happycow's G25 results

[Abdelnour]

Pretty cool results.

Toppo, I will get with you to get my G25 results. I can pay tomorrow. I am curious on mines.

[Kamal900]

The results he gave you. Check post #28 on this thread.

Well, he only gave me the coordinates.

[Abdelnour]

Well, he only gave me the coordinates.

Cool. I have to figure out how to get the coordinates.

[Kamal900]

Not really. Its normal for calcs comprising of ancient components to get such large fits.

Also, fits below 1-1.5 are overfitting and considered bad as well.

Really? I mean, isn't fits that are below 1 considered to be very accurate?

I used unscaled coordinates using only modern samples for Palestinian Muslims:

"sample": "Test1alestinian",
"fit": 0.7489,
"Lebanese_Christian": 63.33,
"Egyptian": 17.5,
"BedouinB": 15,
"Dinka": 4.17,

Isn't the results accurate? Here's using ancient samples:

"sample": "Test1alestinian",
"fit": 0.761,
"Levant_Canaanite_MBA": 58.33,
"Levant_JOR_EBA": 19.17,
"IRN_Hajji_Firuz_BA": 13.33,
"Canary_Islands_Guanche": 5,
"TZA_Pemba_600BP": 4.17,

Here's the results for the Palestinian Muslims using scaled ancient coordinates:

"sample": "Test1alestinian",
"fit": 1.5156,
"Levant_Canaanite_MBA": 56.67,
"Levant_JOR_EBA": 17.5,
"Canary_Islands_Guanche": 10.83,
"TKM_IA": 6.67,
"IRN_Hajji_Firuz_BA": 5.83,
"TZA_Pemba_600BP": 2.5,

[NPKTO]

Really? I mean, isn't fits that are below 1 considered to be very accurate?

I used unscaled coordinates using only modern samples for Palestinian Muslims:

"sample": "Test1alestinian",
"fit": 0.7489,
"Lebanese_Christian": 63.33,
"Egyptian": 17.5,
"BedouinB": 15,
"Dinka": 4.17,

Isn't the results accurate? Here's using ancient samples:

"sample": "Test1alestinian",
"fit": 0.761,
"Levant_Canaanite_MBA": 58.33,
"Levant_JOR_EBA": 19.17,
"IRN_Hajji_Firuz_BA": 13.33,
"Canary_Islands_Guanche": 5,
"TZA_Pemba_600BP": 4.17,

Here's the results for the Palestinian Muslims using scaled ancient coordinates:

"sample": "Test1alestinian",
"fit": 1.5156,
"Levant_Canaanite_MBA": 56.67,
"Levant_JOR_EBA": 17.5,
"Canary_Islands_Guanche": 10.83,
"TKM_IA": 6.67,
"IRN_Hajji_Firuz_BA": 5.83,
"TZA_Pemba_600BP": 2.5,

If you use overlapping components, there will definitely be overfitting. You get lower fits, but the result is overfitted. For example, to model SW Asians like Palestinians using ancients, I would use Levant_N, Anatolia_N, Iran_Neolithic, CHG, and Iberomaurisian. There's some similarities between Anatolia_N and Levant_N. But these are less overlapping because Anatolia_N is significantly drifted towards WHG compared to Levant_N. I guess you could use similar idea to model with bronze age and iron age pops maybe?

[Kamal900]

If you use overlapping components, there will definitely be overfitting. You get lower fits, but the result is overfitted. For example, to model SW Asians like Palestinians using ancients, I would use Levant_N, Anatolia_N, Iran_Neolithic, CHG, and Iberomaurisian. There's some similarities between Anatolia_N and Levant_N. But these are less overlapping because Anatolia_N is significantly drifted towards WHG compared to Levant_N. I guess you could use similar idea to model with bronze age and iron age pops maybe?

What's levant_N? Do you mean Levantine Natufian or Canaanite from Syria?

[NPKTO]

What's levant_N? Do you mean Levantine Natufian or Canaanite from Syria?

It was Levantine Neolithic population. You could also replace it with Levant_PPNB (Pre Pottery Neolithic Culture B).

[Kamal900]

It was Levantine Neolithic population. You could also replace it with Levant_PPNB (Pre Pottery Neolithic Culture B).

What about CHG since it's not found in the spreadsheets?

[Mingle]

Really? I mean, isn't fits that are below 1 considered to be very accurate?

I used unscaled coordinates using only modern samples for Palestinian Muslims:

"sample": "Test1alestinian",
"fit": 0.7489,
"Lebanese_Christian": 63.33,
"Egyptian": 17.5,
"BedouinB": 15,
"Dinka": 4.17,

Isn't the results accurate? Here's using ancient samples:

"sample": "Test1alestinian",
"fit": 0.761,
"Levant_Canaanite_MBA": 58.33,
"Levant_JOR_EBA": 19.17,
"IRN_Hajji_Firuz_BA": 13.33,
"Canary_Islands_Guanche": 5,
"TZA_Pemba_600BP": 4.17,

Here's the results for the Palestinian Muslims using scaled ancient coordinates:

"sample": "Test1alestinian",
"fit": 1.5156,
"Levant_Canaanite_MBA": 56.67,
"Levant_JOR_EBA": 17.5,
"Canary_Islands_Guanche": 10.83,
"TKM_IA": 6.67,
"IRN_Hajji_Firuz_BA": 5.83,
"TZA_Pemba_600BP": 2.5,

If you use ancient components and you get a low fit, I'd say its pretty good. Overfitting is when genetically similar components steal from each other.

[Kamal900]

If you use ancient components and you get a low fit, I'd say its pretty good. Overfitting is when genetically similar components steal from each other.

I see. So, over-fitting is bad when using ancient samples but it's good when using modern samples like in the first results, right?