Improved JointByConditioningDistribution

Now, the compute[Sequential]Conditional[PDF|CDF[Quantile] methods have a dedicated implementation, resulting into a huge performance boost.
regislebrun · Oct 9, 2024 · 5c107ab · 5c107ab
1 parent dfd5ff6
commit 5c107ab
Show file tree

Hide file tree

Showing 4 changed files with 162 additions and 0 deletions.
diff --git a/lib/src/Uncertainty/Distribution/JointByConditioningDistribution.cxx b/lib/src/Uncertainty/Distribution/JointByConditioningDistribution.cxx
@@ -503,6 +503,127 @@ void JointByConditioningDistribution::computeCovariance() const
   isAlreadyComputedCovariance_ = true;
 }
 
+/* Compute the PDF of Xi | X1, ..., Xi-1. x = Xi, y = (X1,...,Xi-1) */
+Scalar JointByConditioningDistribution::computeConditionalPDF(const Scalar x,
+							      const Point & y) const
+{
+  const UnsignedInteger conditioningDimension = y.getDimension();
+  if (conditioningDimension >= getDimension()) throw InvalidArgumentException(HERE) << "Error: cannot compute a conditional PDF with a conditioning point of dimension greater or equal to the distribution dimension.";
+  // Special case for a conditioning only in the conditioning part
+  const UnsignedInteger conditioningDistributionDimension = conditioningDistribution_.getDimension();
+  if ((conditioningDimension < conditioningDistributionDimension))
+    return conditioningDistribution_.computeConditionalPDF(x, y);
+  // The conditioning part is fully conditioned, let's evaluate the link function
+  Point fixedConditioningPart(conditioningDistributionDimension);
+  std::copy(y.begin(), y.begin() + conditioningDistributionDimension, fixedConditioningPart.begin());
+  Distribution conditioned(conditionedDistribution_);
+  conditioned.setParameter(linkFunction_(fixedConditioningPart));
+  Point fixedConditionedPart(conditioningDistributionDimension - conditioningDistributionDimension);
+  if (fixedConditionedPart.getDimension() > 0)
+    std::copy(y.begin() + conditioningDistributionDimension, y.end(), fixedConditionedPart.begin());
+  return conditioned.computeConditionalPDF(x, fixedConditionedPart);
+}
+
+/* Compute the PDF of Xi | X1, ..., Xi-1. x = Xi, y = (X1,...,Xi-1) */
+Point JointByConditioningDistribution::computeSequentialConditionalPDF(const Point & x) const
+{
+  if (x.getDimension() != dimension_) throw InvalidArgumentException(HERE) << "Error: expected a point of dimension=" << dimension_ << ", got dimension=" << x.getDimension();
+  const UnsignedInteger conditioningDistributionDimension = conditioningDistribution_.getDimension();
+  Point result(dimension_);
+  Point conditioningArgument(conditioningDistributionDimension);
+  std::copy(x.begin(), x.begin() + conditioningDistributionDimension, conditioningArgument.begin());
+  const Point conditioningConditionalPDF(conditioningDistribution_.computeSequentialConditionalPDF(conditioningArgument));
+  std::copy(conditioningConditionalPDF.begin(), conditioningConditionalPDF.begin() + conditioningDistributionDimension, result.begin());
+  Distribution conditioned(conditionedDistribution_);
+  conditioned.setParameter(linkFunction_(conditioningArgument));
+  Point conditionedArgument(conditioned.getDimension());
+  std::copy(x.begin() + conditioningDistributionDimension, x.end(), conditionedArgument.begin());
+  const Point conditionedConditionalPDF(conditioned.computeSequentialConditionalPDF(conditionedArgument));
+  std::copy(conditionedConditionalPDF.begin(), conditionedConditionalPDF.end(), result.begin() + conditioningDistributionDimension);
+  return result;
+}
+
+/* Compute the CDF of Xi | X1, ..., Xi-1. x = Xi, y = (X1,...,Xi-1) */
+Scalar JointByConditioningDistribution::computeConditionalCDF(const Scalar x,
+							      const Point & y) const
+{
+  const UnsignedInteger conditioningDimension = y.getDimension();
+  if (conditioningDimension >= getDimension()) throw InvalidArgumentException(HERE) << "Error: cannot compute a conditional CDF with a conditioning point of dimension greater or equal to the distribution dimension.";
+  // Special case for a conditioning only in the conditioning part
+  const UnsignedInteger conditioningDistributionDimension = conditioningDistribution_.getDimension();
+  if ((conditioningDimension < conditioningDistributionDimension))
+    return conditioningDistribution_.computeConditionalCDF(x, y);
+  // The conditioning part is fully conditioned, let's evaluate the link function
+  Point fixedConditioningPart(conditioningDistributionDimension);
+  std::copy(y.begin(), y.begin() + conditioningDistributionDimension, fixedConditioningPart.begin());
+  Distribution conditioned(conditionedDistribution_);
+  conditioned.setParameter(linkFunction_(fixedConditioningPart));
+  Point fixedConditionedPart(conditioningDimension - conditioningDistributionDimension);
+  if (fixedConditionedPart.getDimension() > 0)
+    std::copy(y.begin() + conditioningDistributionDimension, y.end(), fixedConditionedPart.begin());
+  return conditioned.computeConditionalCDF(x, fixedConditionedPart);
+}
+
+/* Compute the CDF of Xi | X1, ..., Xi-1. x = Xi, y = (X1,...,Xi-1) */
+Point JointByConditioningDistribution::computeSequentialConditionalCDF(const Point & x) const
+{
+  if (x.getDimension() != dimension_) throw InvalidArgumentException(HERE) << "Error: expected a point of dimension=" << dimension_ << ", got dimension=" << x.getDimension();
+  const UnsignedInteger conditioningDistributionDimension = conditioningDistribution_.getDimension();
+  Point result(dimension_);
+  Point conditioningArgument(conditioningDistributionDimension);
+  std::copy(x.begin(), x.begin() + conditioningDistributionDimension, conditioningArgument.begin());
+  const Point conditioningConditionalCDF(conditioningDistribution_.computeSequentialConditionalCDF(conditioningArgument));
+  std::copy(conditioningConditionalCDF.begin(), conditioningConditionalCDF.begin() + conditioningDistributionDimension, result.begin());
+  Distribution conditioned(conditionedDistribution_);
+  conditioned.setParameter(linkFunction_(conditioningArgument));
+  Point conditionedArgument(conditioned.getDimension());
+  std::copy(x.begin() + conditioningDistributionDimension, x.end(), conditionedArgument.begin());
+  const Point conditionedConditionalCDF(conditioned.computeSequentialConditionalCDF(conditionedArgument));
+  std::copy(conditionedConditionalCDF.begin(), conditionedConditionalCDF.end(), result.begin() + conditioningDistributionDimension);
+  return result;
+}
+
+/* Compute the quantile of Xi | X1, ..., Xi-1, i.e. x such that CDF(x|y) = q with x = Xi, y = (X1,...,Xi-1) */
+Scalar JointByConditioningDistribution::computeConditionalQuantile(const Scalar q,
+								   const Point & y) const
+{
+  const UnsignedInteger conditioningDimension = y.getDimension();
+  if (conditioningDimension >= getDimension()) throw InvalidArgumentException(HERE) << "Error: cannot compute a conditional quantile with a conditioning point of dimension greater or equal to the distribution dimension.";
+  if ((q < 0.0) || (q > 1.0)) throw InvalidArgumentException(HERE) << "Error: cannot compute a conditional quantile for a probability level outside of [0, 1]";
+  // Special case for a conditioning only in the conditioning part
+  const UnsignedInteger conditioningDistributionDimension = conditioningDistribution_.getDimension();
+  if ((conditioningDimension < conditioningDistributionDimension))
+    return conditioningDistribution_.computeConditionalQuantile(q, y);
+  // The conditioning part is fully conditioned, let's evaluate the link function
+  Point fixedConditioningPart(conditioningDistributionDimension);
+  std::copy(y.begin(), y.begin() + conditioningDistributionDimension, fixedConditioningPart.begin());
+  Distribution conditioned(conditionedDistribution_);
+  conditioned.setParameter(linkFunction_(fixedConditioningPart));
+  Point fixedConditionedPart(conditioningDimension - conditioningDistributionDimension);
+  if (fixedConditionedPart.getDimension() > 0)
+    std::copy(y.begin() + conditioningDistributionDimension, y.end(), fixedConditionedPart.begin());
+  return conditioned.computeConditionalQuantile(q, fixedConditionedPart);
+}
+
+/* Compute the quantile of Xi | X1, ..., Xi-1. x = Xi, y = (X1,...,Xi-1) */
+Point JointByConditioningDistribution::computeSequentialConditionalQuantile(const Point & q) const
+{
+  if (q.getDimension() != dimension_) throw InvalidArgumentException(HERE) << "Error: expected a point of dimension=" << dimension_ << ", got dimension=" << q.getDimension();
+  const UnsignedInteger conditioningDistributionDimension = conditioningDistribution_.getDimension();
+  Point result(dimension_);
+  Point conditioningArgument(conditioningDistributionDimension);
+  std::copy(q.begin(), q.begin() + conditioningDistributionDimension, conditioningArgument.begin());
+  const Point conditioningConditionalQuantile(conditioningDistribution_.computeSequentialConditionalQuantile(conditioningArgument));
+  std::copy(conditioningConditionalQuantile.begin(), conditioningConditionalQuantile.begin() + conditioningDistributionDimension, result.begin());
+  Distribution conditioned(conditionedDistribution_);
+  conditioned.setParameter(linkFunction_(conditioningConditionalQuantile));
+  Point conditionedArgument(conditioned.getDimension());
+  std::copy(q.begin() + conditioningDistributionDimension, q.end(), conditionedArgument.begin());
+  const Point conditionedConditionalQuantile(conditioned.computeSequentialConditionalQuantile(conditionedArgument));
+  std::copy(conditionedConditionalQuantile.begin(), conditionedConditionalQuantile.end(), result.begin() + conditioningDistributionDimension);
+  return result;
+}
+
 /* Method save() stores the object through the StorageManager */
 void JointByConditioningDistribution::save(Advocate & adv) const
 {

diff --git a/lib/src/Uncertainty/Distribution/openturns/JointByConditioningDistribution.hxx b/lib/src/Uncertainty/Distribution/openturns/JointByConditioningDistribution.hxx
@@ -102,6 +102,21 @@ public:
   /** Parameters description accessor */
   Description getParameterDescription() const override;
 
+  /** Compute the PDF of Xi | X1, ..., Xi-1. x = Xi, y = (X1,...,Xi-1) */
+  using DistributionImplementation::computeConditionalPDF;
+  Scalar computeConditionalPDF(const Scalar x, const Point & y) const override;
+  Point computeSequentialConditionalPDF(const Point & x) const override;
+
+  /** Compute the CDF of Xi | X1, ..., Xi-1. x = Xi, y = (X1,...,Xi-1) */
+  using DistributionImplementation::computeConditionalCDF;
+  Scalar computeConditionalCDF(const Scalar x, const Point & y) const override;
+  Point computeSequentialConditionalCDF(const Point & x) const override;
+
+  /** Compute the quantile of Xi | X1, ..., Xi-1, i.e. x such that CDF(x|y) = q with x = Xi, y = (X1,...,Xi-1) */
+  using DistributionImplementation::computeConditionalQuantile;
+  Scalar computeConditionalQuantile(const Scalar q, const Point & y) const override;
+  Point computeSequentialConditionalQuantile(const Point & q) const override;
+
   /** Method save() stores the object through the StorageManager */
   void save(Advocate & adv) const override;
 

diff --git a/python/test/t_JointByConditioningDistribution_std.expout b/python/test/t_JointByConditioningDistribution_std.expout
@@ -23,3 +23,9 @@ anotherSample covariance= [[  0.0830803    0.000313403  0.0767514   ]
 Point point=  [0.5, 1.5, 1.0]  pdf=0.251589  cdf=0.181637
 Quantile= [0.982796,1.9828,3.87457]
 CDF(quantile)= 0.95
+conditional PDF=1.09340e-01
+conditional CDF=9.08789e-01
+conditional quantile=2.50000e+00
+sequential conditional PDF= [1,1,0.10934]
+sequential conditional CDF( [0.5, 1.5, 2.5] )= [0.5,0.5,0.908789]
+sequential conditional quantile( [0.5,0.5,0.908789] )= [0.5,1.5,2.5]
diff --git a/python/test/t_JointByConditioningDistribution_std.py b/python/test/t_JointByConditioningDistribution_std.py
@@ -56,6 +56,26 @@
 print("Quantile=", quantile)
 print("CDF(quantile)= %.12g" % distribution.computeCDF(quantile))
 
+x = 2.5
+y = [0.5, 1.5]
+print("conditional PDF=%.5e" % distribution.computeConditionalPDF(x, y))
+condCDF = distribution.computeConditionalCDF(x, y)
+print("conditional CDF=%.5e" % condCDF)
+q = condCDF
+print("conditional quantile=%.5e" % distribution.computeConditionalQuantile(q, y))
+pt = [i + 0.5 for i in range(dim)]
+print(
+    "sequential conditional PDF=", distribution.computeSequentialConditionalPDF(pt)
+)
+resCDF = distribution.computeSequentialConditionalCDF(pt)
+print("sequential conditional CDF(", pt, ")=", resCDF)
+print(
+    "sequential conditional quantile(",
+    resCDF,
+    ")=",
+    distribution.computeSequentialConditionalQuantile(resCDF),
+)
+
 ot.Log.Show(ot.Log.TRACE)
 checker = ott.DistributionChecker(distribution)
 checker.skipMoments()  # slow