systemml-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Matthias Boehm <mboe...@googlemail.com>
Subject Re: PCA data gen script, no output
Date Sat, 16 Sep 2017 07:03:02 GMT
ok great - I also did some more debugging: (1) when the filename is
specified inside the dml script, the file is indeed written to a directory
~ in HDFS, but (2) when passed from command line the ~ is of course
resolved before it's even passed into SystemML.

So let's do the following: (1) use a small scenario of say 10K x 1K, (2)
run it with absolute file name, and see what happens. If this does not
work, I would suspect some permission issues next - maybe the bridge from
python to the jvm hides some error output. If this is also not the case,
please provide the -explain output and I have a closer look.

Regards,
Matthias

On Fri, Sep 15, 2017 at 11:52 PM, Krishna Kalyan <krishnakalyan3@gmail.com>
wrote:

> Thank you so much for trying this Matthias. I will try this again with
> absolute path.
>
> Regards,
> Krishna
>
> On Sat, Sep 16, 2017 at 12:09 PM, Matthias Boehm <mboehm7@googlemail.com>
> wrote:
>
> > ok, I just tried it with multiple different memory configurations (for
> 6GB
> > driver mem I got the same number of spark instructions as you reported)
> and
> > it ran just fine and produced the outputs. So please, give it a try
> without
> > the ~ (i.e., use an absolute or relative path).
> >
> > Also, even with 2GB mem, this data generation for 1M x 1K ran in about
> 60s
> > (including spark context creation) in my environment. Since your log
> shows
> > a runtime of 5000s, you might want to reduce the data size a bit.
> >
> > Regards,
> > Matthias
> >
> > On Fri, Sep 15, 2017 at 11:08 PM, Krishna Kalyan <
> krishnakalyan3@gmail.com
> > >
> > wrote:
> >
> > > Thanks for the reply,
> > > I have tested with systemml-standalone.py too. I am still faced with
> the
> > > same problem. Currently my spark is configured to work on local fs
> > instead
> > > of HDFS hence I did not have a problem with the ~.
> > >
> > > Regards,
> > > Krishna
> > >
> > >
> > >
> > >
> > > On Sat, Sep 16, 2017 at 7:24 AM, Matthias Boehm <
> mboehm7@googlemail.com>
> > > wrote:
> > >
> > > > well, I don't think any HDFS fs implementation resolves '~' - so it
> has
> > > > probably created a directory called '~/open-source/scripts/PCA_data'
> > in
> > > > your user path in HDFS or current directory in local FS.
> > > >
> > > > Regards,
> > > > Matthias
> > > >
> > > > On Fri, Sep 15, 2017 at 5:47 PM, Krishna Kalyan <
> > > krishnakalyan3@gmail.com>
> > > > wrote:
> > > >
> > > > > Hello,
> > > > > I using PCA
> > > > > <https://github.com/apache/systemml/blob/master/scripts/
> > > > > datagen/genRandData4PCA.dml>
> > > > > data
> > > > > generation scripts to generate data. Unfortunately they do not
> > produce
> > > > any
> > > > > output in the specified target directory.
> > > > >
> > > > > Command used:
> > > > >
> > > > > systemml/bin/systemml-spark-submit.py -f genRandData4PCA.dml
> -nvargs
> > > > > R=1000000 C=1000 OUT=~/open-source/scripts/PCA_data
> > > > >
> > > > > logs
> > > > > https://gist.github.com/krishnakalyan3/70796b13735743886e41d
> > 3da6b75d7
> > > d5
> > > > >
> > > > > This job also does not throw any errors during exection.
> > > > >
> > > > > Thank you so much,
> > > > > Krishna
> > > > >
> > > >
> > >
> >
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message