Thank you for participating in this user study of the decision support tool "Mimer". We ask you to perform several tasks and answer a questionnaire about your experience.

1. Before starting the user study, please thoroughly read through both tutorials found here: https://assar.his.se/mimer/html/tutorials.html
The first tutorial showcases the different plots available in Mimer, the "Filter & Highlights" feature, and how to save and load "views".
The second tutorial shows a full use-case of Mimer and demonstrate knowledge discovery using FPM.
Please return to the tutorials as you need during the study.

 

2. The tasks to perform can be found here: https://drive.google.com/drive/folders/1AQQh3UuRMEH7WASVlsElgRCPBSYXpDR2?usp=sharing

3. Mimer can be found here: https://assar.his.se/mimer/html/

In this task we ask you to use the different visualization methods available in Mimer to visually identify the number of separated clusters available in three different dataset of varying number of dimensions. Please perform the following tasks:

(1) Load the view-file task1a.mimerview.
(2) Visually inspect dataset 1 and identify the number of clusters using:
    3D scatterplot,
    PCP, and
    RadViz.
(3) Visually inspect dataset 2 and identify the number of clusters using:
    PCP,
    RadViz, and
    t-SNE.
(3.1) Run t-SNE with all dimensions selected (https://assar.his.se/mimer/html/tutorial1.html#toc_tsne).
(4) Visually inspect dataset 3 and identify the number of clusters using:
    PCP,
    RadViz, and
    t-SNE.
(4.1) Run t-SNE with all dimensions selected (https://assar.his.se/mimer/html/tutorial1.html#toc_tsne).
(5) Save the current state of Mimer as a view.

In this task we ask you to use the linked plots feature of Mimer, and identify a knee point in the data in a 3D scatter plot, and use different plot to highlight solutions close to the knee point. Please perform the following tasks:

(1) Load the view-file task1b.mimerview.
(2) Open the dataset knee using a 3D scatterplot.
(3) Visually identify the knee point close to (f1 = 0.2, f2 = 0.2, f3 = 0.2).
(4) Use either PCP or RadViz to highlight a handful of solutions close to the knee point.
(5) Save the current state of Mimer as a view.

A "knee-point" is a point in the objective space where where small change in one objective will equal a large change in another objective. A knee-point is categorized by a bump or "knee" on the Pareto-optimal front. The figure below illustrates a knee-point in a 2D objective space with solutions close to the knee highlighted.

In this task we ask you to use FPM to generate knowledge about certain variables’ impacts on the non-dominated solutions from the data. You are given a dataset of solutions generated by optimizing a clutch break design problem (see figure), with two objectives: (f1) minimize total mass, (f2) and minimize stopping time, and five variables: (x1) inner disk radius, (x2) outer disk radius, (x3) disk thickness, (x4) actuating force, and (x5) number of disks. The dataset contains both dominated and non-dominated solutions and ranks according to non-dominated sorting. Please perform the following tasks:

(1) Load the view-file task1c.mimerview.
(2) Open the dataset clutch using a 2D scatterplot.
(3) Using the "Filter & Highlight" feature, highlight the non-dominated solutions using the rule "nds_rank == 0" (https://assar.his.se/mimer/html/tutorial1.html#toc_filter).
(4) Use FPM to find the values required for the disk thickness and the actuating force for these solutions. Use the highlighted solutions as the selected set and the default values from "Min. significance" and "Max. levels" (https://assar.his.se/mimer/html/tutorial2.html#toc_generatingfpmrules).
(5) Open the rules with the FPM graph plot (https://assar.his.se/mimer/html/tutorial2.html#toc_visualizingfpmrules).
(6) Save the current state of Mimer as a view.

 

In this task we ask you to use FPM to filter rules and find the most descriptive rule for a selection in the solutions. You are given a dataset of non-dominated solutions from an optimization of the DTLZ2 problem with three objectives. DTLZ2 is a mathematical test problem where some of the variables have a bigger influence on a solutions placement on the Pareto-optimal front than others. Solutions around the reference point (y0 = 0.5, y1 = 0.7, y2 = 0.5) are highlighted. Please perform the following tasks:

(1) Load the view-file task1d.mimerview.
(2) Open the dataset dtlz2 using a 3D scatterplot.
(3) Use FPM to find rules over all variables (x0−x11) for the highlighted solutions. Use the default values for "Min. significance" and "Max. levels".
(4) Open the rules with the FPM graph plot.
(5) By changing the "Levels" value and using the sliders, filter the rules to find the single rule-interaction that seems most descriptive to you with a significance over 90% (https://assar.his.se/mimer/html/tutorial2.html#toc_visualizingfpmrules).
(6) Save the current state of Mimer as a view.

A "descriptive" rule would be one that with a high significance and a low unsignificance at the same time, i.e. describing solutions in the selected set while not describing solutions in the unselected set.

The "significance" of an FPM-rule is determined by the fraction of solutions in the selected set that are covered by the rule.
The "unselected significance" or "unsignificance" is determined by the fraction of solutions in the unselected set that are also covered by the rule.

In this task we combine the previous subtasks into a full use case. You are given a dataset of non-dominated solutions from an optimization run of a vehicle crash-worthiness design problem with three objectives. The objectives are to minimize the weight of the vehicle (f0), acceleration characteristics (f1), and toe-board instruction in a crash (f2), the five variables of the problem each relate to the thickness of one support beam in the front of the vehicle. Please perform the following tasks:

(1) Load the view-file task2.mimerview.
(2) Open the dataset RE354 using a 3D scatterplot and visually identify the three clusters of solutions.
(3) Highlight all solutions in the cluster with a value of f2 > 0.15, using any method.
(4) Use FPM to find rules for all variables (x0-x4) that best describe this cluster.
(5) Identify the most important variable for this cluster.

(6) Reset the highlight and make a new selection of the solutions within this cluster that has values for f0 < 1678 and f2 < 0.2.
(7) Use FPM to find rules for all variables (x0-x4) for this selection, and by changing the "Levels" value and using the sliders, filter the rules to find the single rule-interaction that seems most descriptive to you with a significance over 90%.
(8) Save the current state of Mimer as a view.