List of usage examples for org.apache.commons.collections.buffer CircularFifoBuffer toArray
@SuppressWarnings("unchecked") public <T> T[] toArray(T[] a)
From source file:edu.berkeley.compbio.sequtils.strings.MarkovTreeNode.java
/** * Computes the total log probability of generating the given sequence fragment under the model. This differs from * {@link #totalProbability(byte[])} in that the sequence fragment is not given explicitly but only as metadata. Thus * its probability may be computed from summary statistics that are already available in the given SequenceFragment * rather than from the raw sequence. Also, because these probabilities are typically very small, the result is * returned in log space (indeed implementations will likely compute them in log space). * * @param sequenceFragment the SequenceFragment whose probability is to be computed * @return the natural logarithm of the conditional probability (a double value between 0 and 1, inclusive) *//*from w w w. j av a 2s. c o m*/ public double fragmentLogProbability(final SequenceFragment sequenceFragment, final boolean perSample) throws SequenceSpectrumException { // the RonPSA implementation uses backlinks and so is vastly more efficient. // We can't use backlinks here because they might point to nodes outside of this subtree synchronized (sequenceFragment.getReaderForSynchronizing()) // because of resetting the reader { final SequenceReader in; try { in = sequenceFragment.getResetReader(); } catch (NotEnoughSequenceException e) { throw new SequenceSpectrumRuntimeException(e); } final int requiredPrefixLength = getMaxDepth() - 1; double logprob = 0; final CircularFifoBuffer prefix = new CircularFifoBuffer(requiredPrefixLength); int samples = 0; while (true) { try { final byte c = in.read(); try { // PERF converting array prefix from circularFifoBuffer to byte[] is terribly inefficient final byte[] prefixAsBytes = DSArrayUtils .toPrimitive((Byte[]) prefix.toArray(new Byte[prefix.size()])); // these log probabilities could be cached, e.g. logConditionalProbability(c, prefix) logprob += MathUtils.approximateLog(conditionalProbability(c, prefixAsBytes)); samples++; prefix.add(c); } catch (SequenceSpectrumException e) { // probably just an invalid character logger.debug("Invalid character " + (char) c); // ignore this character as far as the probability is concerned prefix.clear(); } } catch (NotEnoughSequenceException e) { break; } catch (IOException e) { logger.error("Error", e); throw new SequenceSpectrumException(e); } catch (FilterException e) { logger.error("Error", e); throw new SequenceSpectrumException(e); } } if (perSample) { // we have ln(product(p) == sum(ln(p)). // The geometric mean is exp(sum(ln(p))/n), so to get ln(geometric mean) we need only divide by n. logprob /= samples; } return logprob; } }