ApolloChoiceModelling forum

Hi Stephane and David,

I am having a challenging situation for some time now regarding my model and your assistance would be of great help.
I have my model set up to the best of my knowledge and I think everything runs pretty well up to the estimation stage.

That is at the stage where I have to run this code
model=apollo_estimate(apollo_beta,
apollo_fixed,
apollo_probabilities,
apollo_inputs)

Anytime I run the above code, I get the error message:
Error in apollo_estimate(apollo_beta, apollo_fixed, apollo_probabilities, :
Log-likelihood calculation fails at values close to the starting values!
I have worked around the problem through the helpful resource "http://www.apollochoicemodelling.com/faq.html" and have tried the possible suggested solutions, i.e. "using the apollo_searchStart function", changing the initial value on the log-parameter to -3. and all else zero. I still have the error.

Below are the codes I used.

Thank you for your assistance.
Frederick
********************************************************************************************************************************************************
********************************************************************************************************************************************************
********************************************
###Code I have so far ###
********************************************


rm(list = ls())

### Set Directory ###
setwd("")

### Load Apollo ###
library(tidyverse)
library(magrittr)
library(kableExtra)
library(apollo)


#Start by initializing the code and setting up core controls
##################################################

### Initialize ###
apollo_initialise()

### Set core controls ###
apollo_control= list(
modelName ="MNL_basicmodel_fully_correlated_600",
modelDescr = "Basic MMNL model with fully correlated random coefficient in Preference Space - Max Likelihood 600 MLHS Draws",
indivID = "respondent_id",
outputDirectory = "C:/Users/fnyanzu2/Box/A_ILLINOIS FOLDER/Amy Ando/A_Survey Analysis summer 2022",
mixing = TRUE,
nCores = 6,
panelData = TRUE,
seed = 1995
)


# ##########################################################
### LOAD DATA AND APPLY ANY TRANSFORMATIONS ###
# #########################################################

### Loading data
database = read.csv(".csv", header=TRUE)

### The betta values here uses the last estimate from the apollo_searchStart ###

apollo_beta=c(mu_asc = 0.9440,
sigma_asc = -0.2631,
mu_log_cost = -3.000,
sigma_asc_log_cost = 0.0000,
sigma_log_cost = 0.0000,
mu_nature = 0.0073,
sigma_asc_nature = -0.0082,
sigma_log_cost_nature = 0.0030,
sigma_nature = 0.0075,
mu_farmland = 0.0019,
sigma_asc_farmland = 0.0037,
sigma_log_cost_farmland = -0.0089,
sigma_nature_farmland = 0.0049,
sigma_farmland = -0.0036,
mu_meals_nature = 0.1057,
sigma_asc_meals_nature = -0.0859,
sigma_log_cost_meals_nature = 0.0236,
sigma_nature_meals_nature = 0.0491,
sigma_farmland_meals_nature = 0.0321,
sigma_meals_nature = -0.0160,
mu_meals_farmland = 0.0000,
sigma_asc_meals_farmland = 0.0332,
sigma_log_cost_meals_farmland = 0.1461,
sigma_nature_meals_farmland = -0.1116,
sigma_farmland_meals_farmland = -0.0598,
sigma_meals_nature_meals_farmland = 0.0484,
sigma_meals_farmland = 0.1235,
mu_access_MT = 0.4781,
sigma_asc_access_MT = -0.2186,
sigma_log_cost_access_MT = -0.0342,
sigma_nature_access_MT = -0.1143,
sigma_farmland_access_MT = 0.0120,
sigma_meals_nature_access_MT = -0.0699,
sigma_meals_farmland_access_MT = -0.0196,
sigma_access_MT = 0.0014,
mu_access_PT = -0.0841,
sigma_asc_access_PT = 0.0628,
sigma_log_cost_access_PT = -0.0573,
sigma_nature_access_PT = 0.2299,
sigma_farmland_access_PT = 0.0569,
sigma_meals_nature_access_PT = 0.0672,
sigma_meals_farmland_access_PT = -0.0035,
sigma_access_MT_access_PT = 0.0081,
sigma_access_PT = 0.0030,
mu_access_TT = 0.7235,
sigma_asc_access_TT = 0.0000,
sigma_log_cost_access_TT = 0.0139,
sigma_nature_access_TT = -0.3151,
sigma_farmland_access_TT = 0.0436,
sigma_meals_nature_access_TT = -0.0513,
sigma_meals_farmland_access_TT = -0.0775,
sigma_access_MT_access_TT = 0.0461,
sigma_access_PT_access_TT = 0.0247,
sigma_access_TT = 0.0082,
mu_distance = -0.0403,
sigma_asc_distance = 0.0538,
sigma_log_cost_distance = 0.0020,
sigma_nature_distance = 0.0254,
sigma_farmland_distance = -0.0272,
sigma_meals_nature_distance = 0.0315,
sigma_meals_farmland_distance = -0.0018,
sigma_access_MT_distance = -0.0041,
sigma_access_PT_distance = 0.0192,
sigma_access_TT_distance = -0.0102,
sigma_distance = 0.0005
)

apollo_fixed=c()

apollo_draws=list(
interDrawsType = "halton",
interNDraws = 500,
interNormDraws = c("draws_asc","draws_log_cost","draws_nature",
"draws_farmland","draws_meals_nature","draws_meals_farmland",
"draws_access_MT","draws_access_PT","draws_access_TT",
"draws_distance")
)



apollo_randCoeff=function(apollo_beta,apollo_inputs){
randcoeff=list()
randcoeff[["b_asc"]] = (mu_asc + sigma_asc*draws_asc)
randcoeff[["b_cost"]] = exp(mu_log_cost + sigma_asc_log_cost * draws_asc + sigma_log_cost * draws_log_cost)
randcoeff[["b_nature"]] = (mu_nature + sigma_asc_nature * draws_asc + sigma_log_cost_nature * draws_log_cost + sigma_nature * draws_nature)
randcoeff[["b_farmland"]] = (mu_farmland + sigma_asc_farmland * draws_asc + sigma_log_cost_farmland * draws_log_cost + sigma_nature_farmland * draws_nature + sigma_farmland * draws_farmland)
randcoeff[["b_meals_nature"]] = (mu_meals_nature + sigma_asc_meals_nature * draws_asc + sigma_log_cost_meals_nature * draws_log_cost + sigma_nature_meals_nature * draws_nature + sigma_farmland_meals_nature * draws_farmland + sigma_meals_nature * draws_meals_nature)
randcoeff[["b_meals_farmland"]] = (mu_meals_farmland + sigma_asc_meals_farmland * draws_asc + sigma_log_cost_meals_farmland * draws_log_cost + sigma_nature_meals_farmland * draws_nature + sigma_farmland_meals_farmland * draws_farmland + sigma_meals_nature_meals_farmland * draws_meals_nature + sigma_meals_farmland * draws_meals_farmland)
randcoeff[["b_access_MT"]] = (mu_access_MT + sigma_asc_access_MT * draws_asc + sigma_log_cost_access_MT * draws_log_cost + sigma_nature_access_MT * draws_nature + sigma_farmland_access_MT * draws_farmland + sigma_meals_nature_access_MT * draws_meals_nature + sigma_meals_farmland_access_MT * draws_meals_farmland + sigma_access_MT * draws_access_MT)
randcoeff[["b_access_PT"]] = (mu_access_PT + sigma_asc_access_PT * draws_asc + sigma_log_cost_access_PT * draws_log_cost + sigma_nature_access_PT * draws_nature + sigma_farmland_access_PT * draws_farmland + sigma_meals_nature_access_PT * draws_meals_nature + sigma_meals_farmland_access_PT * draws_meals_farmland + sigma_access_MT_access_PT * draws_access_MT + sigma_access_PT * draws_access_PT)
randcoeff[["b_access_TT"]] = (mu_access_TT + sigma_asc_access_TT * draws_asc + sigma_log_cost_access_TT * draws_log_cost + sigma_nature_access_TT * draws_nature + sigma_farmland_access_TT * draws_farmland + sigma_meals_nature_access_TT * draws_meals_nature + sigma_meals_farmland_access_TT * draws_meals_farmland + sigma_access_MT_access_TT * draws_access_MT + sigma_access_PT_access_TT * draws_access_PT + sigma_access_TT * draws_access_TT)
randcoeff[["b_distance"]] = (mu_distance + sigma_asc_distance * draws_asc + sigma_log_cost_distance * draws_log_cost + sigma_nature_distance * draws_nature + sigma_farmland_distance * draws_farmland + sigma_meals_nature_distance * draws_meals_nature + sigma_meals_farmland_distance * draws_meals_farmland + sigma_access_MT_distance * draws_access_MT + sigma_access_PT_distance * draws_access_PT + sigma_access_TT_distance * draws_access_TT + sigma_distance * draws_distance)

return(randcoeff)
}

apollo_inputs=apollo_validateInputs()

apollo_probabilities=function(apollo_beta, apollo_inputs, functionality="estimate"){


apollo_attach(apollo_beta, apollo_inputs)
on.exit(apollo_detach(apollo_beta, apollo_inputs))
P = list()


## ### Alternative Utilities or list of utilities: these must use the same names as in mnl_settings, order is irrelevant
V = list()
V[["prj"]] = b_asc * asc1 + b_nature * nature1 + b_farmland * farmland1 + b_meals_nature * meals_nature1 + b_meals_farmland * meals_farmland1 + b_access_MT * access_MT1 + b_access_PT * access_PT1 + b_access_TT * access_TT1 + b_distance * distance1 + b_cost + cost1
V[["statusquo"]] = b_asc * asc2 + b_nature * nature2 + b_farmland * farmland2 + b_meals_nature * meals_nature2 + b_meals_farmland * meals_farmland2 + b_access_MT * access_MT2 + b_access_PT * access_PT2 + b_access_TT * access_TT2 + b_distance * distance2 + b_cost + cost2


### Define MNL Settings ###
mnl_settings = list(
alternatives = c(prj=1, statusquo=2),
choiceVar = choice,
utilities = V
)

### Computing probabilities using MNL Model ###
P[["model"]] = apollo_mnl(mnl_settings, functionality)

### Taking product across observation for the same individual ###
P = apollo_panelProd(P, apollo_inputs, functionality)

### Average Across Inter-individual Draws ###
P = apollo_avgInterDraws(P, apollo_inputs, functionality)

### Prepare and Return Outputs of Function ###
P = apollo_prepareProb(P, apollo_inputs, functionality)
return(P)
}


### Search for Starting Values ###
## Stop at 3 iterations ##
#searchStart_settings=list(maxStages = 3,
# nCandidates = 10)


#apollo_beta=apollo_searchStart(apollo_beta,
# apollo_fixed,
# apollo_probabilities,
# apollo_inputs,
# searchStart_settings)



model=apollo_estimate(apollo_beta,
apollo_fixed,
apollo_probabilities,
apollo_inputs)



********************************************
###output results###
*********************************************
Preparing user-defined functions.

Testing likelihood function...
Setting "avail" is missing, so full availability is assumed.

Overview of choices for MNL model component :
prj statusquo
Times available 4344.00 4344.00
Times chosen 2290.00 2054.00
Percentage chosen overall 52.72 47.28
Percentage chosen when available 52.72 47.28

Log-likelihood calculation fails at starting values!
Affected individuals:
ID LL
4 -Inf
13 -Inf
14 -Inf
...
...
...
830 -Inf
832 -Inf
834 -Inf
837 -Inf
Error in apollo_estimate(apollo_beta, apollo_fixed, apollo_probabilities, :
Log-likelihood calculation fails at values close to the starting values!

Hi

did you start off with a simple MNL model first? And when moving to the mixed model, did you look first at univariate distributions? What you have below is a very complex specification, and there are many reasons why it could go wrong and we need to try to isolate the issue

Stephane

Hi Stephane,

Thank you for your response.
I tried the simple MNL model first but with different data (I use the apollo_swissRouteChoiceData example provided). And all was ok. When I started off with this data, I get stuck with the error even with the simple model.
I think maybe I might be doing something wrong that I am not able to figure out.

Thank you.

Best,
Frederick

Hi

if it fails with a simple model, then that needs fixing first. Can you show us the code and output for that one

Thanks

Stephane

Hi Stephane,

Below is the output for the simple preference space model:

Thank you.
************************************************


>
> ## Clear memory
> rm(list = ls())
>
> ### Set Directory ###
>
> ### Load Apollo ###
> library(tidyverse)
> library(magrittr)
> library(kableExtra)
> library(apollo)
>
>
> #Start by initializing the code and setting up core controls
> ############################################################
>
> ### Initialize ###
> apollo_initialise()
Apollo ignition sequence completed
>
> ### Set core controls ###
> apollo_control= list(
+ modelName ="MNL_basicmodel_fully_correlated_600",
+ modelDescr = "Basic MMNL model in Preference Space - Max Likelihood 600 MLHS Draws",
+ indivID = "respondent_id",
+ outputDirectory = "",
+ mixing = TRUE,
+ nCores = 6,
+ panelData = TRUE,
+ seed = 1995
+ )
>
>
> # ##########################################################
> ### LOAD DATA AND APPLY ANY TRANSFORMATIONS ###
> # #########################################################
>
> ### Loading data
> database = read.csv("C:/Users/fnyanzu2/Box/A_ILLINOIS FOLDER/Amy Ando/A_Survey Analysis summer 2022/data_apollo1.csv", header=TRUE)
>
>
> #Determine the attributes of the model and define the priors on the model parameters.

> apollo_beta=c(mu_asc = 0.00,
+ sigma_asc = 0.000,
+ mu_log_cost = -3.00,
+ sigma_log_cost = -0.01,
+ mu_nature = 0.000,
+ sigma_nature = 0.000,
+ mu_farmland = 0.000,
+ sigma_farmland = 0.000,
+ mu_meals_nature = 0.000,
+ sigma_meals_nature = 0.000,
+ mu_meals_farmland = 0.000,
+ sigma_meals_farmland = 0.000,
+ mu_access_MT = 0.000,
+ sigma_access_MT = 0.000,
+ mu_access_PT = 0.000,
+ sigma_access_PT = 0.000,
+ mu_access_TT = 0.000,
+ sigma_access_TT = 0.000,
+ mu_distance = 0.000,
+ sigma_distance = 0.000
+ )
>
> apollo_fixed = c()
>
>
> ###########################################################
> # Setting parameters to define draws & Random parameters
> ##########################################################
>
> ### Draws
> apollo_draws=list(
+ interDrawsType = "mlhs",
+ interNDraws = 600,
+ interNormDraws = c("draws_asc","draws_log_cost","draws_nature",
+ "draws_farmland","draws_meals_nature","draws_meals_farmland",
+ "draws_access_MT","draws_access_PT","draws_access_TT",
+ "draws_distance")
+ )
>
>
> ### Random parameters
>
> apollo_randCoeff=function(apollo_beta,apollo_inputs){
+ randcoeff=list()
+ randcoeff[["b_asc"]] = (mu_asc + sigma_asc*draws_asc)
+ randcoeff[["b_cost"]] = -exp(mu_log_cost + sigma_log_cost * draws_log_cost)
+ randcoeff[["b_nature"]] = (mu_nature + sigma_nature * draws_nature)
+ randcoeff[["b_farmland"]] = (mu_farmland + sigma_farmland * draws_farmland)
+ randcoeff[["b_meals_nature"]] = (mu_meals_nature + sigma_meals_nature * draws_meals_nature)
+ randcoeff[["b_meals_farmland"]] = (mu_meals_farmland + sigma_meals_farmland * draws_meals_farmland)
+ randcoeff[["b_access_MT"]] = (mu_access_MT + sigma_access_MT * draws_access_MT)
+ randcoeff[["b_access_PT"]] = (mu_access_PT + sigma_access_PT * draws_access_PT)
+ randcoeff[["b_access_TT"]] = (mu_access_TT + sigma_access_TT * draws_access_TT)
+ randcoeff[["b_distance"]] = (mu_distance + sigma_distance * draws_distance)
+
+ return(randcoeff)
+ }
>
>
>
> ### Validating inputs
> apollo_inputs=apollo_validateInputs()
apollo_draws and apollo_randCoeff were found, so apollo_control$mixing was set to TRUE
All checks on apollo_control completed.
All checks on database completed.
Generating inter-individual draws .......... Done
>
>
> ### Likelihood function and model defining
> apollo_probabilities=function(apollo_beta, apollo_inputs, functionality="estimate"){
+
+ ### Function initialisation: do not change the following three commands ###
+ ### Attach inputs and detach after function exit ###
+ apollo_attach(apollo_beta, apollo_inputs)
+ on.exit(apollo_detach(apollo_beta, apollo_inputs))
+
+ ### Create list of probabilities P ###
+ P = list()
+
+
+ ## ### Alternative Utilities or list of utilities: these must use the same names as in mnl_settings, order is irrelevant
+ V = list()
+ V[["prj"]] = b_asc * asc1 + b_nature * nature1 + b_farmland * farmland1 + b_meals_nature * meals_nature1 + b_meals_farmland * meals_farmland1 + b_access_MT * access_MT1 + b_access_PT * access_PT1 + b_access_TT * access_TT1 + b_distance * distance1 + b_cost + cost1
+ V[["statusquo"]] = b_asc * asc2 + b_nature * nature2 + b_farmland * farmland2 + b_meals_nature * meals_nature2 + b_meals_farmland * meals_farmland2 + b_access_MT * access_MT2 + b_access_PT * access_PT2 + b_access_TT * access_TT2 + b_distance * distance2 + b_cost + cost2
+
+
+ ### Define MNL Settings ###
+ mnl_settings = list(
+ alternatives = c(prj=1, statusquo=2),
+ choiceVar = choice,
+ utilities = V
+ )
+
+ ### Compute Probabilities using MNL Model ###
+ P[["model"]] = apollo_mnl(mnl_settings, functionality)
+
+ ### Take Product Across Observation for Same Individual ###
+ P = apollo_panelProd(P, apollo_inputs, functionality)
+
+ ### Average Across Inter-individual Draws ###
+ P = apollo_avgInterDraws(P, apollo_inputs, functionality)
+
+ ### Prepare and Return Outputs of Function ###
+ P = apollo_prepareProb(P, apollo_inputs, functionality)
+ return(P)
+ }
> model = apollo_estimate(apollo_beta, apollo_fixed,apollo_probabilities, apollo_inputs)
Preparing user-defined functions.

Testing likelihood function...
Setting "avail" is missing, so full availability is assumed.

Overview of choices for MNL model component :
prj statusquo
Times available 4344.00 4344.00
Times chosen 2290.00 2054.00
Percentage chosen overall 52.72 47.28
Percentage chosen when available 52.72 47.28

Log-likelihood calculation fails at starting values!
Affected individuals:
ID LL
4 -Inf
13 -Inf
...
...
...
832 -Inf
834 -Inf
837 -Inf
Error in apollo_estimate(apollo_beta, apollo_fixed, apollo_probabilities, :
Log-likelihood calculation fails at values close to the starting values!
>

Hi

This is still mixed logit. Did the MNL work? Can you show us that?

Stephane

Dear Stephane,

No, the MNL as mentioned earlier, had similar error.
Please see below, the codes I used and the output as well.

Thank you.

*******************************************************************************************


## Clear memory
> rm(list = ls())
> #gc()
>
> ### Set Directory ###
> setwd("")
>
> ### Load Apollo ###
> library(tidyverse)
> library(magrittr)
> library(kableExtra)
> library(apollo)
>
>
> #Start by initializing the code and setting up core controls
> ############################################################
>
> ### Initialise code
> apollo_initialise()
Apollo ignition sequence completed
>
> ### Set core controls
> apollo_control = list(
+ modelName = "MNL_Farm_Nature",
+ modelDescr = "Simple MNL model on mode farmland value",
+ indivID = "respondent_id",
+ outputDirectory = "output"
+ )
>
>
> # ##########################################################
> ### LOAD DATA AND APPLY ANY TRANSFORMATIONS ###
> # #########################################################
>
> ### Loading data
> database = read.csv(" ## Clear memory
> rm(list = ls())
> #gc()
>
> ### Set Directory ###
> setwd("C:/Users/fnyanzu2/Box/A_ILLINOIS FOLDER/Amy Ando/A_Survey Analysis summer 2022")
>
> ### Load Apollo ###
> library(tidyverse)
> library(magrittr)
> library(kableExtra)
> library(apollo)
>
>
> #Start by initializing the code and setting up core controls
> ############################################################
>
> ### Initialise code
> apollo_initialise()
Apollo ignition sequence completed
>
> ### Set core controls
> apollo_control = list(
+ modelName = "MNL_Farm_Nature",
+ modelDescr = "Simple MNL model on mode farmland value",
+ indivID = "respondent_id",
+ outputDirectory = "output"
+ )
>
>
> # ##########################################################
> ### LOAD DATA AND APPLY ANY TRANSFORMATIONS ###
> # #########################################################
>
> ### Loading data
> database = read.csv("C:/Users/fnyanzu2/Box/A_ILLINOIS FOLDER/Amy Ando/A_Survey Analysis summer 2022/data_apollo1.csv", header=TRUE)
>
>
> # ################################################################# #
> #### DEFINE MODEL PARAMETERS ####
> # ################################################################# #
>
> ### Vector of parameters, including any that are kept fixed in estimation
> apollo_beta=c(b_asc = 0,
+ b_nature = 0,
+ b_farmland = 0,
+ b_meals_nature = 0,
+ b_meals_farmland = 0,
+ b_access_MT = 0,
+ b_access_PT = 0,
+ b_access_TT = 0,
+ b_distance = 0,
+ b_cost = 0)
>
> ### Vector with names (in quotes) of parameters to be kept fixed at their starting value in apollo_beta, use apollo_beta_fixed = c() if none
> apollo_fixed = c("b_cost")
> apollo_inputs = apollo_validateInputs()
Several observations per individual detected based on the value of respondent_id. Setting panelData in apollo_control set to TRUE.
All checks on apollo_control completed.
All checks on database completed.
> apollo_probabilities=function(apollo_beta, apollo_inputs, functionality="estimate"){
+
+ ### Attach inputs and detach after function exit
+ apollo_attach(apollo_beta, apollo_inputs)
+ on.exit(apollo_detach(apollo_beta, apollo_inputs))
+
+ ### Create list of probabilities P
+ P = list()
+
+
+
+
+ ## ### Alternative Utilities or list of utilities: these must use the same names as in mnl_settings, order is irrelevant
+ V = list()
+ V[["prj"]] = b_asc * asc1 + b_nature * nature1 + b_farmland * farmland1 + b_meals_nature * meals_nature1 + b_meals_farmland * meals_farmland1 + b_access_MT * access_MT1 + b_access_PT * access_PT1 + b_access_TT * access_TT1 + b_distance * distance1 + b_cost + cost1
+ V[["statusquo"]] = b_asc * asc2 + b_nature * nature2 + b_farmland * farmland2 + b_meals_nature * meals_nature2 + b_meals_farmland * meals_farmland2 + b_access_MT * access_MT2 + b_access_PT * access_PT2 + b_access_TT * access_TT2 + b_distance * distance2 + b_cost + cost2
+
+
+ ### Define MNL Settings ###
+ mnl_settings = list(
+ alternatives = c(prj=1, statusquo=2),
+ choiceVar = choice,
+ utilities = V
+ )
+
+
+
+ ### Compute probabilities using MNL model
+ P[["model"]] = apollo_mnl(mnl_settings, functionality)
+
+ ### Take product across observation for same individual
+ P = apollo_panelProd(P, apollo_inputs, functionality)
+
+ ### Prepare and return outputs of function
+ P = apollo_prepareProb(P, apollo_inputs, functionality)
+ return(P)
+ }
> model = apollo_estimate(apollo_beta, apollo_fixed, apollo_probabilities, apollo_inputs)
Preparing user-defined functions.

Testing likelihood function...
Setting "avail" is missing, so full availability is assumed.

Overview of choices for MNL model component :
prj statusquo
Times available 4344.00 4344.00
Times chosen 2290.00 2054.00
Percentage chosen overall 52.72 47.28
Percentage chosen when available 52.72 47.28

Log-likelihood calculation fails at starting values!
Affected individuals:
ID LL
4 -Inf
13 -Inf
14 -Inf
41 -Inf
...
...
...
828 -Inf
830 -Inf
832 -Inf
834 -Inf
837 -Inf

Error in apollo_estimate(apollo_beta, apollo_fixed, apollo_probabilities, :
Log-likelihood calculation fails at values close to the starting values!
>

I can look into this for you if you are able to share the code and data with me outside the forum

Hi

easy fix

You have
intead of
Same for the second alternative. Your uitlities then exploded for large values for cost

And why do you keep b_cost fixed?

Stephane

Oh! Thank you for that catch.
The cost was fixed for the mixlogit model. I should have deleted that.
Thank you!!!