[Bioc-devel] Fwd: Re: Build error due to TensorFlow installation
Martin Morgan
mtmorg@n@b|oc @end|ng |rom gm@||@com
Thu Sep 5 15:35:12 CEST 2019
I'm with Simon on the value of using TensorFlow here, and it's too bad the build system is unable to support more flexible use of python (there are additional challenges, for instance inter-operability between packages each depending on & starting their own python interpreter...)
It seems like the keras model
https://github.com/rstudio/keras
https://keras.rstudio.com/
is reasonable -- providing an `install_keras()` and related convenience functions, but writing the package so that it does not depend on an available python for build & check. Obviously not checking the TensorFlow interface on a daily basis compromises the robustness of the software, and hopefully the build system will be able to provide a more robust testing environment in the future.
Martin
On 9/5/19, 9:10 AM, "Bioc-devel on behalf of Kasper Daniel Hansen" <bioc-devel-bounces using r-project.org on behalf of kasperdanielhansen using gmail.com> wrote:
It might be true that the code is much easier on your end, but it comes
with the complexity of using TensorFlow, which is a huge dependency. And
that library can be easy or impossible to install depending on your box. I
am not saying people shouldn't use TensorFlow, but I am saying it brings a
lot of complexities with it.
And (trying to) installing TensorFlow as part of .onLoad seems ... not the
right way to deal with this.
On Thu, Sep 5, 2019 at 6:27 AM Simon Dirmeier <simon.dirmeier using web.de> wrote:
>
> Hi Herve,
>
> > All this means that if you replace some old C++ code with TensorFlow
> > then we will need to mark your package as unavailable on Windows and
> > Mac, at least for now. Sorry. I wonder if there was something wrong
> > with this old C++ code. I would recommend that you stick to it if you
> can.
> >
> The code is fine, but still an impractical complexity that doesn't need
> to exist:
>
> 1) It's hard to extend and read for others.
>
> 2) It needs a custom configure.ac.
>
> 3) Extending the package to other models/families is a huge pain, as one
> needs to derive custom coordinate descents (or other optimizers for that
> matter) for each.
>
> On the other side:
>
> 1) TF allowed me to replace like 5000 lines of source code with 100
> lines of R.
>
> 2) TF allows to easily extend with other models with just a few lines.
>
> 3) I don't need a huge test suite.
>
> 4) On GPUs it's a huge speedup.
>
> So, I'll revert the changes back on Bioc devel for now and continue
> development on another branch.
>
> Cheers,
>
> Simon
>
>
> Am 04.09.19 um 16:53 schrieb Pages, Herve:
> >
> > Hi Simon,
> >
> > On 9/3/19 09:11, Simon Dirmeier wrote:
> >> ...
> >> Do you think it would be possible to install TensorFlow and
> >> TensorFlow-Probability on the builders? I'd assume that many would
> >> profit from that.
> >>
> > As Lori mentioned at the end of her email (see below), we can't make
> > the tensorflow Python module available on our Windows builders at the
> > moment because we need to update Python to Python 3 on these machines
> > first (AFAIK tensorflow is only available for Python 3 on Windows).
> > This is something that we are currently working on.
> >
> > As for the Mac builders, we have tensorflow there but unfortunately
> > it's an old version because recent versions of are broken on El
> > Capitan (this is the Mac OS version that, for various reasons, we are
> > stuck with at the moment). This prevents us from installing the
> > tensorflow_probability module which requires a recent version of
> > tensorflow.
> >
> > The tensorflow and tensorflow_probability modules are available on our
> > Linux builders.
> >
> > All this means that if you replace some old C++ code with TensorFlow
> > then we will need to mark your package as unavailable on Windows and
> > Mac, at least for now. Sorry. I wonder if there was something wrong
> > with this old C++ code. I would recommend that you stick to it if you
> can.
> >
> > Best,
> >
> > H.
> >
> >
> >>> Right now tensorflow is unavailable on our windows server (tokay1)�as
> >>> we have not updated to python 3 (but will be in the near future)� The
> >>> windows error can be ignored for now.
> >>>
> >>>
> >>>
> >>> Lori Shepherd
> >>>
> >>> Bioconductor Core Team
> >>>
> >>> Roswell Park Cancer Institute
> >>>
> >>> Department of Biostatistics & Bioinformatics
> >>>
> >>> Elm & Carlton Streets
> >>>
> >>> Buffalo, New York 14263
> >>>
> >>>
> ------------------------------------------------------------------------
> >>> *From:* Bioc-devel<bioc-devel-bounces using r-project.org> on behalf of
> >>> Simon Dirmeier<simon.dirmeier using web.de>
> >>> *Sent:* Monday, September 2, 2019 6:08:45 AM
> >>> *To:*bioc-devel using r-project.org <bioc-devel using r-project.org>
> >>> *Subject:* [Bioc-devel] Build error due to TensorFlow installation
> >>> Dear all,
> >>>
> >>> since I replaced some old C++ code with TensorFlow I am getting some
> >>> build errors on merida1 and tokay1 regarding installation (even though
> I
> >>> install TF and TF Probability during /.onLoad/)
> >>>
> >>>
> https://urldefense.proofpoint.com/v2/url?u=http-3A__bioconductor.org_checkResults_devel_bioc-2DLATEST_netReg_merida1-2Dbuildsrc.html&d=DwIFaQ&c=eRAMFD45gAfqt84VtBcfhQ&r=BK7q3XeAvimeWdGbWY_wJYbW0WYiZvSXAJJKaaPhzWA&m=Iy8WHNH6twF-8QkJjKa7MOIXm9mYjGg4pBlLb2CGiq8&s=wjC8b2uCm0PJgQ1LRO_CTrqNvfGKpEKN0N6QN23Sbd0&e=
> >>>
> >>> Does anyone know how I can fix this or did anyone use TF with
> >>> Bioconductor so far?
> >>>
> >>> Many thanks in advance.
> >>>
> >>> Best,
> >>>
> >>> Simon
> >>>
> >>>
> >>>
> >>>
> >>>
> >>> ������� [[alternative HTML version deleted]]
> >>>
> >>> _______________________________________________
> >>> Bioc-devel using r-project.org mailing list
> >>>
> https://urldefense.proofpoint.com/v2/url?u=https-3A__stat.ethz.ch_mailman_listinfo_bioc-2Ddevel&d=DwIFaQ&c=eRAMFD45gAfqt84VtBcfhQ&r=BK7q3XeAvimeWdGbWY_wJYbW0WYiZvSXAJJKaaPhzWA&m=Iy8WHNH6twF-8QkJjKa7MOIXm9mYjGg4pBlLb2CGiq8&s=98vYdk5e2ryxlWN99O68ty2dUHnoE23VPU9TwSF8rPc&e=
> >>>
> >>> This email message may contain legally privileged and/or confidential
> >>> information. If you are not the intended recipient(s), or the employee
> >>> or agent responsible for the delivery of this message to the intended
> >>> recipient(s), you are hereby notified that any disclosure, copying,
> >>> distribution, or use of this email message is prohibited. If you have
> >>> received this message in error, please notify the sender immediately
> >>> by e-mail and delete this email message from your computer. Thank you.
> >> [[alternative HTML version deleted]]
> >>
> >>
> >> _______________________________________________
> >> Bioc-devel using r-project.org mailing list
> >>
> https://urldefense.proofpoint.com/v2/url?u=https-3A__stat.ethz.ch_mailman_listinfo_bioc-2Ddevel&d=DwICAg&c=eRAMFD45gAfqt84VtBcfhQ&r=BK7q3XeAvimeWdGbWY_wJYbW0WYiZvSXAJJKaaPhzWA&m=Iy8WHNH6twF-8QkJjKa7MOIXm9mYjGg4pBlLb2CGiq8&s=98vYdk5e2ryxlWN99O68ty2dUHnoE23VPU9TwSF8rPc&e=
> > --
> > Hervé Pagès
> >
> > Program in Computational Biology
> > Division of Public Health Sciences
> > Fred Hutchinson Cancer Research Center
> > 1100 Fairview Ave. N, M1-B514
> > P.O. Box 19024
> > Seattle, WA 98109-1024
> >
> > E-mail:hpages using fredhutch.org
> > Phone: (206) 667-5791
> > Fax: (206) 667-1319
>
> [[alternative HTML version deleted]]
>
> _______________________________________________
> Bioc-devel using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/bioc-devel
>
--
Best,
Kasper
[[alternative HTML version deleted]]
_______________________________________________
Bioc-devel using r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/bioc-devel
More information about the Bioc-devel
mailing list