Important: Read this before posting to this forum

  1. This forum is for questions related to the use of Apollo. We will answer some general choice modelling questions too, where appropriate, and time permitting. We cannot answer questions about how to estimate choice models with other software packages.
  2. There is a very detailed manual for Apollo available at http://www.ApolloChoiceModelling.com/manual.html. This contains detailed descriptions of the various Apollo functions, and numerous examples are available at http://www.ApolloChoiceModelling.com/examples.html.
  3. Before asking a question on the forum, users are kindly requested to follow these steps:
    1. Check that the same issue has not already been addressed in the forum - there is a search tool.
    2. Ensure that the correct syntax has been used. For any function, detailed instructions are available directly in Apollo, e.g. by using ?apollo_mnl for apollo_mnl
    3. Check the frequently asked questions section on the Apollo website, which discusses some common issues/failures. Please see http://www.apollochoicemodelling.com/faq.html
    4. Make sure that R is using the latest official release of Apollo.
  4. If the above steps do not resolve the issue, then users should follow these steps when posting a question:
    1. provide full details on the issue, including the entire code and output, including any error messages
    2. posts will not immediately appear on the forum, but will be checked by a moderator first. This may take a day or two at busy times. There is no need to submit the post multiple times.

Different results between Windows and Linux server

Report bugs or highlight issues with Apollo functions. At a minimum, please include the part of the output where you believe the bug is manifested. Ideally, please share your model file.
Post Reply
ecd921
Posts: 5
Joined: 17 Nov 2020, 12:38

Different results between Windows and Linux server

Post by ecd921 » 09 Nov 2021, 15:56

Dear Stephane and David,

I have been running some MXL estimations on a Linux server using Apollo 0.2.5 with good results. When I moved the same files to a Windows computer, also with apollo 0.2.5., the same estimations show me "some eigenvalues of Hessian are positive" and I get many "NAN" in the s.e., t.rat, etc. Which was not the case in the Linux server.

Which estimation should I rely on, the Linux server one or the Windows one? Why is this happenning?

Thank you.
Best regards,
Emilia

stephanehess
Site Admin
Posts: 440
Joined: 24 Apr 2020, 16:29

Re: Different results between Windows and Linux server

Post by stephanehess » 12 Nov 2021, 09:24

Emilia

small differences can happen, but this looks like an identification issue in one system. Can you please share the output for both?

Thanks

Stephane
--------------------------------
Stephane Hess
www.stephanehess.me.uk

ecd921
Posts: 5
Joined: 17 Nov 2020, 12:38

Re: Different results between Windows and Linux server

Post by ecd921 » 12 Nov 2021, 10:06

Dear Stephane,

Please find attach a zip with both outputs.

Best regards
Emilia
Attachments
MMNL_NoLux_100_v9_CY_output.zip
Both outputs included
(14.23 KiB) Downloaded 29 times

stephanehess
Site Admin
Posts: 440
Joined: 24 Apr 2020, 16:29

Re: Different results between Windows and Linux server

Post by stephanehess » 29 Nov 2021, 13:57

Emilia

apologies for the slow reply. This is quite odd, and what surprises is that the code doesn't use analytical derivatives. To help diagnose the issue further, would you be able to share the data with us offline?

Stephane
--------------------------------
Stephane Hess
www.stephanehess.me.uk

ecd921
Posts: 5
Joined: 17 Nov 2020, 12:38

Re: Different results between Windows and Linux server

Post by ecd921 » 29 Nov 2021, 14:33

Dear Stephane,

I have sent the data to your e-mail. Please let me know if by any chance you have not receive it.

Thanks.

Best regards,

Emilia

stephanehess
Site Admin
Posts: 440
Joined: 24 Apr 2020, 16:29

Re: Different results between Windows and Linux server

Post by stephanehess » 02 Dec 2021, 21:50

Emilia

I had a look.

A few points:

1. I would strongly recommend using more than 100 draws for the model
2. There is no reason to use the workInLogs setting for your model with only 12 observations per individual
3. Most importantly, you are using the setting

Code: Select all

hessianRoutine="maxLik"
. There is no need for this. It would mean using numerical derivatives for the Hessian instead of analytical derivatives, and this, potentially coupled with the low number of draws, led to the problems you saw

Stephane
--------------------------------
Stephane Hess
www.stephanehess.me.uk

Post Reply