We must add the scalar type for the gaussian processes. Here, the type implies the dimension of the space being emulated.

Currently, queso supports vector types, but not scalars.

To finish this ticket, we must:

• Understand how the matlab code deals with this
• Design the API (explicitly selecting scalar vs. vector? or automatic detection)
• How much of the existing vector code can be reused?

Should also consider if we wish to discuss this in the GPMSA meeting tomorrow.

See examples/statisticalInverse_Fortran

Note I had to reorder MPI_Abort() and queso_error() in Defines.h in #75. This may lead to uncaught exceptions which in turn means that MPI_Abort is not called and processes still go about their business when they shouldn't. We need to check for uncaught exceptions and call MPI_Abort if this is the case.

I think libMesh does this.

Form the Jacobian, e.g. d(Z)/d(theta) where Z is the emulator and theta are the parameters.

As they say in the paper:

With these, one can explore the sensitivities of the simulator model, describe the parameter settings for which the simulator is consistent with the
data, and make predictions for the physical system at settings x for which no experimental data exist.

To reproduce:

#include <vector>
#include <string>
#include <iostream>
#include <queso/Environment.h>
#include <queso/GslVector.h>
#include <queso/GslMatrix.h>
#include <queso/VectorSpace.h>

int main(int argc, char **argv) {
  MPI_Init(&argc, &argv);

  QUESO::EnvOptionsValues options;
  options.m_numSubEnvironments = 1;

  QUESO::FullEnvironment *env = new QUESO::FullEnvironment(MPI_COMM_WORLD, "",
      "", &options);

  std::vector<std::string> names(1);
  names[0] = "my_name";
  QUESO::VectorSpace<QUESO::GslVector, QUESO::GslMatrix> vec_space(*env,
      "vec_prefix", 1, &names);
  QUESO::VectorSpace<QUESO::GslVector, QUESO::GslMatrix> vspace_cpy(vec_space);

  delete env;
  MPI_Finalize();

  return 0;
}

Here's the error:

dyld: lazy symbol binding failed: Symbol not found: __ZN5QUESO9DistArrayISsE4copyERKS1_
  Referenced from: /Users/damon/git/github/queso/build/sep_impl_vectorspace/src/.libs/libqueso-0.47.dylib
  Expected in: flat namespace

dyld: Symbol not found: __ZN5QUESO9DistArrayISsE4copyERKS1_
  Referenced from: /Users/damon/git/github/queso/build/sep_impl_vectorspace/src/.libs/libqueso-0.47.dylib
  Expected in: flat namespace

Looks like our friend, the DistArray copy constructor, is back.

We need a spiffy gravitar icon for queso. To get the dialogue going (and arguably, to use as a placeholder) I'm proposing:

http://users.ices.utexas.edu/~nick/collaborate/queso/queso-icon.jpg

or,

http://users.ices.utexas.edu/~nick/collaborate/queso/queso-icon-yellow.jpg

Thoughts?

Will require adding header files as target dependencies. Workaround is touch doxygen/queso.dox before running make doc.

I saw on Redmine that Ernesto has been updating QUESO code. Paul noticed that he is adding options to disable parameters, but we need to keep an eye on it:

Date Author Comment

41865 09/30/2013 08:42 pm Ernesto Prudencio changes for disabling parameters is user wants to

41858 09/30/2013 01:49 am Ernesto Prudencio For GPMSA tests: adding capability to disable parameters

$ grep -l -r template src/stats/inc

Gives this list:

• ./FiniteDistribution.h
• ./InfoTheory.h (see #106)
• ./JointPdf.h (see #107)
• ./MarkovChainPositionData.h (see #108)
• ./MatrixCovarianceFunction.h (see #109)
• ./MetropolisHastingsSG1.h (see #110)
• ./MetropolisHastingsSG2.h (see #110)
• ./MLSampling1.h (see #111)
• ./MLSampling2.h (see #111)
• ./MLSampling3.h (see #111)
• ./ModelValidation.h (see #112)
• ./MonteCarloSG.h (see #113)
• ./ScalarCdf.h (see #114)
• ./ScalarCovarianceFunction.h (see #115)
• ./ScalarGaussianRandomField.h (see #116)
• ./StatisticalForwardProblem.h (see #117)
• ./StatisticalInverseProblem.h (see #118)
• ./TKGroup.h (see #121)
• ./ValidationCycle.h (see #119)
• ./VectorCdf.h (see #122)
• ./VectorGaussianRandomField.h (see #120)
• ./VectorMdf.h (see #123)
• ./VectorRealizer.h (see #124)
• ./VectorRV.h

Code coverage of this folder is sitting at 0.7% on my laptop. We could probably do better.

The release v0.47.0 went with a few out-of-date information and with an example that had a bug.

This new manual comes with 2 new, described examples (simpleStatisticalInveseProblem and simpleStatisticalForwardProblem), and includes the description of another one (validationCycle).

It should be used as the base ref for future user' s manuals.

This got commented out in the merge mentioned in #65 because I wasn't sure if it was working.

@kemelli let me know if this part of examples/Makefile.am can be uncommented

Queso currently deals with faults by calling exit(1). This is not robust with respect to parallel computation, because any exit calls will tear down the entire routine.

Instead, we should move to exception handling that permits catching failures and dealing with them in an appropriate manner.

I propose we start this process by replacing all exit calls with queso_exit. To start this can invoke exit(1), and later this method can be overloaded. I'm guessing a quick alternative will be grvy_exit(), but long term I believe we should investigate fault tolerant process handling.

grep -rl 'template' src/basic/inc

• ./ArrayOfSequences.h (#51)
• ./InstantiateIntersection.h (see #62)
• ./ScalarFunction.h (see #66)
• ./ScalarFunctionSynchronizer.h (see #64)
• ./ScalarSequence.h (see #82)
• ./SequenceOfVectors.h (see #79)
• ./SequenceStatisticalOptions.h (NOT TEMPLATED)
• ./VectorFunction.h (see #80)
• ./VectorFunctionSynchronizer.h (see #83)
• ./VectorSequence.h (see #84)
• ./VectorSet.h (see #58)
• ./VectorSpace.h (see #56)
• ./VectorSubset.h (see #59)

Are these actual tests? If so, should they be in the top-level test directory?

@koomie suggested this one to lower the barrier of entry for new users. Helps questions like, "I need function foo, which header is it in?"

Now that we have nifty exception handling, let's replace the numerous exit(1) calls with our mpi-friendly queso_exit()

queso-users and queso-dev are probably good names. Google groups ok with everyone? I'm open to suggestions.

@kemelli Would you like to set these up?

I think this code should run without error:

#include <vector>
#include <string>
#include <iostream>
#include <set>
#include <queso/Environment.h>
#include <queso/GslVector.h>
#include <queso/GslMatrix.h>
#include <queso/VectorSpace.h>
#include <queso/SequenceOfVectors.h>

int main(int argc, char **argv) {
  MPI_Init(&argc, &argv);

  QUESO::EnvOptionsValues options;
  options.m_numSubEnvironments = 1;
  options.m_subDisplayAllowAll = 0;
  options.m_subDisplayAllowedSet.insert(0);

  QUESO::FullEnvironment *env = new QUESO::FullEnvironment(MPI_COMM_WORLD, "",
            "", &options);

  // Create a vector space
  std::vector<std::string> names(1);
  names[0] = "my_name";
  QUESO::VectorSpace<QUESO::GslVector, QUESO::GslMatrix> vec_space(*env,
      "vec_prefix", 1, &names);

  // Create some vectors to put in the VectorSequence
  QUESO::GslVector v1(vec_space.zeroVector());
  QUESO::GslVector v2(vec_space.zeroVector());
  v1[0] = 0.0;
  v2[0] = 1.0;

  // Create a sequence of vectors
  QUESO::SequenceOfVectors<QUESO::GslVector, QUESO::GslMatrix> vec_seq(vec_space, 2, "vec_seq");

  vec_seq.setPositionValues(0, v1);
  vec_seq.setPositionValues(1, v2);

  // Now erase just one element (the 0th element)
  vec_seq.erasePositions(0, 1);

  MPI_Finalize();

  return 0;
}

This relies on the copy method in DistArray being implemented.

The Macros in src/core/inc/Defines.h:

• UQ_RC_MACRO
• UQ_TEST_MACRO
• UQ_FATAL_RC_MACRO
• UQ_FATAL_TEST_MACRO

Are all old and do not appropriately understand MPI. Let's replace them with queso_exit (and associated) calls, which can provide catchable error handling as well as better information on what caused the failures.

Ernesto made two commits on the 8th of September. Merge these back into Queso git proper.

It lives in 1D1DFunction.h.

Stick with the iso_c example as a better way to access QUESO from Fortran

git doesn't think too kindly of keyword substitution. But, I do miss a timestamp header at times. This can be pulled off with some additional attribute customization, but it may not be worth it or of sufficient interest. Evaluate and either:

• update to use external attribute markup or
• toss old SVN keywords from src

• Port to markdown so it's prettified on the repository landing page
• Add information on how to contribute
• Add information regarding the licence (just a reference to the LICENSE) file will suffice

We should fix this.

$ grep -rl 'template' src/gp/inc

• ./ExperimentModel.h (see #130)
• ./ExperimentStorage.h (see #131)
• ./GcmExperimentInfo.h (see #132)
• ./GcmJointInfo.h (see #133)
• ./GcmJointTildeInfo.h (see #134)
• ./GcmSimulationInfo.h (see #135)
• ./GcmSimulationTildeInfo.h (see #136)
• ./GcmTotalInfo.h (see #137)
• ./GcmZInfo.h (see #138)
• ./GcmZTildeInfo.h (see #139)
• ./GpmsaComputerModel.h (see #142)
• ./GpmsaComputerModel2.h (see #142)
• ./GpmsaComputerModel3.h (see #142)
• ./GpmsaComputerModel4.h (see #142)
• ./GpmsaComputerModel5.h (see #142)
• ./SimulationModel.h (see #140)
• ./SimulationStorage.h (see #141)

Some of the tests are silently failing out-of-source. Fix imminent.

Dakota has a wrapper for Bayesian calibration with QUESO, which will have to be re-written when the changes in QUESO classes names become official.
This is just so you guys are aware and work with Laura Swiler/Brian Williams when the time comes.
:)
Here is it: https://software.sandia.gov/trac/dakota/browser/core/trunk/src/NonDQUESOBayesCalibration.cpp

branch v0.47.1 has the most current version of the user's manual. It needs to be merged to dev.

MPI is required, so this is unnecessary.

grep -rl 'template' src/core/inc

• ./BasicPdfsBase.h
• ./DistArray.h (#36)
• ./GslMatrix.h
• ./GslVector.h
• ./TeuchosMatrix.h
• ./TeuchosVector.h

This example was created by Gabriel when using library ANN.
It needs to be tested and documented, so while that is not done, we should remove it from Queso distribution.

make check output after setting --without-glpk to configure:

libtool: link: mpic++ -g -O2 -Wall -o .libs/BimodalExample_gsl example_main.o example_compute.o example_likelihood.o -Wl,-rpath -Wl,/opt/local/lib -Wl,-bind_at_load  -L../../src/ /Users/damon/git/github/queso/src/.libs/libqueso.dylib -L/opt/local/lib /opt/local/lib/libgsl.dylib /opt/local/lib/libgslcblas.dylib -lboost_program_options-mt /opt/local/lib/libhdf5.dylib
Undefined symbols for architecture x86_64:
  "_glp_add_cols", referenced from:
      uqMLSamplingClass<uqGslVectorClass, uqGslMatrixClass>::solveBIP_proc0(std::vector<uqExchangeInfoStruct, std::allocator<uqExchangeInfoStruct> >&) in example_compute.o
  "_glp_add_rows", referenced from:
      uqMLSamplingClass<uqGslVectorClass, uqGslMatrixClass>::solveBIP_proc0(std::vector<uqExchangeInfoStruct, std::allocator<uqExchangeInfoStruct> >&) in example_compute.o
  "_glp_create_prob", referenced from:
      uqMLSamplingClass<uqGslVectorClass, uqGslMatrixClass>::solveBIP_proc0(std::vector<uqExchangeInfoStruct, std::allocator<uqExchangeInfoStruct> >&) in example_compute.o
  "_glp_delete_prob", referenced from:
      uqMLSamplingClass<uqGslVectorClass, uqGslMatrixClass>::solveBIP_proc0(std::vector<uqExchangeInfoStruct, std::allocator<uqExchangeInfoStruct> >&) in example_compute.o
  "_glp_get_num_bin", referenced from:
      uqMLSamplingClass<uqGslVectorClass, uqGslMatrixClass>::solveBIP_proc0(std::vector<uqExchangeInfoStruct, std::allocator<uqExchangeInfoStruct> >&) in example_compute.o
  "_glp_get_num_cols", referenced from:
      uqMLSamplingClass<uqGslVectorClass, uqGslMatrixClass>::solveBIP_proc0(std::vector<uqExchangeInfoStruct, std::allocator<uqExchangeInfoStruct> >&) in example_compute.o
  "_glp_get_num_int", referenced from:
      uqMLSamplingClass<uqGslVectorClass, uqGslMatrixClass>::solveBIP_proc0(std::vector<uqExchangeInfoStruct, std::allocator<uqExchangeInfoStruct> >&) in example_compute.o
  "_glp_get_num_nz", referenced from:
      uqMLSamplingClass<uqGslVectorClass, uqGslMatrixClass>::solveBIP_proc0(std::vector<uqExchangeInfoStruct, std::allocator<uqExchangeInfoStruct> >&) in example_compute.o
  "_glp_get_num_rows", referenced from:
      uqMLSamplingClass<uqGslVectorClass, uqGslMatrixClass>::solveBIP_proc0(std::vector<uqExchangeInfoStruct, std::allocator<uqExchangeInfoStruct> >&) in example_compute.o
  "_glp_init_iocp", referenced from:
      uqMLSamplingClass<uqGslVectorClass, uqGslMatrixClass>::solveBIP_proc0(std::vector<uqExchangeInfoStruct, std::allocator<uqExchangeInfoStruct> >&) in example_compute.o
  "_glp_intopt", referenced from:
      uqMLSamplingClass<uqGslVectorClass, uqGslMatrixClass>::solveBIP_proc0(std::vector<uqExchangeInfoStruct, std::allocator<uqExchangeInfoStruct> >&) in example_compute.o
  "_glp_load_matrix", referenced from:
      uqMLSamplingClass<uqGslVectorClass, uqGslMatrixClass>::solveBIP_proc0(std::vector<uqExchangeInfoStruct, std::allocator<uqExchangeInfoStruct> >&) in example_compute.o
  "_glp_mip_col_val", referenced from:
      uqMLSamplingClass<uqGslVectorClass, uqGslMatrixClass>::solveBIP_proc0(std::vector<uqExchangeInfoStruct, std::allocator<uqExchangeInfoStruct> >&) in example_compute.o
  "_glp_mip_obj_val", referenced from:
      uqMLSamplingClass<uqGslVectorClass, uqGslMatrixClass>::solveBIP_proc0(std::vector<uqExchangeInfoStruct, std::allocator<uqExchangeInfoStruct> >&) in example_compute.o
  "_glp_mip_row_val", referenced from:
      uqMLSamplingClass<uqGslVectorClass, uqGslMatrixClass>::solveBIP_proc0(std::vector<uqExchangeInfoStruct, std::allocator<uqExchangeInfoStruct> >&) in example_compute.o
  "_glp_mip_status", referenced from:
      uqMLSamplingClass<uqGslVectorClass, uqGslMatrixClass>::solveBIP_proc0(std::vector<uqExchangeInfoStruct, std::allocator<uqExchangeInfoStruct> >&) in example_compute.o
  "_glp_set_col_bnds", referenced from:
      uqMLSamplingClass<uqGslVectorClass, uqGslMatrixClass>::solveBIP_proc0(std::vector<uqExchangeInfoStruct, std::allocator<uqExchangeInfoStruct> >&) in example_compute.o
  "_glp_set_col_kind", referenced from:
      uqMLSamplingClass<uqGslVectorClass, uqGslMatrixClass>::solveBIP_proc0(std::vector<uqExchangeInfoStruct, std::allocator<uqExchangeInfoStruct> >&) in example_compute.o
  "_glp_set_col_name", referenced from:
      uqMLSamplingClass<uqGslVectorClass, uqGslMatrixClass>::solveBIP_proc0(std::vector<uqExchangeInfoStruct, std::allocator<uqExchangeInfoStruct> >&) in example_compute.o
  "_glp_set_col_stat", referenced from:
      uqMLSamplingClass<uqGslVectorClass, uqGslMatrixClass>::solveBIP_proc0(std::vector<uqExchangeInfoStruct, std::allocator<uqExchangeInfoStruct> >&) in example_compute.o
  "_glp_set_obj_coef", referenced from:
      uqMLSamplingClass<uqGslVectorClass, uqGslMatrixClass>::solveBIP_proc0(std::vector<uqExchangeInfoStruct, std::allocator<uqExchangeInfoStruct> >&) in example_compute.o
  "_glp_set_obj_dir", referenced from:
      uqMLSamplingClass<uqGslVectorClass, uqGslMatrixClass>::solveBIP_proc0(std::vector<uqExchangeInfoStruct, std::allocator<uqExchangeInfoStruct> >&) in example_compute.o
  "_glp_set_prob_name", referenced from:
      uqMLSamplingClass<uqGslVectorClass, uqGslMatrixClass>::solveBIP_proc0(std::vector<uqExchangeInfoStruct, std::allocator<uqExchangeInfoStruct> >&) in example_compute.o
  "_glp_set_row_bnds", referenced from:
      uqMLSamplingClass<uqGslVectorClass, uqGslMatrixClass>::solveBIP_proc0(std::vector<uqExchangeInfoStruct, std::allocator<uqExchangeInfoStruct> >&) in example_compute.o
  "_glp_set_row_name", referenced from:
      uqMLSamplingClass<uqGslVectorClass, uqGslMatrixClass>::solveBIP_proc0(std::vector<uqExchangeInfoStruct, std::allocator<uqExchangeInfoStruct> >&) in example_compute.o
  "_glp_set_row_stat", referenced from:
      uqMLSamplingClass<uqGslVectorClass, uqGslMatrixClass>::solveBIP_proc0(std::vector<uqExchangeInfoStruct, std::allocator<uqExchangeInfoStruct> >&) in example_compute.o
ld: symbol(s) not found for architecture x86_64

GLPK is optional, so we should put some conditional foo around these so they don't get snarfed by accident.