Davidski's G25 components

[Vishap]

What are the components of Davidski's G25 calculator, in order?

[Lucas]

What are the components of Davidski's G25 calculator, in order?

There aren't components. Those are PCA values.

[gixajo]

What are the components of Davidski's G25 calculator, in order?

I want to know it too...

[Nurzat]

There aren't components. Those are PCA values.

does it mean anything if you have identical coordinates with an individual or population? for example PC2.

also, up to which PC are they relevant? the first ones capture most of the differences, right?

also, since PCA plots only use PC1 and PC2, how much is lost? do people give too much credit to PCA plots (because they only have PC1 and PC2, in the end)?

[gixajo]

There aren't components. Those are PCA values.

But each coordinate is related to one ethnic group or origin, isn´t it?

And we don´t know exactly to wich origin is related every coordinate, although I have some of them more or less identified.

Anyway, it´s ok don´t know exactly which origins is related every coordinate, so some people cannot change them in order to fake his results.(as it happens with K13 coordinates and others).

[gixajo]

does it mean anything if you have identical coordinates with an individual or population? for example PC2.

also, up to which PC are they relevant? the first ones capture most of the differences, right?

also, since PCA plots only use PC1 and PC2, how much is lost? do people give too much credit to PCA plots (because they only have PC1 and PC2, in the end)?

Every coordinate is a distance to an ethnicity/origin/component etc, some say that distances are measured in an n=25 multidimension PCA, but in my opinion distances are measured in a 2D or maybe3D projection of n=25 dimensions PCA.

[Morena]

I just read that he's taking out modern samples and just using ancient samples. I disagree with this. There should be a modern sample, even if people get mad. Even if some people may not match because of 'personal identification,' with enough samples, these false samples will be averaged out.

[Vishap]

There aren't components. Those are PCA values.

Okay but what do they refer to? What does, for example, 0.002476 after the second comma mean with regard to a component?

[Lucas]

Okay but what do they refer to? What does, for example, 0.002476 after the second comma mean with regard to a component?

As far as I know this is SmartPCA on raw dataset. https://github.com/chrchang/eigensoft/wiki/smartpca
Vbknethio did it for test (G30) and there is thread about it here.

But theoretically Davidski could make K25 calculator and could make PCA in PAST on it's oracle ( I did it for K36 but additionally make few other conversions to hide real components). Such method would be faster when he wanted to sell coordinates from obvious reasons...

But in such case principal components couldn't be just calculator components (check what is PCA on wikipedia or elswhere) and certainly not in that order like originally. Everyone can test and make PCA on K13 or K15 oracle in PAST for example. So still you can't infer from principal component what was original calc component for first column, or second and so on.

[Lucas]

also, up to which PC are they relevant? the first ones capture most of the differences, right?

also, since PCA plots only use PC1 and PC2, how much is lost? do people give too much credit to PCA plots (because they only have PC1 and PC2, in the end)?

In Vahaduo / NMonte nothing is lost. Calculation is on 25 PCs. Yes, first principal components are most important always. You can easily dismiss last 5 I think... Someone can make sheet with only 20 values and check.

But you lost 23 PCs when you make PCA plot. Or 22 if you make 3D plot.


does it mean anything if you have identical coordinates with an individual or population? for example PC2.

Interesting case, you have such example?

[Nurzat]

In Vahaduo / NMonte nothing is lost. Calculation is on 25 PCs. Yes, first principal components are most important always. You can easily dismiss last 5 I think... Someone can make sheet with only 20 values and check.

But you lost 23 PCs when you make PCA plot. Or 22 if you make 3D plot.



Interesting case, you have such example?



yes, I see three individual samples with same PC1 and PC2 coordinates as me (so we all 4 plot in the same point in the 2D graph) and I want to understand how relevant is it:

Sample,PC1,PC2
ISL_Viking_Age_Pre_Christian:SBT-A-1,0.125205,0.128972
SWE_Viking_Age_Sigtuna:vik_stg026,0.125205,0.12897 2
UKR_Chernyakhiv_Shyshaky:MJ37,0.125205,0.128972
Carpathian_Rusnak-West_Moldovan:Nurzat,0.125205,0.128972