def _computeBinomialPval(cls, x, size, prob, tail):
from proto.RSetup import r
x, size, prob = int(x), int(size), float(prob)
if prob * size >= cls.MIN_SUCCESSES_FOR_NORM_APPROXIMATION <= (
1 - prob) * size:
mean = size * prob
sd = (size * prob * (1 - prob))**0.5
lessPval = r.pnorm(x, mean, sd)
if tail == 'less':
pval = lessPval
elif tail == 'more':
pval = 1 - lessPval
elif tail == 'different':
pval = min(1, 2 * min(lessPval, 1 - lessPval))
else:
from gold.application.LogSetup import logMessage, logging
logMessage(
'Unsupported tail (%s) encountered in _computeBinomialTail.'
% tail,
level=logging.WARN)
elif x > cls.MAX_SUCCESSES_FOR_BINOMIAL_RUNTIME:
return None
#raise NotImplementedError()
else:
if tail == 'less':
pval = r.pbinom(x, size, prob)
elif tail == 'more':
pval = 1 - r.pbinom(x - 1, size, prob)
elif tail == 'different':
pval = min(
1, 2 * min(r.pbinom(x, size, prob),
1 - r.pbinom(x - 1, size, prob)))
return pval
def _calcBinomialPVals(self, countMatrix, colSums, rowSums, N):
from proto.RSetup import r
pValMatrix = numpy.zeros(countMatrix.shape)
for i in xrange(pValMatrix.shape[0]):
for j in xrange(pValMatrix.shape[1]):
q = x = int(countMatrix[i, j])
size = int(colSums[j])
prob = float(rowSums[i]) / N
# print q,size,prob
rawPval = min(
r.pbinom(q, size, prob, True),
r.pbinom(q, size, prob, False) + r.dbinom(q, size, prob))
# rawPval = min( r.pbinom(q, size, prob, True) - 0.5 * r.dbinom(q, size, prob), \
# r.pbinom(q, size, prob, False) + 0.5 * r.dbinom(q, size, prob) )
pValMatrix[i, j] = min(rawPval * 2, 1) #correct for two-tail..
return pValMatrix
def _compute(self):
Ois = self._countOfSegmentsInBin.getResult()
globalOis = self._countOfSegmentsAcrossBins.getResult()
binCoverageProportion = self._proportionOfAnalysisRegionInBin.getResult(
)
#print self._region
#print "<p>**** %d, %d, %f</p>" % (Ois, globalOis, binCoverageProportion)
#Eis =float(globalOis)*binCoverageProportion
from proto.RSetup import r
return 1.0 - r.pbinom(
Ois, globalOis, binCoverageProportion
) #handle tails.. Do some trick to send in lower.tail..
def _compute(self):
from proto.RSetup import r
#r("require(Defaults)")
#r('setDefaults(q, save="no")')
#r("useDefaults(q)")
x = self._numPointsInside.getResult()
size = self._numPointsTotal.getResult()
prob = self._segmentCoverProportion.getResult()
se = math.sqrt(1.0 * (prob) * (1 - prob) / size)
if size < 1 or prob in [0, 1]:
return None
if self._tail == 'less':
pval = r.pbinom(x, size, prob)
elif self._tail == 'more':
pval = 1 - r.pbinom(x - 1, size, prob)
elif self._tail == 'different':
pval = min(
1, 2 *
min(r.pbinom(x, size, prob), 1 - r.pbinom(x - 1, size, prob)))
#return {'P-value':pval, 'SegCover':prob, 'PointsInside':x, 'PointsTotal':size}
return OrderedDict([ ('P-value', float(pval)), ('Test statistic: PointsInside', x), ('E(Test statistic): ExpPointsInside', prob*size), \
('DiffFromExpected', x-prob*size), ('PointsTotal', size), ('SegCoverage', prob) ])