Training

Dakota logo
Overview
Image of DAKOTA_Arrow_Name_Tag

There is no video for the Overview section

Duration: 45 minutes

  • What is Dakota?
  • Why use Dakota?
  • Prerequisites
Model Characterization

Duration: 100 minutes

  • See how Dakota can automate what you are already doing
  • Know what model characteristics will affect how you use Dakota
  • Be able to run a basic study to characterize a model
Input Syntax/Building Blocks

Duration: 60 Minutes

  • Develop an accurate “mental model” of Dakota components
  • Understand how to configure Dakota components using a Dakota input file
  • Become familiar with the Dakota Reference Manual
Interfacing I

Duration: 130 Minutes

  • Mechanics of how Dakota communicates with and runs a simulation
  • Requirements this places on the user and interface
  • Basic strategies for developing a simulation interface
  • Convenience features Dakota provides for managing simulation runs
  • Note: This module covers "black box" interfacing, not "library mode" Dakota
Interfacing II

Duration: 130 minutes

  • Mechanics of how Dakota communicates with and runs a simulation
  • Requirements this places on the user and interface
  • Basic strategies for developing a simulation interface
  • Convenience features Dakota provides for managing simulation runs
  • Note: This module covers "black box" interfacing, not "library mode" Dakota
Sensitivity Analysis

Duration: 90 minutes

  • Sensitivity analysis goals and examples
  • Global sensitivity analysis approaches and metrics available in Dakota
  • Dakota examples for parameter studies and global sensitivity analysis
Surrogate Models

Duration: 50

  • Define a surrogate model
  • Identify situations where it may be appropriate to use a surrogate model
  • Learn how to specify a surrogate model in Dakota
  • Run a surrogate model in Dakota and examine outputs based on the surrogate model
  • Identify some common diagnostics for surrogates
  • Understand different ways surrogates are used in Dakota
Optimization

Duration: 100

  • Understand potential goals of optimization and optimization terminology
  • Learn how to communicate the relevant problem information to Dakota
  • Become familiar with several types of optimization solvers and how to choose from among them based on problem type and goals
Calibration

Duration: 70 minutes

  • Why you might want to tune models to match data via calibration (parameter estimation)
  • How to formulate calibration problems and present them to Dakota
  • What Dakota methods can help you achieve calibration goals
Uncertainty Quantification

Duration: 125 minutes

  • Uncertainty quantification goals and examples
  • Examples for uncertainty quantification
  • Focus on forward propagation
Parallel Options

Duration: 60

  • Discuss what to consider when designing a parallelized study
  • Understand what Dakota provides and its limitations
  • Be able to choose the best parallelism approach
  • Know how to configure Dakota and your interface for your parallelism approach

Dakota training was offered in 2015-2016. Video recordings, slides, and other materials from the training are offered here. Although the information is dated in some ways, we believe it remains highly relevant and a valuable resource for learning about Dakota use.

Viewers may follow along with the exercises by downloading the materials for each module. The exercises were created for use with Dakota 6.3 on macOS or Linux, but users of slightly different versions of Dakota (6.0 or greater) and Windows users of Dakota will encounter few difficulties.

Plotting Tools

Exercises in the Model Characterization and Sensitivity Analysis modules make use of plotting tools created especially for the training.

Python 2.7 and the matplotlib and pandas libraries are required by the tools. Installing Anaconda is a convenient way to satisfy these requirements.

Updates and Errata

Updated (2016) materials and presentations are available, but they may not match those used in the videos.

Cantilever beam errata: In the content below (slides and simulation drivers), the stress equation for the cantilever beam incorrectly has a fixed length L = 100.

The analysis driver is corrected in versions of Dakota newer than 2019-09-05. Thanks to Anjali Sandip for reporting.