search

bvilhjal / ldpred

MIT License
95 stars 58 forks source link

LDpred rho values #90

Closed Minta821 closed 4 years ago

Minta821 commented 5 years ago

Hi Bjarni J. Vilhja´lmsson, While I was working on the ldpred with different rho values ( the sample size is 125000), ldpred works well on many rho values, but when it reaches rho 0.03 (even it is converged), it shows a sudden drop in AUC... Again it improves the AUC and offers the best AUC at rho 0.01. But with the ldpred-inf model, it offer almost the same performance of rho 0.01. Why do this pattern happens in the AUC with different rho values? Is it due to any convergence problem which can be changed by adjusting any parameters in MCMC procedures? Do the ldpred-inf model using any MCMC process? Regards, Thomas

wavefancy commented 5 years ago

It seems that the large N problem still needs to be fixed, having the same problem with a large sample size.

bvilhjal commented 4 years ago

I believe I've improved this now. Please try it again..