/**
* Compute the histogram of distances between sample points on the shape. The histogram bins uniformly subdivide the range
* of distances from zero to \a max_distance. If \a max_distance is negative, the shape scale specified in the constructor
* will be used.
*
* @param histogram The histogram to be computed.
* @param dist_type The type of distance metric.
* @param max_distance The maximum separation between points to consider for the histogram. A negative value indicates the
* entire shape is to be considered (in which case \a max_distance is set to the shape scale specified in the
* constructor). The histogram range is set appropriately.
* @param pair_reduction_ratio The fraction of the available set of point pairs -- expressed as a number between 0 and 1 --
* that will be randomly selected and used to actually build the histogram. Note that the final set is a subsampling of
* the set of all possible pairs, not the set of all possible pairs of a subsampled set of points. This may be useful for
* getting a more evenly sampled set of pairwise distances, with an extra-large initial set of points. A value of 1
* indicates all ordered pairs will be used, but this counts every distance twice, so a value of 0.5 or so may be more
* appropriate. A negative value picks a default of 0.5, or the ratio that gives a maximum of ~1M ordered pairs, whichever
* is smaller.
*/
void compute(Histogram & histogram, DistanceType dist_type, Real max_distance = -1, Real pair_reduction_ratio = -1) const
{
long num_samples = ldh.numSamples();
long num_distinct_unordered = (num_samples - 1) * num_samples;
if (pair_reduction_ratio < 0)
pair_reduction_ratio = (Real)std::min(0.5, 1000000.0 / num_distinct_unordered); // don't count (x, x)
histogram.setZero(); // don't bother setting the range
Real local_reduction_ratio = std::sqrt(pair_reduction_ratio);
long num_queries = Math::clamp((long)std::ceil(local_reduction_ratio * num_samples), 0, num_samples - 1);
if (num_queries <= 0)
return;
TheaArray<int32> query_indices((array_size_t)num_queries);
Random::common().sortedIntegers(0, (int32)num_samples - 1, (int32)num_queries, &query_indices[0]);
Histogram local_histogram(histogram.numBins());
for (array_size_t i = 0; i < query_indices.size(); ++i)
{
Vector3 p = ldh.getSamplePosition(query_indices[i]);
ldh.compute(p, local_histogram, dist_type, max_distance, local_reduction_ratio);
// Remove the zero distance from the query point to itself
local_histogram.remove(0.0);
if (i == 0)
histogram.setRange(local_histogram.minValue(), local_histogram.maxValue());
histogram.insert(local_histogram);
}
}
virtual bool operator()(bool enabled, const Args &args) {
ncalls++;
std::pair<Histogram::iterator, bool> inserted = calls.insert(std::make_pair<int, size_t>(args.callno, 1));
if (!inserted.second)
inserted.first->second++;
return enabled;
}
virtual bool operator()(bool enabled, const Args &args) {
SgAsmx86Instruction *insn = isSgAsmx86Instruction(args.insn);
assert(insn);
ninsns++;
std::string kind = rose::stringifyX86InstructionKind(insn->get_kind(), "x86_");
std::pair<Histogram::iterator, bool> inserted = insns.insert(std::make_pair<std::string, size_t>(kind, 1));
if (!inserted.second)
inserted.first->second++;
return enabled;
}
/** Return the most likely distance between two contigs and the number
* of pairs that support that distance estimate.
* @param len0 the length of the first contig in bp
* @param len1 the length of the second contig in bp
* @param rf whether the fragment library is oriented reverse-forward
* @param[out] n the number of samples with a non-zero probability
*/
int maximumLikelihoodEstimate(unsigned l,
int first, int last,
const vector<int>& samples, const PMF& pmf,
unsigned len0, unsigned len1, bool rf,
unsigned& n)
{
assert(first < last);
assert(!samples.empty());
// The aligner is unable to map reads to the ends of the sequence.
// Correct for this lack of sensitivity by subtracting l-1 bp from
// the length of each sequence, where the aligner requires a match
// of at least l bp. When the fragment library is oriented
// forward-reverse, subtract 2*(l-1) from each sample.
assert(l > 0);
assert(len0 >= l);
assert(len1 >= l);
len0 -= l - 1;
len1 -= l - 1;
if (len0 > len1)
swap(len0, len1);
if (rf) {
// This library is oriented reverse-forward.
Histogram h(samples.begin(), samples.end());
int d;
tie(d, n) = maximumLikelihoodEstimate(
first, last, h,
pmf, len0, len1);
return d;
} else {
// This library is oriented forward-reverse.
// Subtract 2*(l-1) from each sample.
Histogram h;
typedef vector<int> Samples;
for (Samples::const_iterator it = samples.begin();
it != samples.end(); ++it) {
assert(*it > 2 * (int)(l - 1));
h.insert(*it - 2 * (l - 1));
}
int d;
tie(d, n) = maximumLikelihoodEstimate(
first, last, h,
pmf, len0, len1);
return max(first, d - 2 * (int)(l - 1));
}
}
static void handleAlignmentPair(const ReadAlignMap::value_type& curr,
const ReadAlignMap::value_type& pair)
{
const string& currID = curr.first;
const string& pairID = pair.first;
// Both reads must align to a unique location.
// The reads are allowed to span more than one contig, but
// at least one of the two reads must span no more than
// two contigs.
const unsigned MAX_SPAN = 2;
if (curr.second.empty() && pair.second.empty()) {
stats.bothUnaligned++;
} else if (curr.second.empty() || pair.second.empty()) {
stats.oneUnaligned++;
} else if (!checkUniqueAlignments(curr.second)
|| !checkUniqueAlignments(pair.second)) {
stats.numMulti++;
} else if (curr.second.size() > MAX_SPAN
&& pair.second.size() > MAX_SPAN) {
stats.numSplit++;
} else {
// Iterate over the vectors, outputting the aligments
bool counted = false;
for (AlignmentVector::const_iterator refAlignIter
= curr.second.begin();
refAlignIter != curr.second.end(); ++refAlignIter) {
for (AlignmentVector::const_iterator pairAlignIter
= pair.second.begin();
pairAlignIter != pair.second.end();
++pairAlignIter) {
const Alignment& a0 = flipAlignment(*refAlignIter,
currID);
const Alignment& a1 = flipAlignment(*pairAlignIter,
pairID);
bool sameTarget = a0.contig == a1.contig;
if (sameTarget
&& curr.second.size() == 1
&& pair.second.size() == 1) {
// Same target and the only alignment.
if (a0.isRC != a1.isRC) {
// Correctly oriented. Add this alignment to
// the distribution of fragment sizes.
int size = fragmentSize(a0, a1);
histogram.insert(size);
if (!opt::fragPath.empty()) {
fragFile << size << '\n';
assert(fragFile.good());
}
} else
stats.numFF++;
counted = true;
}
bool outputSameTarget = opt::fragPath.empty()
&& opt::histPath.empty();
if (!sameTarget || outputSameTarget) {
cout << SAMRecord(a0, a1) << '\n'
<< SAMRecord(a1, a0) << '\n';
assert(cout.good());
}
}
}
if (!counted)
stats.numDifferent++;
}
}
