Hi,
I’m trying to tune the model. I’m running on a c5a.24xlarge 96 CPUs / 192 GiB instance under Jupyter.
The perf step found 6 components as the optimum.
I have 3 blocks: mRNA (36K genes), protein (~1400 proteins), and methylation (~64K probes) (all have been filtered before to eliminate near zero variance)
The runtime for tuning the first 2 components was 7 hours (on a 36 core instance)
Now I’m tryin to tune the other 4 components (I tried all components as ncomp=6, I tried adding just 2 ncomp=4) . the design matrix is and code is:
A matrix: 3 × 3 of type dbl||
mRNA|meth|protein|
|mRNA|0.0|0.1|0.1|
|meth|0.1|0.0|0.1|
|protein|0.1|0.1|0.0|
BPPARAM ← BiocParallel::MulticoreParam(workers = parallel::detectCores()-3)
tune.diablo58.sleNZV_next1 = tune.block.splsda(X , Y,
ncomp = 3, test.keepX = test.keepX, design = design,
already.tested.X = already.tested.X,validation = ‘Mfold’, folds = 5, nrepeat = 4,
BPPARAM = BPPARAM,
dist = “centroids.dist”,near.zero.var = TRUE)
I’m getting the message error:
Design matrix has changed to include Y; each block will be linked to Y.
You have provided a sequence of keepX of length: 9 for block mRNA and 11 for block meth and 5 for block protein.
This results in 495 models being fitted for each component and each nrepeat, this may take some time to run, be patient!
R_zmq_msg_send errno: 4 strerror: Interrupted system call
This occurs after 14 hours of runtime, then the notebook shows an unknown status of the kernel. I monitor each process in the cores, and they are maxed out with 20 tasks, 20Gb of memory used and 96 running cores at ~100%. I see the same problem using a 48, a 64 and 96 cpu instance. The version used is mixOmics 6.26.0
Please let me know if you know how to solve the issue.
Regards
Julio