[NLPL Task Force (A)] abel - pip3 install issue

[Stephan Oepen]

the software installations on Abel will remain functional for as long
as the system remains available, probably sometime towards the end of
this year.  nobody is in any hurry to purge python 2.7, even though we
may say that we do not officially support it :-).

cheers, oe

On Wed, Aug 14, 2019 at 7:32 PM Asad Sayeed <asad.sayeed at gu.se> wrote:
>
> Hi all,
>
> Actually, we figured out what dependencies it needs, and the
> nlpl-tensorflow module probably is too advanced for it:
>
>    pyhocon
>    tensorflow 1.10.0, tfhub 0.2.0
>    numpy 1.15.4
>    h5py 2.7.0
>
> Any possibility of rolling a module with these versions, on py3?
> Previous experience suggests that I can't compile my own tensorflow
> easily in userspace, though I could be misremembering. Alternatively I
> can just stick to py2.7 if you're still OK with supporting it.
>
> Thanks.
>
> Yours,
> --Asad.
>
> On 2019-08-14 17:11, Asad Sayeed wrote:
> > Hi Stephan,
> >
> > That worked now, but then it turns out there is another problem. The
> > tool we are using requires the compiling of some C++ code into a .so
> > file.  Under the nlpl-tensorflow with python3, it doesn't compile
> > anymore, possibly due to changes to tensorflow contrib stuff that the
> > original author of the tool is depending on.  Yuval seems to have
> > gotten it working on an offsite machine at his end, but it appears
> > that I will still have to fall back to the python2.7 version because I
> > am just not going to be able to debug tensorflow C++ code in any
> > reasonable time.
> >
> > I realize 2.7 is being discontinued but there are still a lot of
> > "orphaned" graduate-student projects on github that represent
> > comparatively recent results in "niche" research areas, so I don't
> > think it will be possible to completely abandon support for py2.7 for
> > a while...
> >
> > Yours,
> > --Asad.
> >
> > On 2019-08-14 16:29, Stephan Oepen wrote:
> >> hi asad (and yuval),
> >>
> >> please feel free to use the 'infrastructure at nlpl.eu' contact address
> >> for all technical requests.
> >>
> >>> -bash-4.1$ pip3 install pyhocon
> >>> -bash: /projects/nlpl/software/tensorflow/1.13/bin/3.7/pip3:
> >>> /projects/nlpl/software/tensorflow/1.13/bin/python3: bad
> >>> interpreter: No
> >>> such file or directory
> >> i am sorry about this.  i have made it a habit to run 'python3 -m
> >> pip', hence had not noticed before that this module had an invalid
> >> 'pip3'.  that should now be fixed, so i imagine 'pip3 install --user'
> >> should allow you to add minor missing dependencies into your home
> >> directory?
> >>
> >> oe