peak calling device initialization issue.

[LnLogN]

I'm getting "Error in device initialization" error when running peak calling with dREG, but the prediction function runs fines. .

I'm running on a node with:

OS: redhat 7.4 kernel 3.10.0-693
GPU: GTX1080
Cuda: cuda9.0
dreg version: latest version checked out from github
rgtsvm version: latest version
bedops: 2.4.14

All the data are test data downloaded from ftp server provided, below is the test run output:

Using: R --vanilla --slave --args test/K562_long_plus.bw test/K562_long_minus.bw test/k562/K562.TEST test/asvm.gdm.6.6M.20170828.rdata 4 0 < /home/installs/packs/rgtsvm/dREG/run_dREG.R
WARNING: ignoring environment value of R_HOME
Loading required package: dREG
Loading required package: bigWig
Loading required package: e1071
Loading required package: rphast
Loading required package: snowfall
Loading required package: snow
Loading required package: data.table
Loading required package: rmutil

Attaching package: ‘rmutil’

The following object is masked from ‘package:stats’:

    nobs

The following objects are masked from ‘package:base’:

    as.data.frame, units

Loading required package: mvtnorm
Loading required package: randomForest
randomForest 4.6-14
Type rfNews() to see new features/changes/bug fixes.
Warning message:
replacing previous import ‘stats::nobs’ by ‘rmutil::nobs’ when loading ‘dREG’ 
------------ Parameters ------------- 
Bigwig(plus): test/K562_long_plus.bw 
Bigwig(minus): test/K562_long_minus.bw 
Output: test/k562/K562.TEST 
dREG model: test/asvm.gdm.6.6M.20170828.rdata 
CPU cores: 4 
GPU ID: 0 
Using Rgtsvm: TRUE 
-------------------------------------
 [ 2019-01-23 11:21:43 ] 1) Checking bigWig files.
[ 2019-01-23 11:21:49 ] 2) Starting peak calling.
Loading required package: Rgtsvm

Attaching package: ‘Rgtsvm’

The following objects are masked from ‘package:e1071’:

    svm, tune.control, tune.svm

Error in checkForRemoteErrors(val) : 
  one node produced an error: Error in device initialization.
Calls: system.time ... clusterApply -> staticClusterApply -> checkForRemoteErrors
Timing stopped at: 117.8 3.9 144.7
Execution halted

What have I tried:

1. running with 1 cpu and 1 gpu - same error.
2. Tried to run with no gpu by not specifying a number after cpu or using "FALSE", got this error:
   Error in if (gpu_id > 0) { : missing value where TRUE/FALSE needed
   Execution halted

Please let me know of any insights to this problem.

[LnLogN]

Found the problem, I only have 1 gpu on the node, and the gpu_cores is hard coded to 1, but rgtsvm uses this number as gpu id. change this to gpu_cores=gpu_id worked, not sure if it'll work for multi-GPUs.

run.time <- system.time(r <- peak_calling( asvm, gdm, ps_plus_path, ps_minus_path, cpu_cores=cpu_cores, use_rgtsvm=use_rgtsvm, gpu_cores=1));

[wzhy2000]

Thank you for your feedback.
I tried to fix this question. Can you try the new codes again?

Or
you can use the online gateway to run your data.
http://dreg.dnasequence.org

Best,
Zhong Wang

[gulcinozer]

I tried to run dReg without GPU option and received an error about rgtsvm library. When I check the code in run_dREG.R closely, I noticed that there is a parameter initiation error.

use_rgtsvm <- FALSE;
gpu_id <- as.integer(args[6])
if (!is.na(gpu_id)) use_rgtsvm<-TRUE

If I provide 0 for gpu_id, it is not NA anymore and use_rgtsvm is set to TRUE. Then, this will cause an error in peak_calling function.

If I don't provide any value for gpu_id, use_rgtsvm would be set to FALSE, but then the next step would return an error:
if(gpu_id>0)
{
library(Rgtsvm);
ret <- selectGPUdevice(gpu_id);
}

I manually edited run_dREG.R to bypass this error.

[wzhy2000]

Thank you for your feedback. I will change it as you did.
But I don't suggest running dREG without GPU. You can use our online computing resource: dREG Gateway ( https://dreg.dnasequence.org/ )