tnagler / kdecopula Goto Github PK

Hello,

When using kdevine, I observed an odd effect of order of observations on density calculation. I traced it in a debugger to interp_2d function. Here is a pattern of behavior:

Browse[5]> u
             zr2       zr1
[1,] 0.001724833 0.0010000
[2,] 0.858983907 0.7091684

Browse[5]> interp_2d(u[1, , drop = F], obj$estimate, obj$grid, numeric(4), 
+           numeric(4))
[1] 161.8257
Browse[5]> interp_2d(u[2, , drop = F], obj$estimate, obj$grid, numeric(4), 
+           numeric(4))
[1] 1.141078

Browse[5]> interp_2d(u[c(1, 1), , drop = F], obj$estimate, obj$grid, numeric(4), 
+           numeric(4))
[1] 161.8257 161.8257
Browse[5]> interp_2d(u[c(2, 2), , drop = F], obj$estimate, obj$grid, numeric(4), 
+           numeric(4))
[1] 1.141078 1.141078
Browse[5]> interp_2d(u[c(1, 2), , drop = F], obj$estimate, obj$grid, numeric(4), 
+           numeric(4))
[1] 161.825726   1.141078
Browse[5]> interp_2d(u[c(2, 1), , drop = F], obj$estimate, obj$grid, numeric(4), 
+           numeric(4))
[1] 1.141078e+00 1.000588e-15

Note the effect of switching from (1, 2) to (2, 1). Overall, it appears that having two observations in a particular order can crash the function.

I glanced at the interp_2d function, but that leads to other functions, which may take me a while to figure out. Do you have any thoughts on how to stabilize the calculation?

Thanks,

Slava

hello，Thomas
I wonder if you have a copy of the EquilibriumModelsADMM.jl, I cannot get into connection with the developer，I saw you comments under the Project. Hope you have some hints

sorry for bothering you. thank you for your time and consideration.

Dear Nagler

How could I solve the AIC of the types TTCV and TTPI, please?

when I fit data and estimate it using these types it doesn't give the AIC values.

Regards