Mixed Logit Model Running time

[Alejandro Montes]

Dear Prof. Bierlaire,

For my MSc Thesis I'm running several Mixed Logit models using PandasBiogeme. For each model I am only including one random variable as to reduce complexity. However, some of the models are taking too long to run when the numberOfDraws is at around 32000. I wanted to ask you if there is any recommendation on how to accelerate said times. For instance, I was taking that maybe biogeme is not using all the cores available in my laptop, but I'm not sure about that.

I am generating the random draws as Normal, following the example 05 available in the webpage

Thanks in advance for your help,

Alejandro Montes

[Bierlaire Michel]

If you have only one random variable, you can use numerical integration instead of Monte-Carlo. 

If you want to use Monte-Carlo, you can also check if the results stabilize with a lower number of draws. You can also use variance reduction methods, such as antithetic draws. 

By default, Biogeme is using as many cores as reported by the function multiprocessing.cpu_count(). The actual number of cores used is reported in the HTML report file. 

See

https://transp-or.epfl.ch/documents/technicalReports/Bier19.pdf

--
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/73092ca2-a781-47e1-a060-52eea869fcf8n%40googlegroups.com.

[Alejandro Montes]

Thanks for the quick reply,

What about halton draws? would they help to reduce the number of draws? I'm modelling one random parameter at a time, but then I would like to run one model with the ones that I found have a significant effect.

[Bierlaire Michel]

On 30 Aug 2021, at 13:38, Alejandro Montes <amont...@gmail.com> wrote:

Thanks for the quick reply,

What about halton draws? would they help to reduce the number of draws? I'm modelling one random parameter at a time, but then I would like to run one model with the ones that I found have a significant effect.

Sure. There is a comparison on table 6, p. 16 in the report. 

On Monday, August 30, 2021 at 1:26:55 PM UTC+2 michel.b...@epfl.ch wrote:

If you have only one random variable, you can use numerical integration instead of Monte-Carlo. 

If you want to use Monte-Carlo, you can also check if the results stabilize with a lower number of draws. You can also use variance reduction methods, such as antithetic draws. 

By default, Biogeme is using as many cores as reported by the function multiprocessing.cpu_count(). The actual number of cores used is reported in the HTML report file. 

See

https://transp-or.epfl.ch/documents/technicalReports/Bier19.pdf

On 30 Aug 2021, at 12:46, Alejandro Montes <amont...@gmail.com> wrote:

Dear Prof. Bierlaire,

For my MSc Thesis I'm running several Mixed Logit models using PandasBiogeme. For each model I am only including one random variable as to reduce complexity. However, some of the models are taking too long to run when the numberOfDraws is at around 32000. I wanted to ask you if there is any recommendation on how to accelerate said times. For instance, I was taking that maybe biogeme is not using all the cores available in my laptop, but I'm not sure about that.

I am generating the random draws as Normal, following the example 05 available in the webpage

Thanks in advance for your help,

Alejandro Montes

--
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/73092ca2-a781-47e1-a060-52eea869fcf8n%40googlegroups.com.

--
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/2a3368c2-de94-449b-846c-2a943d066f14n%40googlegroups.com.