Tasmanian has integration with the libEnsemble project, which is a Python library to coordinate the concurrent evaluation of dynamic ensembles of calculations. The workflow management system of libEnsemble can be wrapped around the Tasmanian sparse grid methods for constructing surrogate models and libEnsemble can serve as an intermediary between Tasmanian and the model of interest.

The functionality provided by libEnsemble is similar to that of TasGrid::constructSurrogate() and TasGrid::mpiConstructSurrogate(). While the native Tasmanian methods are functional, libEnsemble provides much more sophisticated workflow algorithms and much wider support for hardware architectures. One of the main advantages of libEnsemble is that the code is portable on shared memory multi-threaded environments, as well as numerous distributed cluster systems.

Requirements

Using Tasmanian and libEnsemble together requires:

Tasmanian 7.0 or newer
Tasmanian Python bindings are enabled, e.g., with Tasmanian_ENABLE_PYTHON=ON or the pip-installer
libEnsemble from the latest development branch
see the libEnsemble documentation for any additional requirements

Installation

The easiest way to install the latest development version of libEnsemble is to clone the repo and use a local pip-installer:

git clone https://github.com/Libensemble/libensemble.git
cd libensemble
git checkout develop
python3 setup.py sdist
python3 -m pip uninstall libensemble
python3 -m pip install dist/*.tar.gz

See the Installation instructions for Tasmanian.

Notes:

the pip approach required both git and the Python pip modules to be installed
see also the libEnsemble instructions for more details
both Tasmanian and libEnsemble works well within a Python venv environment

Basic Usage

An example of the Tasmanian-libEnsemble integration can be found in the libEnsemble testing suite. The Tasmanian sparse grid construction algorithm is implemented as the sparse_grid_batched libEnsemble native gen_f function:

from libensemble.gen_funcs.persistent_tasmanian import sparse_grid_batched as gen_f_batched

The construction method accepts the following user parameters:

gen_specs['user']['tasmanian_init'] -> a callable function with no inputs that returns an initialized object of Tasmanian.SparseGrid

gen_specs['user']['tasmanian_checkpoint_file'] -> a file for intermediary check pointing as well as the final grid

No additional parameters are required for static grid construction, e.g., similar to TasGrid::loadNeededPoints(). Iterative refinement can be enabled with the additional options for either anisotropic strategy

gen_specs['user']['refinement'] = 'setAnisotropicRefinement'
gen_specs['user']['sType'] = 'iptotal'
gen_specs['user']['iMinGrowth'] = 10
gen_specs['user']['iOutput'] = 0

or surplus strategy

gen_specs['user']['refinement'] = 'setSurplusRefinement'
gen_specs['user']['fTolerance'] = 1.E-2
gen_specs['user']['sCriteria'] = 'classic'
gen_specs['user']['iOutput'] = 0

See TasGrid::TasmanianSparseGrid::setAnisotropicRefinement() and TasGrid::TasmanianSparseGrid::setSurplusRefinement() for details.