work_directory

Perform each function evaluation in a separate working directory

Specification

• Alias: None

• Arguments: None

• Default: no work directory

Child Keywords:

Required/Optional	Description of Group	Dakota Keyword	Dakota Keyword Description
Optional		named	The base name of the work directory created by Dakota
Optional		directory_tag	Tag each work directory with the function evaluation number
Optional		directory_save	Preserve the work directory after function evaluation completion
Optional		link_files	Paths to be linked into each working directory
Optional		copy_files	Files and directories to be copied into each working directory
Optional		replace	Overwrite existing files within a work directory

Description

When performing concurrent evaluations, it is typically necessary to cloister simulation input and output files in separate directories to avoid conflicts. When the work_directory feature is enabled, Dakota will create a directory for each evaluation, with optional tagging ( directory_tag) and saving ( directory_save ), as with files, and execute the analysis driver from that working directory.

The directory may be named with a string, or left anonymous to use an automatically-generated directory in the system’s temporary file space, e.g., /tmp/dakota_work_c93vb71z/. The optional link_files and copy_files keywords specify files or directories which should appear in each working directory.

When using work_directory, the analysis_drivers may be given by an absolute path, located in (or relative to) the startup directory alongside the Dakota input file, in the list of template files linked or copied, or on the $PATH (Path% on Windows).