Easy way to use parallel options of scikit-learn functions on HPC
SKLearn manages its parallelism with Joblib. Joblib can swap out the multiprocessing backend for other distributed systems like dask.distributed or IPython Parallel. See this issue on the sklearn
github page for details.
Example using Joblib with Dask.distributed
Code taken from the issue page linked above.
from sklearn.externals.joblib import parallel_backend
search = RandomizedSearchCV(model, param_space, cv=10, n_iter=1000, verbose=1)
with parallel_backend('dask', scheduler_host='your_scheduler_host:your_port'):
search.fit(digits.data, digits.target)
This requires that you set up a dask.distributed
scheduler and workers on your cluster. General instructions are available here: http://dask.readthedocs.io/en/latest/setup.html
Example using Joblib with ipyparallel
Code taken from the same issue page.
from sklearn.externals.joblib import Parallel, parallel_backend, register_parallel_backend
from ipyparallel import Client
from ipyparallel.joblib import IPythonParallelBackend
digits = load_digits()
c = Client(profile='myprofile')
print(c.ids)
bview = c.load_balanced_view()
# this is taken from the ipyparallel source code
register_parallel_backend('ipyparallel', lambda : IPythonParallelBackend(view=bview))
...
with parallel_backend('ipyparallel'):
search.fit(digits.data, digits.target)
Note: in both the above examples, the n_jobs
parameter seems to not matter anymore.
Set up dask.distributed with SLURM
For SLURM the easiest way to do this is probably to use the dask-jobqueue project
>>> from dask_jobqueue import SLURMCluster
>>> cluster = SLURMCluster(project='...', queue='...', ...)
>>> cluster.scale(20)
You could also use dask-mpi or any of several other methods mentioned at Dask's setup documentation
Use dask.distributed directly
Alternatively you can set up a dask.distributed or IPyParallel cluster and then use these interfaces directly to parallelize your SKLearn code. Here is an example video of SKLearn and Joblib developer Olivier Grisel, doing exactly that at PyData Berlin: https://youtu.be/Ll6qWDbRTD0?t=1561
Try Dask-ML
You could also try the Dask-ML package, which has a RandomizedSearchCV
object that is API compatible with scikit-learn but computationally implemented on top of Dask
https://github.com/dask/dask-ml
pip install dask-ml
user3271237
Updated on July 09, 2022Comments
-
user3271237 almost 2 years
In many functions from scikit-learn implemented user-friendly parallelization. For example in
sklearn.cross_validation.cross_val_score
you just pass desired number of computational jobs inn_jobs
argument. And for PC with multi-core processor it will work very nice. But if I want use such option in high performance cluster (with installed OpenMPI package and using SLURM for resource management) ? As I knowsklearn
usesjoblib
for parallelization, which usesmultiprocessing
. And, as I know (from this, for example, Python multiprocessing within mpi) Python programs parallelized withmultiprocessing
easy to scale oh whole MPI architecture withmpirun
utility. Can I spread computation ofsklearn
functions on several computational nodes just usingmpirun
andn_jobs
argument? -
IVlad almost 8 yearsI'm trying to get the first example working, the one also described here: distributed.readthedocs.io/en/latest/joblib.html. I used
dask-ssh
to set up my scheduler and workers. That works fine, if I print the scheduler object I get the right number of cores (240
). Next, I wrapped the call to the randomizedsearch'sfit
in thewith
statement. If I look in the console window where i executeddask-ssh
, I see a connection from the node I run the python script in. However, there is no distributed work going on. It doesn't scale, and it doesn't even see the GPUs that the workers have. -
IVlad almost 8 yearsI also tried tinkering with RandomizedSearchCV's
n_jobs
parameter, setting to-1
,1
,100
,240
. Each value above20
leads to about the same performance, which makes me think that nothing is actually running on the distributed workers, but on the node I run the python script on (gensim also prints a message that there is no GPU. There is a GPU on the worker nodes, but there isn't one on the node I run the script from). -
MRocklin almost 8 yearsAt this point you're beyond my expertise. You could raise an issue with the joblib maintainers. I've e-mailed one and alerted him to this question, but they're busy people. I've also appended the answer to point to the experimental dask-learn package
-
IVlad almost 8 yearsOk, thanks. I tried dklearn, but unfortunately it just gets stuck for me, seems to never finish. Will keep at it.
-
IVlad almost 8 yearsUpdate: also tried
ipyparallel
, same thing I described withdask
. The workers (engines in ipyparallel) are successfully created, the client sees them, but my grid searches do not run on them. -
IVlad almost 8 yearsI took the liberty to edit your answer with working
sklearn
examples, as I figured them out with the help ofsklearn
developers. Please let me know if you're happy with it, in which case I'll award the bounty. -
MRocklin almost 8 yearsCool. I'm surprised that you had to call
register_parallel_backend('distributed', DistributedBackend)
. This should already be handled indistributed.joblib
. Perhaps sklearn is packing along their own version of the joblib library now? -
IVlad almost 8 yearsYes, it is apparently. That's why you have to import the ones they use, not the one installed on your platform. And that's what made things confusing for me, since all examples were importing the platform joblib, not sklearn's. Ah well, at least it's taken care of.
-
MRocklin almost 8 yearsIn the end I hope that the solution ends up working out well for you
-
Kai over 6 years@IVlad, I ran your ipyparallel example above and I can see all 8 workers busy executing. Thank you for providing it. However, when I use RandomizedSearchCV with different model (sklearn_crfsuite) only one worker is active. crf model provides the same methods as other sklearn models, so I'm not sure what's happening.
-
George Pamfilis about 6 yearswhen using ipyparallel do i have to sync the imports? stackoverflow.com/questions/33722330/…