Skip to content

[6.4][ML] Correct the multimodal prior confidence interval calculation #178

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Aug 2, 2018
Merged

[6.4][ML] Correct the multimodal prior confidence interval calculation #178

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 3 additions & 0 deletions docs/CHANGELOG.asciidoc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -50,6 +50,9 @@ Fix corner case failing to calculate lgamma values and the correspoinding log er
Influence count per bucket for metric population analyses was wrong and lead to wrong influencer scoring ({pull}#150[#150])
Fix a possible SIGSEGV for jobs with multivariate by fields enabled which would lead to the job failing ({pull}#170[#170])

Correct the model bounds and typical value calculation for time series models which use a multimodal distribution.
This issue could cause "Unable to bracket left percentile =..." errors to appear in the logs. ({pull}#176[#176])

=== Regressions

=== Known Issues
Expand Down
4 changes: 2 additions & 2 deletions include/maths/CMultimodalPriorUtils.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -206,8 +206,8 @@ class MATHS_EXPORT CMultimodalPriorUtils : private core::CNonInstantiatable {
return support;
}

double p1 = std::log((1.0 - percentage) / 2.0);
double p2 = std::log((1.0 + percentage) / 2.0);
double p1 = maths_t::count(weights) * std::log((1.0 - percentage) / 2.0);
double p2 = maths_t::count(weights) * std::log((1.0 + percentage) / 2.0);

CLogCdf<PRIOR> fl(CLogCdf<PRIOR>::E_Lower, prior, weights);
CLogCdf<PRIOR> fu(CLogCdf<PRIOR>::E_Upper, prior, weights);
Expand Down
31 changes: 29 additions & 2 deletions lib/maths/unittest/CMultimodalPriorTest.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -941,10 +941,10 @@ void CMultimodalPriorTest::testMarginalLikelihoodConfidenceInterval() {

using TMeanAccumulator = maths::CBasicStatistics::SSampleMean<double>::TAccumulator;

test::CRandomNumbers rng;

LOG_DEBUG(<< "Synthetic");
{
test::CRandomNumbers rng;

double w1 = 0.2;
double location1 = 0.1;
double squareScale1 = 0.2;
Expand Down Expand Up @@ -1047,6 +1047,33 @@ void CMultimodalPriorTest::testMarginalLikelihoodConfidenceInterval() {
CPPUNIT_ASSERT_DOUBLES_EQUAL(-111.0, i90.first, 0.5);
CPPUNIT_ASSERT_DOUBLES_EQUAL(158952.0, i90.second, 0.5);
}

LOG_DEBUG(<< "Non-unit count weight");
{
// The confidence interval should be independent of the sample
// count weight.

TDoubleVec modes[2];
rng.generateNormalSamples(10.0, 2.0, 50, modes[0]);
rng.generateNormalSamples(20.0, 2.0, 50, modes[1]);
TDoubleVec samples(modes[0].begin(), modes[0].end());
samples.insert(samples.end(), modes[1].begin(), modes[1].end());

CMultimodalPrior filter(makePrior());
filter.addSamples(samples);

CPPUNIT_ASSERT_EQUAL(std::size_t(2), filter.numberModes());

TDoubleDoublePr interval{filter.marginalLikelihoodConfidenceInterval(
90.0, maths_t::countWeight(1.0))};
TDoubleDoublePr weightedInterval{filter.marginalLikelihoodConfidenceInterval(
90.0, maths_t::countWeight(0.3))};
LOG_DEBUG(<< "interval = " << core::CContainerPrinter::print(interval));
LOG_DEBUG(<< "weightedInterval = "
<< core::CContainerPrinter::print(weightedInterval));
CPPUNIT_ASSERT_EQUAL(core::CContainerPrinter::print(interval),
core::CContainerPrinter::print(weightedInterval));
}
}

void CMultimodalPriorTest::testSampleMarginalLikelihood() {
Expand Down