2 replies Latest Post - ‏2013-07-22T13:46:14Z by JohnDoews
1 Post

Pinned topic STATS_RELIMP, Shapley Regression, and Memory Restrictions?

‏2012-10-12T21:41:32Z |

I am encountering memory problems when running Shapley Value Regressions and have tried a number of solutions. I am running SPSS v20 on Windows XP 32-bit.

I have successfully been able to run a Shapley Regression using the STATS_RELIMP R plugin with <12 predictors and 82 cases, but when I increase the number of predictors I receive the message "Error: cannot allocate the vector size of n Mb."

I have tried a number of the suggested online solutions, including increasing the memory limit in R and adding the /3GB switch in the Windows boot.ini file, but I haven't been able to run my full model (65 predictors, 82 cases).

One option I may have is to install SPSS and R on a server to get more RAM, but I'd love to know if there are other ways around this problem. Does anyone have any suggestions?

If you need more information, I'd be happy to provide it.

Updated on 2012-10-16T17:19:06Z at 2012-10-16T17:19:06Z by SystemAdmin
  • SystemAdmin
    396 Posts

    Re: STATS_RELIMP, Shapley Regression, and Memory Restrictions?

    ‏2012-10-16T17:19:06Z  in response to WilliamWelch
    You are running up against the limitations of 32-bit addressing and the R Relimp implementation. Moving to a system with more RAM will not help.

    The 64-bit version of R can be used with Statistics V21 64-bit and might solve this problem, but I haven't tried this myself.

    Otherwise, I suggest that you look at subsets of variables and then perhaps force in ones that you are sure matter and look at the Shapley (just awarded the Nobel prize in Economics) value for others.

    Jon Peck
    • JohnDoews
      1 Post

      Re: STATS_RELIMP, Shapley Regression, and Memory Restrictions?

      ‏2013-07-22T13:46:14Z  in response to SystemAdmin

      I tried running on a 64-bit system. It would work with up to 15 independent variables but not more for a database of about 800 cases. 

      I went and tried to run the analysis in R directly. It did not help. The limitation is not from SPSS but from R. So unless someone addresses the issue in R, I would not expect any change soon ...