German Shepherd Dog Forums

German Shepherd Dog Forums (http://www.germanshepherds.com/forum/)
-   Bloodlines & Pedigrees (http://www.germanshepherds.com/forum/bloodlines-pedigrees/)
-   -   Computer based statistical analysis of bloodlines (http://www.germanshepherds.com/forum/bloodlines-pedigrees/203834-computer-based-statistical-analysis-bloodlines.html)

hunterisgreat 01-15-2013 12:38 PM

Computer based statistical analysis of bloodlines
 
So it's fairly easy to get all the dogs and their lineage into a computer. hip/elbow data is largely available, and other tests like DM, coat genes, etc, are becoming commonplace. A computer, properly employed, could predict trends and recognize patterns in potential breedings a far more accurately than a human. After all, computers make most of the stock trades on all major exchanges with superior accuracy. I've been kicking around the idea of making such a system. Thoughts? Ideas for publicly available data sources?

jocoyn 01-15-2013 12:48 PM

One of the problems is.........a valid sampling of offspring is not performed (at least over here) and bad results are often not posted.

Isn't that what the ZW is supposed to do over in Germany for hips but there a greater percentage of dogs are submittted for screening (or are they)?

hunterisgreat 01-15-2013 01:18 PM

Quote:

Originally Posted by jocoyn (Post 2743378)
One of the problems is.........a valid sampling of offspring is not performed (at least over here) and bad results are often not posted.

Isn't that what the ZW is supposed to do over in Germany for hips but there a greater percentage of dogs are submittted for screening (or are they)?

True, however with the various tools out there in the computer world, you can quantify, for example, bad hips, good hips, excellent hips, and "no data"... which is to say, a statistical analysis for an excellent hip dog bred to a "no data" dog would be something along the lines of "excellent hips, 33% confidence"... a level of confidence to a given answer is part of the computational approach.

In other words, if out of the breeding of 2 dogs, where i know half the hips scores, whatever they may be, I can still draw meaningful conclusions that carry less confidence than if I knew every dog's hips

stealthq 01-15-2013 02:21 PM

You can't come up with a remotely accurate confidence level using just the skewed data without some kind of reasonable estimate for the distribution of values for the overall population. Otherwise, your residuals will be so high that you might as well not bother.

Perhaps there are more generic statistics reported by veterinary schools, etc, that you can find? That would give you a place to start for some health-related values.

wildo 01-15-2013 02:28 PM

Haven spoken to Lee (wolfstraum) on the phone about a pedigree in the past, my thought is that the data she knew and used to decipher the pedigree is simply not available via queryable data sources. What data DOES live out there (like hip scores and some titles) may be helpful in generating an entry point for analysis, but almost certainly not to the level that one would be able to setup expectations on potential breedings. It appears to me that such info comes from years of first (or perhaps second) hand experience, and is stored in brains only, not silicon.

[EDIT]- I would say that some dogs have been discussed at length via forums like these and PDB. But in such cases, the data is not easily queried since it is in paragraph/sentence form. Sure- it lives in a DB somewhere, but the software would have to be amazingly good at contextual recognition. Not likely unless you can finance Watson.

hunterisgreat 01-15-2013 03:17 PM

Quote:

Originally Posted by stealthq (Post 2744138)
You can't come up with a remotely accurate confidence level using just the skewed data without some kind of reasonable estimate for the distribution of values for the overall population. Otherwise, your residuals will be so high that you might as well not bother.

Perhaps there are more generic statistics reported by veterinary schools, etc, that you can find? That would give you a place to start for some health-related values.

You could make more accurate predictions than a human could, with the same data (this is demonstrated to be true in *many* applications).

hunterisgreat 01-15-2013 03:19 PM

Quote:

Originally Posted by wildo (Post 2744234)
Haven spoken to Lee (wolfstraum) on the phone about a pedigree in the past, my thought is that the data she knew and used to decipher the pedigree is simply not available via queryable data sources. What data DOES live out there (like hip scores and some titles) may be helpful in generating an entry point for analysis, but almost certainly not to the level that one would be able to setup expectations on potential breedings. It appears to me that such info comes from years of first (or perhaps second) hand experience, and is stored in brains only, not silicon.

[EDIT]- I would say that some dogs have been discussed at length via forums like these and PDB. But in such cases, the data is not easily queried since it is in paragraph/sentence form. Sure- it lives in a DB somewhere, but the software would have to be amazingly good at contextual recognition. Not likely unless you can finance Watson.

Right, I wouldn't have access to someone's personal experiences and opinions on given dogs... but should dogs be accurately assessed via some sort of score sheet, such data could be used

jocoyn 01-15-2013 03:30 PM

What percentage of dogs do you think would have to have that data generated in order to be a viable tool? I believe it is very very low.

Liesje 01-15-2013 03:36 PM

There is a "breed betterment registry" or whatever it's called but so few people use it.

hunterisgreat 01-15-2013 04:10 PM

Quote:

Originally Posted by jocoyn (Post 2744802)
What percentage of dogs do you think would have to have that data generated in order to be a viable tool? I believe it is very very low.

The short answer is, the more the data, the better the results.


All times are GMT -4. The time now is 08:25 AM.

Powered by vBulletin® Copyright ©2000 - 2014, Jelsoft Enterprises Ltd.
SEO by vBSEO 3.3.2