Error in LCNL model
Posted: 06 May 2023, 09:43
Dear Apollo team,
I'm estimating individuals' mode choice using LCNL (Latent Class Nested Logit) model. The model is made-up of 4 choice sets (bus, railway, highway and airline), each of which contains 4 continuous variables (cost, access time, journey time and frequency) with 3 levels and 1 dummy variable (seat comfort) with 3 levels. However, when estimating the model I run into the following warning message:
(b_seatAm= seat_comfort_average*male
b_seatGm= seat_comfort_good*male)
Warning message:
In sqrt(diag(varcov)) : NaNs produced
I am using the following code to estimate the model:
And, the outputs are as follows:
Do you have any idea what is happening and is there any way to resolve this issue?
Best regards, Chunyu Song
I'm estimating individuals' mode choice using LCNL (Latent Class Nested Logit) model. The model is made-up of 4 choice sets (bus, railway, highway and airline), each of which contains 4 continuous variables (cost, access time, journey time and frequency) with 3 levels and 1 dummy variable (seat comfort) with 3 levels. However, when estimating the model I run into the following warning message:
(b_seatAm= seat_comfort_average*male
b_seatGm= seat_comfort_good*male)
Warning message:
In sqrt(diag(varcov)) : NaNs produced
I am using the following code to estimate the model:
Code: Select all
### Clear memory
rm(list = ls())
### Load Apollo library
library(apollo)
### Initialise code
apollo_initialise()
apollo_control = list(
modelName ="Apollo_example_20_LC-NL-SCY",
modelDescr ="LC model with class allocation model",
indivID ="ID",
nCores = 6
)
database = read.csv("D:\\QDU\\zhoudao\\Apollo\\mode choice-Apollo\\Apollo-mode choice-SCY.csv",header = TRUE)
database = subset(database,database$SP==1)
## scaled variables
apollo_beta = c(asc_bus_c1 = 0,
asc_train_c1 = 0,
asc_high_c1 = 0,
asc_air_c1 = 0,
b_cost_c1 = 0,
b_acct_c1 = 0,
b_jourt_c1 = 0,
b_freq_c1 = 0,
b_seat_P_c1 = 0,
b_seat_A_c1 = 0,
b_seat_G_c1 = 0,
lambda_A_c1 = 1,
lambda_B_c1 = 0.5,
lambda_C_c1 = 1,
delta_c1 = 0,
asc_bus_c2 = 0,
asc_train_c2 = 0,
asc_high_c2 = 0,
asc_air_c2 = 0,
b_cost_c2 = 0,
b_acct_c2 = 0,
b_jourt_c2 = 0,
b_freq_c2 = 0,
b_seat_P_c2 = 0,
b_seat_A_c2 = 0,
b_seat_G_c2 = 0,
lambda_A_c2 = 1,
lambda_B_c2 = 0.5,
lambda_C_c2 = 1,
delta_c2 = 0,
b_seatAm_c1 = 0,
b_seatGm_c1 = 0,
b_seatAm_c2 = 0,
b_seatGm_c2 = 0
)
apollo_fixed = c("asc_bus_c1","asc_bus_c2","delta_c2","lambda_A_c1","lambda_A_c2","lambda_C_c1","lambda_C_c2","b_seat_P_c1","b_seat_P_c2")
## Define latent class components
apollo_lcPars = function(apollo_beta, apollo_inputs){
lcpars = list()
lcpars[["b_cost"]] = list(b_cost_c1,b_cost_c2)
lcpars[["b_acct"]] = list(b_acct_c1,b_acct_c2)
lcpars[["b_jourt"]] = list(b_jourt_c1,b_jourt_c2)
lcpars[["b_freq"]] = list(b_freq_c1, b_freq_c2)
lcpars[["b_seat_P"]] = list(b_seat_P_c1,b_seat_P_c2)
lcpars[["b_seat_A"]] = list(b_seat_A_c1,b_seat_A_c2)
lcpars[["b_seat_G"]] = list(b_seat_G_c1,b_seat_G_c2)
lcpars[["asc_train"]] = list(asc_train_c1,asc_train_c2)
lcpars[["asc_high"]] = list(asc_high_c1,asc_high_c2)
lcpars[["asc_air"]] = list(asc_air_c1,asc_air_c2)
lcpars[["b_seatAm"]] = list (b_seatAm_c1,b_seatAm_c2)
lcpars[["b_seatGm"]] = list(b_seatGm_c1,b_seatGm_c2)
#Utilities of class allocation model
V = list()
V[["class_1"]] = delta_c1
V[["class_2"]] = delta_c2
classAlloc_settings = list(
classes = c(class_1=1, class_2=2),
utilities = V
)
lcpars[["pi_values"]] = apollo_classAlloc(classAlloc_settings)
return(lcpars)
}
### Group and validate inputs
apollo_inputs = apollo_validateInputs()
### Define model and likelihood function
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()
### Specify nests for NL model
nlNests = list()
nlNests[[1]] = list(root=1, A_c1=lambda_A_c1, B_c1 = lambda_B_c1, C_c1 = lambda_C_c1)
nlNests[[2]] = list(root=1,A_c2=lambda_A_c2, B_c2 = lambda_B_c2, C_c2 = lambda_C_c2)
nlStructure = list()
nlStructure[[1]] = list()
nlStructure[[2]] = list()
nlStructure[[1]][["root"]] = c("A_c1","B_c1","C_c1")
nlStructure[[2]][["root"]] = c("A_c2","B_c2","C_c2")
nlStructure[[1]][["A_c1"]] = c("bus")
nlStructure[[1]][["B_c1"]] = c("train","high")
nlStructure[[1]][["C_c1"]] = c("air")
nlStructure[[2]][["A_c2"]] = c("bus")
nlStructure[[2]][["B_c2"]] = c("train","high")
nlStructure[[2]][["C_c2"]] = c("air")
nl_settings <- list(
alternatives = c(bus=1, train=2, high=3, air=4),
avail = list(bus=av_bus, train=av_train, high=av_high, air=av_air),
choiceVar = choice,
nlNests = nlNests,
nlStructure = nlStructure
)
### Loop over classes
s=1
while(s<=2){
### Compute class-specific utilities
V=list()
V[['bus']] = b_cost[[s]] * cost_bus + b_acct[[s]] * acct_bus + b_jourt[[s]] * jourt_bus + b_freq[[s]] * freq_bus + b_seat_P[[s]] * (seat_bus==1) + b_seat_A[[s]] * (seat_bus==2) + b_seat_G[[s]] * (seat_bus==3) + b_seatAm[[s]]*((seat_bus==2)*male) + b_seatGm[[s]]*((seat_bus==3)*male)
V[['train']] = asc_train[[s]] + b_cost[[s]] * cost_train + b_acct[[s]] * acct_train + b_jourt[[s]] * jourt_train + b_freq[[s]] * freq_train + b_seat_P[[s]] * (seat_train==1) + b_seat_A[[s]] * (seat_train==2) + b_seat_G[[s]] * (seat_train==3)+ b_seatAm[[s]]*((seat_train==2)*male)+b_seatGm[[s]]*((seat_train==3)*male)
V[['high']] = asc_high[[s]] + b_cost[[s]] * cost_high + b_acct[[s]] * acct_high + b_jourt[[s]] * jourt_high + b_freq[[s]] * freq_high + b_seat_P[[s]] * (seat_high==1) + b_seat_A[[s]] * (seat_high==2) + b_seat_G[[s]] * (seat_high==3) + b_seatAm[[s]]*((seat_high==2)*male)+ b_seatGm[[s]]*((seat_high==3)*male)
V[['air']] = asc_air[[s]] + b_cost[[s]] * cost_air + b_acct[[s]] * acct_air + b_jourt[[s]] * jourt_air + b_freq[[s]] * freq_air + b_seat_P[[s]] * (seat_air==1) + b_seat_A[[s]] * (seat_air==2) + b_seat_G[[s]] * (seat_air==3) + b_seatAm[[s]]*((seat_air==2)*male)+ b_seatGm[[s]]*((seat_air==3)*male)
nl_settings$V = V
nl_settings$nlNests = nlNests[[s]]
nl_settings$nlStructure = nlStructure[[s]]
P[[s]] = apollo_nl(nl_settings, functionality)
P[[s]] = apollo_panelProd(P[[s]],apollo_inputs,functionality)
s=s+1
}
lc_settings = list(inClassProb = P, classProb=pi_values)
P[["model"]] = apollo_lc(lc_settings,apollo_inputs,functionality)
P = apollo_prepareProb(P,apollo_inputs,functionality)
}
#apollo_control$worklnLogs=TRUE
model = apollo_estimate(apollo_beta, apollo_fixed, apollo_probabilities,apollo_inputs)
apollo_modelOutput(model,modelOutput_settings = list(printPVal=TRUE))Code: Select all
Warning message:
In sqrt(diag(varcov)) : NaNs produced
> apollo_modelOutput(model,modelOutput_settings = list(printPVal=TRUE))
Model run by lenovo using Apollo 0.2.7 on R 4.2.1 for Windows.
www.ApolloChoiceModelling.com
Model name : Apollo_example_20_LC-NL-SCY
Model description : LC model with class allocation model
Model run at : 2023-05-06 16:04:02
Estimation method : bfgs
Model diagnosis : successful convergence
Number of individuals : 439
Number of rows in database : 2634
Number of modelled outcomes : 7902
1 : 2634
2 : 2634
model : 2634
Number of cores used : 6
Model without mixing
LL(start) : -3936.22
LL(0, whole model) : -3651.5
LL(C, whole model) : -3028.74
LL(final, whole model) : -2629.62
Rho-square (0) : 0.2799
Adj.Rho-square (0) : 0.273
Rho-square (C) : 0.1318
Adj.Rho-square (C) : 0.1235
AIC : 5309.25
BIC : 5483.62
LL(0,component_1) : -3651.5
LL(final,component_1) : -Inf Likelihood equal to zero for at least
one individual in this component.
LL(0,component_2) : -3651.5
LL(final,component_2) : -Inf Likelihood equal to zero for at least
one individual in this component.
Estimated parameters : 25
Time taken (hh:mm:ss) : 00:02:9.77
pre-estimation : 00:00:9.32
estimation : 00:01:23.39
post-estimation : 00:00:37.06
Iterations : 129
Min abs eigenvalue of Hessian : 1431139
Some eigenvalues of Hessian are positive, indicating potential problems!
Unconstrained optimisation.
Estimates:
Estimate s.e. t.rat.(0) p(1-sided) Rob.s.e. Rob.t.rat.(0) p(1-sided)
asc_bus_c1 0.000000 NA NA NA NA NA NA
asc_train_c1 3.248857 NaN NaN NaN 0.001106 2938.38622 0.00000
asc_high_c1 3.096513 NaN NaN NaN 0.001508 2053.82939 0.00000
asc_air_c1 1.846577 0.065261 28.29523 0.000000 0.123262 14.98096 0.00000
b_cost_c1 -0.001809 NaN NaN NaN 7.643e-06 -236.62763 0.00000
b_acct_c1 -0.150757 NaN NaN NaN 0.026421 -5.70598 5.784e-09
b_jourt_c1 -0.197729 NaN NaN NaN 1.5580e-04 -1269.09841 0.00000
b_freq_c1 0.014161 NaN NaN NaN 7.9082e-04 17.90612 0.00000
b_seat_P_c1 0.000000 NA NA NA NA NA NA
b_seat_A_c1 0.167447 0.008071 20.74793 0.000000 0.002400 69.78094 0.00000
b_seat_G_c1 -0.040720 NaN NaN NaN 0.002171 -18.75382 0.00000
lambda_A_c1 1.000000 NA NA NA NA NA NA
lambda_B_c1 0.007403 NaN NaN NaN 0.001313 5.63643 8.681e-09
lambda_C_c1 1.000000 NA NA NA NA NA NA
delta_c1 0.162121 0.099651 1.62689 0.051880 0.104840 1.54638 0.06101
asc_bus_c2 0.000000 NA NA NA NA NA NA
asc_train_c2 0.872793 0.131832 6.62049 1.790e-11 0.161702 5.39753 3.378e-08
asc_high_c2 0.946578 0.150436 6.29221 1.565e-10 0.144131 6.56750 2.558e-11
asc_air_c2 -0.013408 0.228424 -0.05870 0.476597 0.247061 -0.05427 0.47836
b_cost_c2 -0.003953 6.1395e-04 -6.43865 6.027e-11 6.7848e-04 -5.82625 2.834e-09
b_acct_c2 -0.255237 0.099688 -2.56035 0.005228 0.079069 -3.22804 6.2322e-04
b_jourt_c2 -0.038368 0.014965 -2.56384 0.005176 0.012041 -3.18652 7.1997e-04
b_freq_c2 0.002302 0.005912 0.38938 0.348499 0.004756 0.48400 0.31419
b_seat_P_c2 0.000000 NA NA NA NA NA NA
b_seat_A_c2 0.577226 0.112626 5.12518 1.486e-07 0.115537 4.99603 2.926e-07
b_seat_G_c2 0.767851 0.110701 6.93629 2.013e-12 0.130821 5.86947 2.186e-09
lambda_A_c2 1.000000 NA NA NA NA NA NA
lambda_B_c2 0.659898 0.148464 4.44483 4.398e-06 0.134554 4.90435 4.687e-07
lambda_C_c2 1.000000 NA NA NA NA NA NA
delta_c2 0.000000 NA NA NA NA NA NA
b_seatAm_c1 0.007009 0.004399 1.59340 0.055535 0.004567 1.53484 0.06241
b_seatGm_c1 0.005471 0.003246 1.68586 0.045912 0.004424 1.23663 0.10811
b_seatAm_c2 -0.586488 0.147873 -3.96618 3.652e-05 0.172829 -3.39347 3.4507e-04
b_seatGm_c2 -0.639784 0.159948 -3.99994 3.168e-05 0.188321 -3.39731 3.4026e-04
Nesting structure for NL model component 1:
Nest: root (1)
|-Nest: A_c1 (1)
| '-Alternative: bus
|-Nest: B_c1 (0.0074)
| |-Alternative: train
| '-Alternative: high
'-Nest: C_c1 (1)
'-Alternative: air
Nesting structure for NL model component 2:
Nest: root (1)
|-Nest: A_c2 (1)
| '-Alternative: bus
|-Nest: B_c2 (0.6599)
| |-Alternative: train
| '-Alternative: high
'-Nest: C_c2 (1)
'-Alternative: air
Summary of class allocation for LC model component :
Mean prob.
Class_1 0.7317
Class_2 0.2683
Do you have any idea what is happening and is there any way to resolve this issue?
Best regards, Chunyu Song