search

amcdavid / HurdleNormal

Estimation and sampling for multi-dimensional Hurdle models on a Normal density with applications to single-cell co-expression
5 stars 5 forks source link

G, H, K #7

Open galaxywall opened 3 years ago

galaxywall commented 3 years ago

I am a little confused about how you set the matrices G, H, and K for the simulations in terms of the codes below? How did you pick the entry values, the diagonal and off-diagonal ones(e.g. -0.25, -0.4, 1), for G, H, and K in both cases?

Also, could you please give me a little more idea how can I draw a graph given G,H, and K? Could you lead me to the relevant codes?

modelList[[i]] <- HurdleNormal:::simulateHurdle210(N=thisN, thisP, c('G', 'Hupper', 'Hlower', 'K'), structure='chain', structureArgs=list(sparsity=thisSparsity), intensityArgs=list(G=-.25,K=-.4, Hupper=-.75, Hlower=-.75), Gdiag=-.5, Hdiag=1, gibbs_args=list(mvnorm=mvnorm, kcell=kcell, post_fun=contam))

modelList[[i]] <- HurdleNormal:::simulateHurdle210(N=thisN, thisP, c('G'), structure='chain', structureArgs=list(sparsity=thisSparsity), intensityArgs=list(G=1), Gdiag=-4.5, Hdiag=2, gibbs_args=list(mvnorm=mvnorm, kcell=kcell, post_fun=contam))
amcdavid commented 3 years ago

How? trial and error. Why? I suspect it’s a setting where the Gaussian model doesn’t fit well since there’s a positive interaction in the conditional variance but negative in “V”.

On Nov 19, 2020, at 4:39 PM, galaxywall notifications@github.com wrote:

 I am a little confused about how you set the matrices G, H, and K for the simulations in terms of the codes below? How did you pick the entry values for G, H, and K in both cases?

modelList[[i]] <- HurdleNormal:::simulateHurdle210(N=thisN, thisP, c('G', 'Hupper', 'Hlower', 'K'), structure='chain', structureArgs=list(sparsity=thisSparsity), intensityArgs=list(G=-.25,K=-.4, Hupper=-.75, Hlower=-.75), Gdiag=-.5, Hdiag=1, gibbs_args=list(mvnorm=mvnorm, kcell=kcell, post_fun=contam))

modelList[[i]] <- HurdleNormal:::simulateHurdle210(N=thisN, thisP, c('G'), structure='chain', structureArgs=list(sparsity=thisSparsity), intensityArgs=list(G=1), Gdiag=-4.5, Hdiag=2, gibbs_args=list(mvnorm=mvnorm, kcell=kcell, post_fun=contam)) — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or unsubscribe.

galaxywall commented 3 years ago

So is this a problematic setting? Could you elaborate a little bit about what is special about this setting? What did you mean by "V"?