非结构化模型由于协方差参数众多，且在REML过程中很可能获得非满秩协方差矩阵，因此往往存在收敛问题和奇异性。您可能需要尝试降秩因子分析模型，以保证收敛并提高速度。常规US模型如下所示

library(sommer)
data(DT_example)
DT <- DT_example
A <- A_example
head(DT)
DT=DT[with(DT, order(Env)), ]
ans2us <- mmec(Yield~Env,
             random= ~ vsc(usc(Env),isc(Name)),
             rcov= ~ vsc(dsc(Env),isc(units)),
             data=DT)
plot(ans2us$llik[1,])
summary(ans2us)$varcomp
pp <- predict(ans2us, D="Name")
pp$pvals

FA实现如下所示

# less accurate but is guaranteed to converge
M <- with(DT, rrc(Env, Name, Yield, nPC=2))
ans2fa <- mmec(Yield~Env,
              random= ~ vsc(usc( M ),isc(Name)),
              rcov= ~ vsc(dsc(Env),isc(units)),
              nIters = 100,
              data=DT)
summary(ans2fa)$varcomp
plot(ans2fa$llik[1,])
ppfa <- predict(ans2fa, D="Name")

Lam <- with(DT, rrc(Env, Name, Yield, nPC=2, returnLam = TRUE))$Lam
scores <- ans2fa$uList[[1]]
BLUP <- scores %*% t(Lam) # environment specific predictions

# highly correlated predictions
plot(pp$pvals$predicted.value, ppfa$pvals$predicted.value)

干杯