Unable to install statsmodels using Python 3.4.1 :: Anaconda 2.0.1 (x86_64)

[Anthony DeMaio]

I tried to install statsmodels with easy_install and it did not work. I'm not sure if this is the right place to ask about this, but I figure it can't hurt. Here is the message the terminal spat out to me:

The package setup script has attempted to modify files on your system that are not within the EasyInstall build area, and has been aborted.

This package cannot be safely installed by EasyInstall, and may not support alternate installation locations even if you run its setup script by hand.  Please inform the package's author and the EasyInstall maintainers to find out if a fix or workaround is available.

Any help or advice is much appreciated. I am new to python and programming in general, so it's possible I am overlooking something, or just not understanding the error, which is why I'm having trouble figuring out a workaround. 

[josef...@gmail.com]

Which statsmodels version are you trying to install?

There were some changes after 0.5 that fixed some of those.  (*)

Doesn't Anaconda come with it's own installer that would work around this?

(*) one case was with importing matplotlib during the install via pandas.

Josef

[Allen Downey]

I just exchanged email with one of the Anaconda folks about this.  They are working with statsmodels 0.5.0, and seem to be aware that it is not working with Python 3.4.  So it sounds like they are waiting for the next version of statsmodels to fix the problem.

I am shuttling information, so I hope I am not misrepresenting Anaconda or statsmodels.  But I am interested in seeing this resolved, so let me know if I can help.

Allen

[Fabien]

Hi,

On 14.07.2014 17:13, Allen Downey wrote:
> They are working with statsmodels 0.5.0, and seem to be aware that it is
> not working with Python 3.4.

I also tried to install statsmodels with "pip3 install statsmodels" on a
python3.4 environment on linux with all required dependencies installed,
but it didn't work.

Just out of curiosity: how long will it take for statsmodels 0.6.0 to be
available on PyPi?

[josef...@gmail.com]

On Wed, Aug 13, 2014 at 7:55 AM, Fabien <fabien....@gmail.com> wrote:
> Hi,
>
>
> On 14.07.2014 17:13, Allen Downey wrote:
>>
>> They are working with statsmodels 0.5.0, and seem to be aware that it is
>> not working with Python 3.4.
>
>
> I also tried to install statsmodels with "pip3 install statsmodels" on a
> python3.4 environment on linux with all required dependencies installed, but
> it didn't work.

If you have (a very recent) cython and are set up for compiling
extensions, then you can pip install from github.

http://stackoverflow.com/questions/23343484/python-3-statsmodels/23519080#23519080

>
> Just out of curiosity: how long will it take for statsmodels 0.6.0 to be
> available on PyPi?

rough guess is 2 weeks for an rc to be on pipy (subject to unforseen problems)

Josef

>
>

[Fabien]

On 13.08.2014 14:06, josef...@gmail.com
wrote:

> If you have (a very recent) cython and are set up for compiling
> extensions, then you can pip install from github.

I did the:

pip3 install git+git://github.com/statsmodels/statsmodels

and it worked perfectly.

Thanks!

[Fabien]

On 13.08.2014 14:57, Fabien wrote:
> and it worked perfectly.

Ups, talked too fast.

Did something change in the way some ols attributes are computed? Some
of my tests failed on 0.6.0. Here are a few lines that reproduce the error:

Python 2.7.6 & statsmodels 0.5.0:
py> import statsmodels.api as sm
py> import numpy as np
py> x = np.random.rand(50)
py> y = np.random.rand(50,8)
py> ols = sm.OLS(x, y)
py> ols.rank
8

Python 3.4.0 & statsmodels 0.6.0.dev-2e806fc:
>>> import statsmodels.api as sm
>>> import numpy as np
>>> x = np.random.rand(50)
>>> y = np.random.rand(50,8)
>>> ols = sm.OLS(x, y)
>>> print(ols.rank)
None

[josef...@gmail.com]

IIRC, this is a consequence that we streamlined the rank calculation
(to avoid some duplicate linear algebra calculations).
I need to verify this.

rank is still available after calling fit()

Josef


>
>
>

[Fabien]

On 13.08.2014 15:55, josef...@gmail.com
wrote:

> IIRC, this is a consequence that we streamlined the rank calculation
> (to avoid some duplicate linear algebra calculations).
> I need to verify this.
>
> rank is still available after calling fit()

If it spares computations, is is actually an improvement.

This ols.rank stuff was just for my test suite, I can change this.

[stone...@gmail.com]

any new about the "rc" ?