s-news
[Top] [All Lists]

Permutation Tests on Large Objects

To: S-News <s-news@wubios.wustl.edu>
Subject: Permutation Tests on Large Objects
From: "Kim Elmore" <Kim.Elmore@noaa.gov>
Date: Thu, 01 Dec 2005 14:56:44 -0600
User-agent: Mozilla Thunderbird 1.0.7 (Windows/20050923)
I'm using S-Plus 7.0.3 under XP on a dual-Opteron box with 4 GB of RAM.

I'm trying to do a two-sample permutation test using large objects. Specifically, I have two vectors: one has a length of 2315245 while the other has a length of 3033371 -- tehse vectors represent geopotential height errors from meteorological models under different conditions. I want to assess the significance level of the RMS value of each vector and I'm trying to use permutationTest2() from the resample library to do this.

Aparently, S-Plus creates an array or indices for resampling as a first step, but because my objects are so large, such an aray with anything more than a few tens of replicates quickly becomes too large to fit within the addressable memory of the XP 32-bit OS. I've tried various ways of trimming down the memory requiremenets, but with no success. For example, I thought I might be able to use addSamples(), but there are no default methods to use on a permutationtest object. I've played with block.size, but that doesn't help, either. I even tried using the bigdata library, but the resample library doesn;t play well with bigdata objects and, in any event, certainly doesn't create any bigdata objects internally.

Do any of you know of a way around this?

Kim Elmore

--
           Kim Elmore, OU/CIMMS/NSSL
  "All of weather is divided into three parts:
Yes, No and Maybe. The greatest of these is Maybe."


<Prev in Thread] Current Thread [Next in Thread>
  • Permutation Tests on Large Objects, Kim Elmore <=