def test_bootstrap(self):
c_bootstrap = CMeanSD_BOOTSTRAP(dimension=8)
two_bootstrap = CMeanSD_BOOTSTRAP(with_openMP=True, dimension=8)
two_bootstrap.omp_n_threads = 2
four_bootstrap = CMeanSD_BOOTSTRAP(with_openMP=True, dimension=8)
four_bootstrap.omp_n_threads = 4
res_serial = c_bootstrap.run(self.data)
res_two = two_bootstrap.run(self.data)
res_four = four_bootstrap.run(self.data)
actual = self.std
p = percent_error(res_serial, actual)
self.assertTrue((p <= self.threshold).all(),
msg=FAIL_MSG % ('serial', actual, res_serial, p))
p = percent_error(res_two, actual)
self.assertTrue((p <= self.threshold).all(),
msg=FAIL_MSG % ('two threads', actual, res_two, p))
p = percent_error(res_four, actual)
self.assertTrue((p <= self.threshold).all(),
msg=FAIL_MSG % ('four threads', actual, res_four, p))
def test_blb(self):
c_blb = CMeanSD_BLB(dimension=8)
two_blb = CMeanSD_BLB(with_openMP=True, dimension=8)
two_blb.omp_n_threads = 2
four_blb = CMeanSD_BLB(with_openMP=True, dimension=8)
four_blb.omp_n_threads = 4
res_serial = c_blb.run(self.data)
res_two = two_blb.run(self.data)
res_four = four_blb.run(self.data)
actual = norm(self.std)
p = percent_error(res_serial, actual)
self.assertTrue(p <= self.threshold,
msg=FAIL_MSG % ('serial', actual, res_serial, p))
p = percent_error(res_two, actual)
self.assertTrue(p <= self.threshold,
msg=FAIL_MSG % ('two threads', actual, res_two, p))
p = percent_error(res_four, actual)
self.assertTrue(p <= self.threshold,
msg=FAIL_MSG % ('four threads', actual, res_four, p))
numpy.put(results, i * offset, y)
numpy.put(results, xrange(i * offset + 1, (i + 1) * offset), x)
return results, mapping
class LinRegBLB(blb.BLB):
def compute_estimate(self, sample):
linreg()
def reduce_bootstraps(self, sample):
mean()
def average(self, sample):
mean()
if __name__ == '__main__':
data, mapping = create_gaussian_data(3000, 50)
print 'mapping is, ', mapping
cblb = LinRegBLB(dimension=51, num_subsamples=30)
estimated_mapping = cblb.run(data)
assert len(mapping) == len(
estimated_mapping), "Mapping dimension are inequal!"
for i in xrange(len(mapping)):
print 'checking covariate ', i
p = percent_error(mapping[i], estimated_mapping[i])
if p > 0.05:
print 'Error: covariate %d is %f, estimated as %f, percent error %f' % (
i, mapping[i], estimated_mapping[i], p)
print estimated_mapping
print 'done'
#y = (numpy.random.rand() * y_eps) + reduce( float.__add__, map( float.__mul__, x, mapping ) )
y = reduce( float.__add__, x*mapping )
numpy.put( results, i*offset, y )
numpy.put( results, xrange( i*offset+1, (i+1)*offset) , x )
return results, mapping
class LinRegBLB( blb.BLB ):
def compute_estimate( self, sample ):
linreg()
def reduce_bootstraps( self, sample ):
mean()
def average( self, sample ):
mean()
if __name__ == '__main__':
data, mapping = create_gaussian_data( 3000, 50 )
print 'mapping is, ', mapping
cblb = LinRegBLB(dimension=51, num_subsamples=30)
estimated_mapping = cblb.run( data )
assert len(mapping) == len(estimated_mapping), "Mapping dimension are inequal!"
for i in xrange( len( mapping ) ):
print 'checking covariate ', i
p = percent_error( mapping[i], estimated_mapping[i] )
if p > 0.05:
print 'Error: covariate %d is %f, estimated as %f, percent error %f' % (i, mapping[i], estimated_mapping[i], p )
print estimated_mapping
print 'done'
