search

shokoohi / fmrs

fmrs package provides estimation and variable selection in Finite Mixture of Accelerated Failure Time Regression (FMAFTR) and Finite Mixture of Regression (FMR) models.
3 stars 2 forks source link

why the estimate of the mixprop is so different from the true value? #2

Open 123xiao4 opened 2 years ago

123xiao4 commented 2 years ago

I'm confused that the estimate of the mixprop is so different from the true value. It looks like the order of the esimate of the mixprop is reversed. In the following code, the true value of mixprop is 0.3 and 0.7, but the estimate is 0.6896571 and 0.3103429. ` nComp = 2 nCov = 10 nObs = 500 dispersion = c(1, 1) mixProp = c(0.3, 0.7) rho = 0.5 coeff1 = c(1,1,0,0,3,0,1,0.6, 0, 3, 0) coeff2 = c(2,-1,2,0,0,3,-1, 0, 0, 4, .7) umax = 40

dat <- fmrs.gendata(nObs = nObs, nComp = nComp, nCov = nCov, coeff = c(coeff1, coeff2), dispersion = dispersion, mixProp = mixProp, rho = rho, umax = umax, disFamily = "norm") res.mle <- fmrs.mle(y = dat$y, x = dat$x, delta = dat$delta, nComp = nComp, disFamily = "norm", initCoeff = rnorm(nComp*nCov+nComp), initDispersion = rep(1, nComp), initmixProp = rep(1/nComp, nComp))

Mixlasso

res.lam1 <- fmrs.tunsel(y = dat$y, x = dat$x, delta = dat$delta, nComp = nComp, disFamily = "norm", initCoeff = c(coefficients(res.mle)), initDispersion = dispersion(res.mle), initmixProp = mixProp(res.mle), penFamily = "lasso") res.var1 <- fmrs.varsel(y = dat$y, x = dat$x, delta = dat$delta, nComp = ncomp(res.mle), disFamily = "norm", initCoeff=c(coefficients(res.mle)), initDispersion = dispersion(res.mle), initmixProp = mixProp(res.mle), penFamily = "lasso", lambPen = slot(res.lam1, "lambPen")) mixProp(res.var1) Comp.1 Comp.2 [1,] 0.6896571 0.3103429

`

shokoohi commented 1 year ago

Hello, It is a well-known fact that FMR is identifiable up to a permutation. What you see is that the place of Component 1 and Component 2 is exchanged which is common in FMR. If you want to get the correct order, try to use different initial values for mixing proportions, say c(0.2, 0.8). This will result in correct order in the fit.