Statsclubproblems

From Computational Intelligence

(Difference between revisions)
Jump to: navigation, search
(Problem descriptions)
Current revision (10:48, 1 December 2020) (edit) (undo)
(Problem descriptions)
 
Line 1: Line 1:
-
== Problem descriptions ==
 
-
Add your problem description here for the discussion in the statistics reading club of 23 Feb
 
-
 
-
==== Wojtek ====
 
-
 
-
[http://www.few.vu.nl/~wojtek/files/StatsSlides.pdf Hereby] a compilation of my slides that I was using during the first two meetings. Pay special attention to slides 19-21 - they contain tasks for some of you
 
-
(Robert-Jan, Willem, Rob, Evert, Vincent) and give an idea of what to expect on Monday 21st (at 14:00).
 
-
 
-
==== Selmar/Gusz ====
 
-
The dataset consists of 9 subsets of experimental data. In each subset the results are given of 100 runs of a specific EA on a specific problem-function.
 
-
There are three different EAs, and three different functions: Rastrigin, Sphere and a handcrafted stepped function.
 
-
 
-
For each run the following metrics are saved:
 
-
* Best fitness in the final population
 
-
* The number of problem evaluations needed to find the solution
 
-
* If the run terminated succesfully
 
-
 
-
The question is: Which EA is the best?
 
-
 
-
Download the dataset [http://www.cs.vu.nl/~sksmit/SRC.zip HERE] (zipped .MAT files)
 
-
 
-
==== Martijn ====
 
-
[[Image:oanec-stats.png|thumb|example results]]
 
-
 
-
We investigate effects of reciprocity and transivity in network formation. The dataset consists of 3 X 4 X 5 subsets over three dimensions:
 
-
 
-
* '''frequency_of_informal_opportunities''' \in [LOW, MEDIUM, HIGH]
 
-
* '''[operational_transitivity, operational_reciprocity]''' \in [[no,no], [yes,no],[no,yes],[yes,yes]]
 
-
* '''specialisation''' \in [admin, electro, nautical, technical, marines]
 
-
 
-
We measured number of reciprocated ties.
 
-
 
-
The question is: are the observed differences significant?
 
-
Or: what dimension has largest impact on number of reciprocated ties?
 
-
 
-
I have the dataset, but not readily available to include here.
 
-
 
-
==== Willem ====
 
-
This is a problem I had with a previous paper.
 
-
 
-
Stripped down, my problem comes down to the following:
 
-
I have 3 algorithms: 2 benchmarks and 1 new algorithm. I want to show
 
-
that my new algorithm outperforms the other algorithms. The output of
 
-
the algorithms is a single number, the 'value'.
 
-
 
-
The goal of the
 
-
algorithm is to locate (static) targets, who are distributed over some
 
-
terrain. I took 10 different target distributions, and tested each
 
-
algorithm on these distributions, with 10 different initial random
 
-
seeds.
 
-
 
-
The problem here is that the random seed makes a lot of difference in
 
-
how well the algorithms perform. This means that the mean performance
 
-
of an algorithm over different random seeds doesn't give me much
 
-
information. But, I can compare the outcomes of different algorithms
 
-
using the same initial random seed.
 
-
 
-
So, at each run, I computed the difference in value for new algorithm
 
-
vs. the two benchmarks. To show that my algorithm outperforms the
 
-
other algorithms, I now only have to show that these differences are
 
-
significantly higher than 0. I did this using the wilcoxon signed rank
 
-
test.
 
-
 
-
Attached you will find 2 data sets and 2 plots. The first data file
 
-
contains the difference between the new algorithm and the first
 
-
benchmark, the second date file contains the difference between the
 
-
new algorithm and the second benchmark. (the original outcomes of the
 
-
algorithms are on a different computer than i'm on right now, so I
 
-
cannot send you these.) Each row in the .dat files are the results for
 
-
one target distribution.
 
-
 
-
The two .pdf files are the plots for these differences. If you look at
 
-
them, you intuitively see that the the differences are generally
 
-
higher than 0. But, what is the best test to show this?
 
-
 
-
[http://www.few.vu.nl/~willem/files/dumb_vs_smart.dat dumb_vs_smart.dat] (plain text)
 
-
[http://www.few.vu.nl/~willem/files/dumb_vs_smart.pdf dumb_vs_smart.pdf] (pdf)
 
-
[http://www.few.vu.nl/~willem/files/det_vs_smart.dat det_vs_smart.dat] (plain text)
 
-
[http://www.few.vu.nl/~willem/files/det_vs_smart.pdf det_vs_smart.pdf] (pdf)
 

Current revision

Personal tools