Databases
StackOverflow2013
Postsdbo
POparallel randomForest with different results using doSNOW
 

Note that there are some explanatory texts on larger screens.

plurals
  1. POparallel randomForest with different results using doSNOW
    primarykey
    data
    text
    <p>I thought I found a way to make a reproducible <code>foreach</code> loop with <code>doSNOW</code> with the following code </p> <pre><code>library(foreach) library(doSNOW) library(parallel) ncores &lt;- 2 cl &lt;- makeCluster(ncores) registerDoSNOW(cl) foreach(i=1:ncores) %dopar% { set.seed(i) rnorm(1) } stopCluster(cl) </code></pre> <p>Because I used a seed inside the foreach loop, I always get the same results (independent from the computer/OS), namely</p> <pre><code>[[1]] [1] -0.6264538 [[2]] [1] -0.8969145 </code></pre> <p>But if I use the randomForest function, I get different results depending on the OS:</p> <pre><code>library(foreach) library(doSNOW) library(parallel) library(randomForest) set.seed(123) ncores &lt;- 2 cl &lt;- makeCluster(ncores) registerDoSNOW(cl) nr &lt;- 1000 x &lt;- matrix(runif(100000), nr) y &lt;- gl(4, nr/4) trainX &lt;- x[1:800,] trainY &lt;- y[1:800] testX &lt;- x[801:nrow(x),] testY &lt;- y[801:length(y)] rf &lt;- foreach(i=1:ncores, ntree=rep(100, ncores), .packages='randomForest', .combine=combine) %dopar% { set.seed(i) randomForest(trainX, trainY, ntree=ntree) } stopCluster(cl) pred &lt;- predict(rf, new=testX) </code></pre> <p>Each windows computer (I have tried 2 windows computers) with </p> <pre><code>R version 3.0.1 (2013-05-16) Platform: i386-w64-mingw32/i386 (32-bit) </code></pre> <p>gave me the following output</p> <pre><code>table(pred) 1 2 3 4 60 68 72 0 </code></pre> <p>Running the same code on a linux computer (I have tried 2 linux computers) with </p> <pre><code>R version 2.15.3 (2013-03-01) Platform: x86_64-pc-linux-gnu (64-bit) </code></pre> <p>gave me the following output</p> <pre><code>table(pred) 1 2 3 4 69 58 73 0 </code></pre> <p>I used a seed inside the foreach loop, so I thought it should give me the same results, but it gives me only the same results for computers with the same OS. Why does this happen only with <code>randomForest</code> and not with <code>rnorm</code>? Can I do something using <code>doSNOW</code> to get the same results between windows and linux computers? I know that using <code>doRNG</code> is better, but I would like to get the same results with <code>doSNOW</code>, if it is possible...</p>
    singulars
    1. This table or related slice is empty.
    1. This table or related slice is empty.
    1. PTQuestion
    1. USGiuseppe
    1. USGiuseppe
    plurals
    1. This table or related slice is empty.
    1. This table or related slice is empty.
    1. This table or related slice is empty.
    1. This table or related slice is empty.
    1. VO
      singulars
      1. POparallel randomForest with different results using doSNOW
      1. This table or related slice is empty.
      1. VTUpMod
    2. VO
      singulars
      1. POparallel randomForest with different results using doSNOW
      1. USGiuseppe
      1. VTFavorite
    1. COSince the versions of R are different, the versions of randomForest (or some other package) may also be different and could use the stream of random numbers in a slightly different way, leading to different results.
      singulars
      1. POparallel randomForest with different results using doSNOW
      1. USVincent Zoonekynd
    2. CODo you get the same results on Windows and Linux when running the same problem sequentially?
      singulars
      1. POparallel randomForest with different results using doSNOW
      1. USSteve Weston
    3. COThe source of the difference does not seem to be the operating system, but whether it is 32-bit or 64-bit.
      singulars
      1. POparallel randomForest with different results using doSNOW
      1. USVincent Zoonekynd
 

Querying!

 
Guidance
A row detail

Detail views are divided into sections. All the information in the data section comes from columns in the selected row. The other sections display data from other, related rows.

Related data can be related in a to-one or a to-many fashion. Captions of data related in a to-many fashion link to a list view showing a filtered view of the table.

Try moving around until you find a non-empty to-many entry and click on the label to get to one. You can move back to the root by clicking on the database name in the header.

SQuiL has stopped working due to an internal error.

If you are curious you may find further information in the browser console, which is accessible through the devtools (F12).

Reload