ApolloChoiceModelling forum

I am getting an error showing " initial value of vmmin is not finite" in the constant-only binary choice model. I have tried updating my software and putting Hessian = FALSE but still showing the same error.

Hi

it's impossible to help you with this without seeing the code, so could you please post the code, along with the output you see

Thanks

Stephane

I forgot to attach the output so I am attaching my console window here

------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
rm(list=ls())
>
> setwd("D:\\PROJECT\\MNL_Analysis") ## set your working directory
>
> getwd() # check whether the working directory you have set is correct; make sure that the code and dataset are in this directory
[1] "D:/PROJECT/MNL_Analysis"
>
> #---------------------------------- install required packages and load the corresponding libraries --------------------------#
>
> #install.packages("reshape") # you need not install a package again if already installed
> library(reshape) # load the corresponding library you require
> #install.packages("optimx")
> library(optimx)
>
>
>
> #------------------------------------------- reading in data ---------------------------------------------------------------#
> Data = read.csv("fourWheelerFaceToFace_V1(coded)_working.csv") # read the data file: your data must be in csv formatto use this command
>
>
> Dmat = as.matrix(Data) # convert the data file into a matrix
>
>
> #number of rows in the data; no. of rows excluding the header: this is the no. of individuals in a case-format data file
> NCS = nrow(Dmat)
>
> # specify the number of choice alternatives
> numuoalt = 2; #can be greater than or equal to 2
>
>
> #------------------------------------ specification of dependent variables ---------------------------------------------------#
> ### extract the columns from your data; be sure that the column number is correct, and in order of the utility you specify below
>
> IDDEP = as.matrix(Dmat[,c(3,4)])
>
>
> #------------------------------------ specification of utlities --------------------------------------------------------------#
> ## please be sure to extract the correct column from the data file using the column number
> ## define for the number of alts. in the model
>
>
> alt1_x = as.matrix(Dmat[,c(5)])
> at1 = t(alt1_x)
>
> alt2_x = as.matrix(Dmat[,c(6)])
> at2 = t(alt2_x)
> #------------------------------------ create a matrix of all explanatory variables based on your utility specification above --------------------------------------------------------------#
>
>
> alt_x = as.matrix(cbind(at1,at2))
>
>
> #-----------------------: specify a vector of starting values: here, all starting values have been kept as 0
> #-----------------------: the number of elements here should be equal to the number of parameters to be estimated
>
> nvar = ncol(at1)
> startv =matrix(0,1,1)
>
>
> #-------------------------------------- loglikelihood function: this is generic; does not depend on number of choice alt -----------------------------------------------------------#
>
> logit.lf <- function(beta) {
+ p = matrix(0, nrow=NCS, ncol=1) ### create a vector for choice prob
+ v1 = matrix(0, NCS, numuoalt) ### matrix of utilities: size NCS*numuoalt
+ v1 = t(beta)%*% alt_x ### compute V = beta*X
+ v1 = matrix(v1, nrow = numuoalt, byrow = TRUE) ### we need an NCS*1 row vector for each alternative
+ v1 = t(v1)
+ v1=exp(v1) ### calculate exp(V)
+ den = as.matrix(rowSums(v1)) ### denominator of logit choice prob.
+ vchoice = v1*IDDEP ### deterministic comp. for chosen alt. only: numerator in logit choice prob.
+ p = rowSums(vchoice)/den ### choice prob.
+ logl = -sum(log(p)) ### calculate log-likelihood
+ return(logl)
+ }
>
>
> #--------------------------------------------------- Model estimation --------------------------------------------------------------#
>
> logitmodel <- optim(startv, logit.lf, method="BFGS", control=list(trace=TRUE, REPORT=1),hessian=TRUE) # calling an R routine for optimization
Error in optim(startv, logit.lf, method = "BFGS", control = list(trace = TRUE, :
initial value in 'vmmin' is not finite
>

------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

This is not using Apollo, so not for this forum