0
![Not allowed!](images/buttons/up_dis.png)
Thumbs Up |
Received: 6,663 Given: 7,052 |
I never got why some people at all deal with K13 data when there is also K36 data available. K13 can nothing that K36 can not. Why would anyone voluntarily use something worse without any corresponding benefit?![]()
Target: rothaer_scaled
Distance: 1.0091% / 0.01009085
39.8 (Balto-)Slavic
39.0 Germanic
19.2 Celtic-like
1.8 Graeco-Roman
0.2 Finnic-like
Thumbs Up |
Received: 8,978 Given: 14,002 |
Target: Oszkar
Distance: 226.5212% / 2.26521222 | R3P
60.2 Czech
29.4 Dutch
10.4 Portuguese
Target: Oszkar
Distance: 241.3569% / 2.41356880 | ADC: 0.25x RC
53.0 German
41.8 Czech
5.2 Portuguese
Distance to: Oszkar
5.96673277 German
5.98472222 Austrian
7.78249960 Slovene
9.00404909 Hungarian
9.03074748 Czech
9.87401134 Slovak
11.79202273 Croat
12.06940346 Belgian
12.55726085 Swedish
13.67268445 Dutch
14.57431302 Danish
14.64322369 English
15.24642253 Norwegian
15.67877546 Icelandic
16.22210221 Welsh
16.71436209 Serb
16.73810921 Scottish
17.19915986 Polish
18.07818852 Irish
18.92773626 Ukrainian
19.11602731 Romanian
21.44843118 Finnish
21.84309044 Russian
22.24666492 Belarusian
22.84712017 Estonian
Thumbs Up |
Received: 8,978 Given: 14,002 |
Target: OszkarArmenianWife
Distance: 240.4959% / 2.40495917
99.2 Armenian
0.8 Ukrainian
Thumbs Up |
Received: 12,626 Given: 32,029 |
too many components in K36, many of them not very informative, data seems spread a bit randomly, you don't know what to make out of many components (Easterners who score Basque, Westerners who score a bit of Armenian etc) and not the same availability of various regional and national scores as for K13.
I myself favour K13 even above G25 - plotting PC1 and PC2 in G25 leaves aside some % of data variation, while a t-SNE plot of K13 components in theory doesn't leave anything out and 13 components are few enough to show some clear geographical consistency
Thumbs Up |
Received: 5,242 Given: 18,144 |
Target: paradox
Distance: 4.4186% / 4.41856076 | ADC: 0.25x RC
73.0 Bulgarian
14.8 Maltese
12.2 Armenian
Target: Father
Distance: 2.2538% / 2.25381481 | ADC: 0.25x RC
72.8 Romanian
15.7 Maltese
11.4 Dutch
0.1 Bulgarian
Thumbs Up |
Received: 6,663 Given: 7,052 |
Too many for what exactly? Is there a problem for Vahaduo to deal with 36 components compared to 13?
Every information is an advantage and even a component with - theoretically - zero information is no disadvantage.
The only wrong is to have labeled the components anything else than component 1, component 2 etc. because are otherwise caused a lot of confusion. However, in a comparion, the labelling of the K36 components is completely irrelevant.
Of course, someone scoring "Italian" or "Basque" is not indicative of having any Italian or Basque ancestry.
K36 is - not surprisingly when having more components - more capable in distinguishing. F. i. it can distinguish a Germanic-Finnic mixture from a Germanic-Slavic mixture, which K13 cannot (unfortunately also G25 not, but the latter has other advantages). It's not by chance that LM Genetics is using K36.
That would to my perception be the only legit reason, if it's applicable. It's a pity that people at all collected K13 data instead of K36 which could as for today be done without any additions effort compatred to K13. I guess, the reason for going on K13 was the once lower effort and the opinion that it's sufficient.
Plotting in a 2-dimensional PCA is topic of its own, because you the have to reduce 13 or 36 components to two which alsways is a huge loss of information. on the other hand you can construct a PCA in 1000 different ways in how to define PC1 and PC2 out of the other components.
When it comes to a Vahaduo calculation all dimensions are used.
What is a "t-SNE plot"?
Target: rothaer_scaled
Distance: 1.0091% / 0.01009085
39.8 (Balto-)Slavic
39.0 Germanic
19.2 Celtic-like
1.8 Graeco-Roman
0.2 Finnic-like
Thumbs Up |
Received: 10,451 Given: 12,912 |
- All calculators web app: https://universe-apricity.github.io/...ors/index.html (work in progress)
- Thread for feedback: https://www.theapricity.com/forum/sh...lators-web-app
- My DNA test results: https://universe-apricity.github.io/.../universe.html
Thumbs Up |
Received: 1,794 Given: 831 |
Thumbs Up |
Received: 10,451 Given: 12,912 |
- All calculators web app: https://universe-apricity.github.io/...ors/index.html (work in progress)
- Thread for feedback: https://www.theapricity.com/forum/sh...lators-web-app
- My DNA test results: https://universe-apricity.github.io/.../universe.html
Thumbs Up |
Received: 11,047 Given: 26,801 |
yes, it leads to overfit apparently. some guys with background in math could explain this better than me. not just for vahaduo, but too many components will give overfit in any model, like using sets of G25 coords to model a population for example, G25 has too many components as well.
https://en.wikipedia.org/wiki/Overfitting
There are currently 1 users browsing this thread. (0 members and 1 guests)
Bookmarks