y_, chip_size_y_);
// Mutex needed for wells access, but not needed for region_reads access
pthread_mutex_lock(®ion_loader_mutex_);
wells_->SetChunk(begin_y, end_y-begin_y, begin_x, end_x-begin_x, 0, flow_order_.num_flows());
wells_->ReadWells();
vector<float> well_buffer(flow_order_.num_flows());
for (int y = begin_y; y < end_y; y++) {
for (int x = begin_x; x < end_x; x++) {
if (get_subset(x,y) != train_subset_)
continue;
if (!mask_->Match(x, y, MaskLive))
continue;
if (!mask_->Match(x, y, MaskBead))
continue;
// A little help from friends in BkgModel
if (mask_->Match(x, y, MaskFilteredBadResidual))
continue;
if (mask_->Match(x, y, MaskFilteredBadPPF))
continue;
if (mask_->Match(x, y, MaskFilteredBadKey))
continue;
int cls = 0;
if (!mask_->Match(x, y, MaskLib)) { // Not a library bead?
cls = 1;
if (!mask_->Match(x, y, MaskTF)) // Not a tf bead?
continue;
}
for (int flow = 0; flow < flow_order_.num_flows(); ++flow)
well_buffer[flow] = wells_->At(y,x,flow);
// Sanity check. If there are NaNs in this read, print warning
vector<int> nanflow;
for (int flow = 0; flow < flow_order_.num_flows(); ++flow) {
if (!isnan(well_buffer[flow]))
continue;
well_buffer[flow] = 0;
nanflow.push_back(flow);
}
if(nanflow.size() > 0) {
fprintf(stderr, "ERROR: BaseCaller read NaNs from wells file, x=%d y=%d flow=%d", x, y, nanflow[0]);
for(unsigned int flow=1; flow < nanflow.size(); flow++) {
fprintf(stderr, ",%d", nanflow[flow]);
}
fprintf(stderr, "\n");
fflush(stderr);
}
region_reads_[region].push_back(BasecallerRead());
if (use_pid_norm_) {
region_reads_[region].back().SetDataAndKeyNormalizeNew(&well_buffer[0],
flow_order_.num_flows(), keys_[cls].flows(), keys_[cls].flows_length()-1, false /*true*/);
} else {
region_reads_[region].back().SetDataAndKeyNormalize(&well_buffer[0],
flow_order_.num_flows(), keys_[cls].flows(), keys_[cls].flows_length()-1);
}
bool keypass = true;
for (int flow = 0; flow < (keys_[cls].flows_length() - 1); flow++) {
if ((int) (region_reads_[region].back().raw_measurements[flow] + 0.5) != keys_[cls][flow])
keypass = false;
if (isnan(region_reads_[region].back().raw_measurements[flow]))
keypass = false;
}
if (!keypass) {
region_reads_[region].pop_back();
continue;
}
}
}
pthread_mutex_unlock(®ion_loader_mutex_);
region_num_reads_[region] = region_reads_[region].size();
return timer.GetMicroSec();
}
size_t PhaseEstimator::LoadRegion(int region)
{
if (region_num_reads_[region] == 0) // Nothing to load ?
return 0;
if (region_reads_[region].size() > 0) // Region already loaded?
return 0;
ClockTimer timer;
timer.StartTimer();
region_reads_[region].reserve(region_num_reads_[region]);
int region_x = region % num_regions_x_;
int region_y = region / num_regions_x_;
int begin_x = region_x * region_size_x_;
int begin_y = region_y * region_size_y_;
int end_x = min(begin_x + region_size_x_, chip_size_x_);
int end_y = min(begin_y + region_size_y_, chip_size_y_);
// Mutex needed for wells access, but not needed for region_reads access
pthread_mutex_lock(®ion_loader_mutex_);
wells_->SetChunk(begin_y, end_y-begin_y, begin_x, end_x-begin_x, 0, flow_order_.num_flows());
wells_->ReadWells();
vector<float> well_buffer(flow_order_.num_flows());
for (int y = begin_y; y < end_y; y++) {
for (int x = begin_x; x < end_x; x++) {
if (train_subset_count_ > 0 and get_subset(x,y) != train_subset_)
continue;
if (!mask_->Match(x, y, MaskLive))
continue;
if (!mask_->Match(x, y, MaskBead))
continue;
// A little help from friends in BkgModel
if (mask_->Match(x, y, MaskFilteredBadResidual))
continue;
if (mask_->Match(x, y, MaskFilteredBadPPF))
continue;
if (mask_->Match(x, y, MaskFilteredBadKey))
continue;
int cls = 0;
if (!mask_->Match(x, y, MaskLib)) { // Not a library bead?
cls = 1;
if (!mask_->Match(x, y, MaskTF)) // Not a tf bead?
continue;
}
for (int flow = 0; flow < flow_order_.num_flows(); ++flow)
well_buffer[flow] = wells_->At(y,x,flow);
// Sanity check. If there are NaNs in this read, print warning
vector<int> nanflow;
for (int flow = 0; flow < flow_order_.num_flows(); ++flow) {
if (!isnan(well_buffer[flow]))
continue;
well_buffer[flow] = 0;
nanflow.push_back(flow);
}
if(nanflow.size() > 0) {
fprintf(stderr, "ERROR: BaseCaller read NaNs from wells file, x=%d y=%d flow=%d", x, y, nanflow[0]);
for(unsigned int flow=1; flow < nanflow.size(); flow++) {
fprintf(stderr, ",%d", nanflow[flow]);
}
fprintf(stderr, "\n");
fflush(stderr);
}
region_reads_[region].push_back(BasecallerRead());
bool keypass = true;
if (key_norm_method_ == "adaptive") {
keypass = region_reads_[region].back().SetDataAndKeyNormalizeNew(&well_buffer[0],
flow_order_.num_flows(), keys_[cls].flows(), keys_[cls].flows_length()-1, false);
} else if (key_norm_method_ == "off") {
keypass = region_reads_[region].back().SetDataAndKeyPass(well_buffer,
flow_order_.num_flows(), keys_[cls].flows(), keys_[cls].flows_length()-1);
} else {
keypass = region_reads_[region].back().SetDataAndKeyNormalize(&well_buffer[0],
flow_order_.num_flows(), keys_[cls].flows(), keys_[cls].flows_length()-1);
}
// *** Compute some metrics - overload read.penalty_residual to store them
if (keypass) {
unsigned int num_zeromer_flows = 0, num_neg_zeromer_flows = 0;
double squared_dist_int = 0.0;
for (int flow=phasing_start_flow_; flow < phasing_end_flow_; ++flow){
if (region_reads_[region].back().raw_measurements.at(flow) < 0.5) {
++num_zeromer_flows;
if (region_reads_[region].back().raw_measurements.at(flow) < 0.0)
++num_neg_zeromer_flows;
}
if (region_reads_[region].back().raw_measurements.at(flow) < inclusion_threshold_) {
double delta = region_reads_[region].back().raw_measurements.at(flow) -
round(region_reads_[region].back().raw_measurements.at(flow));
squared_dist_int += delta * delta;
}
}
// Too few zero-mers or too much noise? Moving on along, don't waste time on investigating hopeless candidates.
if (num_zeromer_flows < 5 or (float)squared_dist_int > residual_threshold_ + 1.5)
keypass = false;
else {
// [0]=percent_neg_zeromer_flows [1]=squared_dist_int
region_reads_[region].back().penalty_residual.assign(2, 0.0f);
region_reads_[region].back().penalty_residual.at(0) = (float)num_neg_zeromer_flows / (float)num_zeromer_flows;
region_reads_[region].back().penalty_residual.at(1) = squared_dist_int;
}
}
// ***
if (not keypass) {
region_reads_[region].pop_back();
continue;
}
}
}
pthread_mutex_unlock(®ion_loader_mutex_);
region_num_reads_[region] = region_reads_[region].size();
return timer.GetMicroSec();
}
