Developing a latent class choice model in Biogeme
Ahmadreza Talebian
Dear Professor Bierlaire,
I am trying to develop a set of latent class choice models. I have used Latent Gold to determine the right number of latent groups. It turned out that there exist two unobserved classes. The software gives the probabilities of each individual belonging to each class as well as the overall probability of each class. Next I want to use Biogeme to develop choice models and here is my plan: I think I can use class probabilities estimated by Latent Gold in Biogeme and manipulate the command such that the software uses the default values. The command would then be
PROB_CLASS1 = Beta('PROB_CLASS1', 0.23, 0, 1, 1)
PROB_CLASS2 = Beta('PROB_CLASS2', 0.77, 0, 1, 1)
From this point, the procedure seems to be straightforward: we should define utility function and characterize the choice model as a discrete mixture of logit, with availability conditions.
The questions that I have are:
1- Am I understanding the overall procedure correctly?
2- Don’t I need to introduce the probabilities of each individual belonging to each class to the software? If yes, how should that be done? I guess it could be done by defining some weight factors.
Many many thanks!
Ahmadreza
Bierlaire Michel
On 13 Oct 2020, at 18:40, Ahmadreza Talebian <ahmadreza...@gmail.com> wrote:
Dear Professor Bierlaire,
I am trying to develop a set of latent class choice models. I have used Latent Gold to determine the right number of latent groups. It turned out that there exist two unobserved classes. The software gives the probabilities of each individual belonging to each class as well as the overall probability of each class. Next I want to use Biogeme to develop choice models and here is my plan: I think I can use class probabilities estimated by Latent Gold in Biogeme and manipulate the command such that the software uses the default values. The command would then be
PROB_CLASS1 = Beta('PROB_CLASS1', 0.23, 0, 1, 1)
PROB_CLASS2 = Beta('PROB_CLASS2', 0.77, 0, 1, 1)
From this point, the procedure seems to be straightforward: we should define utility function and characterize the choice model as a discrete mixture of logit, with availability conditions.
The questions that I have are:
1- Am I understanding the overall procedure correctly?
2- Don’t I need to introduce the probabilities of each individual belonging to each class to the software? If yes, how should that be done? I guess it could be done by defining some weight factors.
Many many thanks!
Ahmadreza
--
You received this message because you are subscribed to the Google Groups "Biogeme" group.
To unsubscribe from this group and stop receiving emails from it, send an email to biogeme+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/biogeme/23a86ae9-862f-442a-94a5-9b2cfa54710dn%40googlegroups.com.