Quick Start Guide ANDURIL

User Manual:

Open the PDF directly: View PDF .
Page Count: 4

Download
Open PDF In Browser	View PDF

Quick start guide
for ANDURIL
Georgios Leontaris & Oswaldo Morales-Nápoles
19 June 2018

Preface
As its title suggests, this document was written with the purpose of increasing the
“friendliness” of ANDURIL towards potential users. It is not the authors’ intention to present
a thorough step-by-step guide, rather than to clarify how the developed toolbox can be used.
It is recommended to be read in combination with the supplementary information (SI) for
ANDURIL - A MATLAB Toolbox for ANalysis and Decisions with UnceRtaInty: Learning
from expert judgments.

Requirements
MATLAB with Statistics and Machine Learning Toolbox installed

Installation
Double click on the ANDURIL.mtlbx
to install the toolbox. Please note that any
previously installed versions of ANDURIL should be uninstalled in advance. This can be
easily done by clicking on the Add-Ons -> Manage Custom Toolboxes in the Home tab of
MATLAB and choose uninstall (see Figure 1). After the successful installation of the toolbox,
the same window can be used to access all the functions by clicking on “Click to view toolbox
files in MATLAB Current Folder window”.

Figure 1: Manage Custom Toolboxes window.

ANDURIL Main Script
ANDURIL_main can be used for i) importing the values of expert judgments, ii) making the
analysis and synthesis of expert judgments and iii) post-processing of the resulting synthetic
decision makers (DMs).

i) Import data
The user has three options to import data in the format required by ANDURIL to perform the
analysis applying Cooke’s classical model. The first option is to enter the expert judgments

regarding seed and target variables as well as information regarding the background measure
and the realizations of every variable, manually in the required format as described in the SI.
The second option is to import the expert judgments regarding the seed and target variables
by using the import interface of MATLAB. It must be mentioned that these assessments could
have been either exported from EXCALIBUR (by using the Export as space delimited->
quantiles functionality) or manually entered in a spreadsheet. In the latter case, attention
should be payed to the format; this should be similar to that presented in Figure 2. Then,
only the values should be imported to MATLAB and the formulate_data.m function can be
used to formulate the assessments in the required format. Information regarding the number
of experts, the number of elicited quantiles, the number of seed and target variables, the
background measure of every variable as well as the realizations should be entered manually
in the required format as described in the SI. Examples are presented in ANRURIL_main.m
and example_ANDURIL.m.
The third option (and probably the favorite for EXCALIBUR users) is to use the .dtt and .rls
files of EXCALIBUR. These should be saved as txt files without any text for the description
of every variable. In case there is text this should be erased, otherwise the import_ascii_files
function will not work properly. Finally, the import_ascii_files.m can be used to formulate
the assessments of seed and target variables, as well as information regarding the
background measure and the realization of every variable. An example is presented in
ANRURIL_main.m.

Figure 2: Example of assessments regarding 8 seed and 2 target variables of 2 experts in a spreadsheet.

ii) Analysis and Synthesis of expert judgments
ANDURIL supports the synthesis of expert judgments based on four different weighting
schemes, i.e. four different types of DMs. Namely, the a) equal, b) user-defined, c) global
and d) item weights. However for the performance-based weighting schemes (i.e. global and
item weights) the option of optimization is available (for more information about this see the
SI). This results in two more possible DMs, the e) global weights with optimization and f)
item weights with optimization. The functions that should be called and the values of the
required parameters for the synthesis of each different DM are presented below (with the
same notation as these are presented in the ANRURIL_main.m). For more information
regarding the arguments of each function, see the SI.
a) Equal weights DM
Parameters:
 k = 0.1; (this is the default value)
 cal_power = 1; (this is the default value)
 optimization = ‘no’;
 weight_type = ‘equal’;
 alpha_eq = 0;
Function(s): calculate_DM_global.m

b) User defined weights DM
Parameters:
 k = 0.1; (this is the default value)
 cal_power = 1; (this is the default value)
 optimization = ‘no’;
 weight_type = ‘user’;
 user_w = [0 0 0 0 0.4 0.6 0 0 0]; (in this example expert 5 and expert 6 were
given weights equal to 0.4 and 0.6 respectively)
 alpha_ud = 0;
Function(s): calculate_DM_global.m
c) Global weights DM
Parameters:
 k = 0.1; (this is the default value)
 cal_power = 1; (this is the default value)
 optimization = ‘no’;
 weight_type = ‘global’;
 alpha = 0.05; (value used in the example presented)
Function(s): global_weights.m; calculate_DM_global.m
d) Item weights DM
Parameters:
 k = 0.1; (this is the default value)
 cal_power = 1; (this is the default value)
 optimization = ‘no’;
 weight_type = ‘global’;
 alpha = 0.05; (value used in the example presented)
Function(s): item_weights; calculate_DM_item
e) Global weights with optimization DM
Parameters:
 k = 0.1; (this is the default value)
 cal_power = 1; (this is the default value)
 optimization = ‘yes’;
 weight_type = ‘global’;
Funtion(s): DM_Optimization
f) Item weights with optimization DM
Parameters:
 k = 0.1; (this is the default value)
 cal_power = 1; (this is the default value)




optimization = ‘yes’;
weight_type = ‘global’;

Funtion(s): DM_Optimization

iii) Post-processing
ANDURIL supports the following post-processing functionalities:
a) Plotting assessments (itemwise). The user can plot the assessments (i.e. 5th, 50th and 95th
%tiles) of every individual expert and DM for a particular item in one plot. The function that
should be used for this purpose is the plotting_itemwise.m. Illustrative examples are
presented in ANDURIL_main.m, example_ANDURIL.m and the SI.
b) Checking robustness (itemwise). The user can investigate the robustness of the obtained
DM (i.e. the values of the statistical accuracy and informativeness) when up to k items are
excluded at the time. The option of including the calibration power is also provided. The
function that should be used for this purpose is the Checking_robustness_items.m.
Illustrative examples are presented in ANDURIL_main.m, example_ANDURIL.m and the
SI.
c) Checking robustness (expertwise). The user can investigate the robustness of the obtained
DM (i.e. the values of the statistical accuracy and informativeness) when up to k experts are
excluded at the time. The function that should be used for this purpose is the
Checking_robustness_experts.m. Illustrative examples are presented in ANDURIL_main.m,
example_ANDURIL.m and the SI.
d) Plotting box-plots of the performance measures (statistical accuracy and informativeness).
The user can produce box-plots of the statistical accuracy and informativeness of the DM
under investigation. Illustrative examples are presented in ANDURIL_main.m,
example_ANDURIL.m and the SI.



---

Source Exif Data:

File Type                       : PDF
File Type Extension             : pdf
MIME Type                       : application/pdf
PDF Version                     : 1.5
Linearized                      : No
Warning                         : Invalid secondary xref table
Page Count                      : 4
Language                        : nl-NL
Tagged PDF                      : Yes
Keywords                        : 
Producer                        : 
Modify Date                     : 
Create Date                     : 
Creator Tool                    : 
Metadata Date                   : 
Format                          : application/pdf
Title                           : 
Creator                         : EVISE
Description                     : 

EXIF Metadata provided by EXIF.tools