1 / 10
Dec 2014

[color=#00AA00][b]Status of information: 2015-01-09[/b][/color]

If you want to submit a Python code, you actually can choose between [b]

5 Python versions

/b.
In addition to the well known differences between Python 2.x and Python 3.x there are some other things that are poorly or not at all documented.

[b]

Pyramid cluster (slow)

can use Python 2.7.9, Python 3.2.3 and Python 3.2.3 nbc here (I don't know the difference between the last two).
[color=#FF0000]You can't use PyPy 2.4 or Python 3.4[/color], even if you can select it from the list of languages. It leads to internal error.

[b]

Cube cluster (fast)

can use all offered Python versions. Moreover you can use the numpy/scipy module with all CPython versions.

  • created

    Dec '14
  • last reply

    Nov '16
  • 9

    replies

  • 3.3k

    views

  • 7

    users

This post should be pinned.

I would like to add that you sometimes could receive internal error on Cube using PyPy. Submiting the same code again (and again) should help. I have noticed this since last maintenance.

This is a very useful post and I agree that it should be pinned.
Thanks numerix!

Now we have Python 2.7.9 in Cube Cluster, with a memory signature of 7.6M and it is faster than previous version.
Do we really have newer version? Its almost similar to PyPy.

Edit: We do have a newer version and we can use numpy/scipy in it too.

Thanks for the hint. I updated the information in the top post.
Moreover the PyPy-problems ("internal error") mentioned by Lisensolikum have been fixed.

And: I did some version checking and figured out that "PyPy" in fact is a CPython 2.7.8.
That explains some major runtime differences between my local machine and SPOJ.

As PyPy is somewhat faster for some type of calculations, it would be great, if admins could offer a "real PyPy" ...

Something strange happened with numpy in [PYPY & Py3.4]

initially both were supporting numpy but now only Py 2.7.9 having numpy why so?

Probably "PYPY" is now real PYPY. I've tested it on Polish SPOJ problems and it acts very differently now. It is faster for some problems (AC on many problems that were previously undoable in Python!) and slower on other ones (especially on those which use many small test cases, because it always consumes minimum 0.02s).

You are right, now "PyPy" is real "PyPy" and numpy is no longer available for that version.
Thanks for the hint.

1 year later

where i can find library list? or its standard plus numpy/scipy on cube?
having any plans about Cython?