On 3/06/2018 8:15 p.m., PhDr. Mgr. Frantisek KALVAS Ph.D. wrote:
Dear Simon,
there is one thing I do not understand in the Figure 2 - the wealth of agents is surprisingly low, the maximum is about 60. I think it should be higher, around 100 at minimum with Tau=0 (100="always stay at Stable"). Some should be luckier, some not. But in case all agents are on risky pools, the average wealth in the long run should be 80...
František, You are right, and it did worry me too. I meant to look at it later, and somehow got sidetracked. I did notice that there were often too many people in the high and low pools. The problem is worse when there there are more coefficients, so I suspect that I'm not allowing enough training cycles for the predictors (Fogel handles this properly). I don't't understand why this should all be in one direction, though.
Regards,
Simon
So, how it comes the agents accumulate only around 60? May be my calculations are wrong and it is natural state, but in that case it interests me much more how it happens - because I could repair my wrong ideas and calculations.
On 1/06/2018 10:47 p.m., PhDr. Mgr. Frantisek KALVAS Ph.D. wrote:
Dear Simon,
here is my submission. Thank you for your constant support! The submission would look different and would be much poorer without interactions with you. Thank you for it! Thank you for the aiming my attention to rMarkdown - I hope I will use it soon.
František, Thanks for sharing your writeup and model. Unfortunately I
get an error message from the model (attached) in Netlogo 6.0.3. I'll
have a look later with a text editor, and see if I can fix. I'll read
your submission with interest today. I apologize for not crediting you
with the inspiration for using Experience in my 2nd group of agents. I
believe that I added them on your suggestion, and it then slipped from
my memory until I looked at the first page of the submission.
In the meantime I have attached my writeup. You will notice that I've
shamelessly exploited the fact that the 5 page limit does not included
figures and table. The code, including the RMarkdown source, can be
found at my github site https://github.com/weka511/201804. I first
heard about RMarkdown from the Data Scientist's Toolbox
https://www.coursera.org/learn/data-scientists-tools, and then I
usedRosanna van Hespen's
https://rosannavanhespenresearch.wordpress.com/ excellent blog post on
writing a thesis in RMarkdown, especially the material on figures and
the bibliography. I wrote my Master's Thesis back in 1971 using an IBM
Selectric typewriter, https://www.youtube.com/watch?v=vNUEUth7qjcwhich
had an "golfball" printing element, which could be changed for a math
symbols golfball. I also used a bucket of corrector fluid. Presentation
is much easier now.
Kind regards,
Simon
I am really curious for your assessment of my work. So in case you would not bump to it during peer review I attached it here. For the future I am really curious about issue im/perfect information vs. experience. The weakest point of my submission is the decision algorithms. I am not really good at it - I am a beginner, so it is not of professional quality. I am really interested in better decision algorithm based on im/perfect information or experience. I would love your advices how to improve these algorithms, I also would love if you join the effort on issue as co-author "algorithm designer". The present wrap-up is just departing point - without strong decision algorithms the results are not very convincing.
In case it is not whole population of agents, but just subsample, it might happen. So, in Fig 2, there are whole populations or subsamples?
František
Neděle, 3 Červen, 2018 12:00 CEST, Simon Crase simon@greenweaves.nz napsal: