Dakota

Dakota, developed at Sandia National Laboratories, is a software toolkit that provides an interface between computational models and a library of iterative systems analysis methods. Dakota helps answer questions such as:

What are the most influential parameters in my model? → sensitivity analysis
How robust and reliable is my model? → uncertainty quantification
What is the best performing design? → optimization
What model parameter values best match experimental data? → calibration

Dakota has an extensive library of analysis techniques, including:

parameter studies
design of experiments
design of computer experiments (DACE)
sampling (MC or LHS)
local and global reliability
adaptive sampling
stochastic expansion
epistemic methods
gradient-based local and global optimization
derivative-free local and global optimization
nonlinear least squares
surrogate models

Setting up Dakota on beach

Dakota 6.1 is installed on the CSDMS HPCC, beach.colorado.edu. To use Dakota on beach, you'll need to modify your user environment. (This only needs to be done once.)

In your favorite text editor, open your .bashrc file (located in your home directory) and add the following lines to the bottom of the file:

PATH=$PATH:/usr/local/dakota/bin:/usr/local/dakota/test
export LD_LIBRARY_PATH=/usr/local/dakota/bin:/usr/local/dakota/lib

Save the file, logoff from beach, then login again. Dakota will now be available for use. Check with:

$ dakota --version
Dakota version 6.1 released Nov 15 2014.
Subversion revision 2951 built Nov  7 2014 09:52:09.

If you encounter any problems in setting up Dakota on beach, please email the CSDMS software engineers at CSDMSsupport@colorado.edu.

Using Dakota on beach

Topics to address:

Link to HPCC rules
At minimum, use qsub -I for a single-processor job
A simple example -- use Greg's, with the qsub submission script
Best practice: copy inputs to compute node (using /state/partition1), perform Dakota experiment, delete inputs, move results back to home directory, or to /scratch if space is an issue.
Running Dakota in parallel with MPI

If you encounter any problems in using Dakota on beach, please email the CSDMS software engineers at CSDMSsupport@colorado.edu.

Links

The Dakota home page is https://dakota.sandia.gov
- Download Dakota
- Install Dakota
The Dakota 6.5 (released November 2016) documentation is available at https://dakota.sandia.gov//sites/default/files/docs/6.5/html-ref/index.html
Adam Stephens and Laura Swiler, members of the Dakota development team at Sandia, gave a clinic at the 2014 CSDMS Annual Meeting which included a detailed discussion of sensitivity analysis, uncertainty quantification, and calibration in Dakota. Their presentation (PDF) is here.
Building Dakota on Ubuntu 12.04 LTS
The CSDMS Dakota Interface, Dakotathon

Todo

A list of topics that should be addressed in this page. Or linked to.

Why Dakota? Talk about about the need for UQ in modeling, and how Dakota provides a framework to help.
An overview of how Dakota works. Write this up in a more digestible manner than the way it's presented in the Dakota documentation.
A simple example of using Dakota; e.g., Greg's. (Make this another transcluded page.)
Installing and running Dakota on a personal computer
Setting up Dakota on beach
Using Dakota on beach