#1 By: jshining, December 15th, 2013 15:03
Is it currently feasible to compile scipy on heroku servers?
posts such as the one below discourage buildpacks and i also read that vulcan may not be reliable; trying to stay away from anvil b/c of lack of documentation and my general lack of expertise in the area of server configuration
i used 'heroku run bash' to try & compile scipy to see if it would work and ran into problems with BLAS libraries not being found and then needing gfortran once i curl'd for BLAS; overall too many dependencies making me wonder if a buildpack could even solve the problem even if i did have the expertise to create one.
git submodules were mentioned as a possible solution but this also seems a little off-label and may be nontrivialas well
I am possibly a bit over my head but wanted to put feelers out there for any solutions before I abandon heroku b/c of my dependency on scipy. Big picture is I need scikit-image working on a cloud server.
Continuing the discussion from OpenCV and Statically compiled python:
#2 By: kermit666, December 17th, 2013 07:03
It's not possible to get this stuff in a Dyno using e.g. requirements.txt? That's a bummer... It would probably be best if Heroku had such stuff installed via Linux packages and simply provide it by default. E.g. if there are such packages listed in
requirements.txt, simply provide a base dyno that has this installed (if it's too expensive in terms of storage to provide such packages on every dyno).
I found this SO answer which mentions being able to compile the stuff locally and then downloading the binary to a Heroku dyno using a custom buildpack. But, boy, that seems like a nasty way to achieve this when precompiled packages for Linux already exist.
#3 By: Kenneth Reitz, January 7th, 2014 15:39
I may be working on a secret project that may make this easier
#4 By: kermit666, January 7th, 2014 17:34
That sounds great! After all, the "sci" side of Python is one of its big strengths.
#5 By: brandon115, June 16th, 2014 23:18
If it helps anyone, I just released a custom buildpack for installing the latest version of NumPy (1.8.1) and SciPy (0.14.0). You can find it here: https://github.com/thenovices/heroku-buildpack-scipy
Please let me know if it's helpful or if you run into any issues! Also let me know if there's interest in seeing a write-up or tutorial for creating such a buildpack.
#6 By: marco542, August 6th, 2015 09:19
I am trying out the
conda-builpack by @kenneth. I need to run
sklearn on Python 3.4. However, the deployment does not succeed because the size limit of 300MB is exceeded by the slug (354MB).
This is what the
conda-requirements.txt look like:
# what is really required
As the above leads to over 300MB, is there something I can get rid of? Is there a better approach to use Python 3.4 and the scientific packages?