0
12 months ago by
Sara0
Sara0 wrote:

Hi all,

I'm running a Fisher exact test with a large amount of data (2 columns and about 3000 rows) on the cluster (Rocks 6.2) with R 3.3.2 , but I got the below workspace error with 1e9 value as workspace:

 FEXACT error 7.
LDSTP is too small for this problem.
Try increasing the size of the workspace.
Calls: [ -> [.data.table -> fisher.test
Execution halted

With increasing workspace, I got another error:

cannot allocate memory block of size 134217728 Tb
Calls: [ -> [.data.table -> fisher.test
In fisher.test(matrix(c(a1, a2, b1, b2), ncol = 2),  :
NAs introduced by coercion to integer range
Execution halted

I used "subR -input=file.R -mem=200G" for submitting and running R on the cluster. I played a lot with workspace value, but one of the two above-mentioned errors appeared. Also, "ulimit" returned me "unlimited" , so sounds that there is no limitation from system admin. Surprisingly, I before run the Fisher test on the same dataset with the same workspace value (1e9) on another server successfully during just 10 minutes. Although I didn't remember its RMA, the RAM and other properties were not higher the current server I'm working on it. Could you please kindly help me out to solve this problem?

R test fisher exact workspace • 310 views
modified 12 months ago by Aaron Lun25k • written 12 months ago by Sara0
0
12 months ago by
Aaron Lun25k
Cambridge, United Kingdom
Aaron Lun25k wrote:
1. This is not a Bioconductor question, you should ask this on a general R programming forum like Stack Overflow, or on the R-help mailing list.
2. This is already a well-encountered "problem" that has much commentary online (just google "fisher exact test workspace", for example). The error is also described in ?fisher.test.
3. Make your life easier and just use a chi-squared test.