[Dependancy] Propose introduction of dependancy on Numba

[mmcky]

Given numba is maturing quickly and hasn't presented any major issues thus far when using the library. Is it the right time to introduce numba as a strict dependancy of the package?

1. Allows for the tailored use of numba when writing new code for the library.
2. Reduce the need to maintain multiple versions and use of replacement jit functions. Older numpy or pure python versions could be moved to another repo or a teaching subpackage for future reference or for the purposes of teaching.
3. Everyone I know is using anaconda which comes with numba automatically.

Technically it is also already in pip-requirements.txt.

[sglyon]

I'm totally in favor of a strict numba dependency. +1 from me

[cc7768]

I am also in favor of this.

[oyamad]

I am strongly in favor of this.

[davidrpugh]

Agreed.

On Wed, Aug 26, 2015 at 4:46 AM, Daisuke Oyama notifications@github.com
wrote:

I am strongly in favor of this.

—
Reply to this email directly or view it on GitHub
#173 (comment)
.

[jstac]

I'm in favor too. As @mmcky says, numba has proved stable, and any dependency hassle for users is more than offset by a higher performance library, which will in turn benefit users and developers.

@albop are you happy for us to go ahead with this?

I don't believe there's a need for a teaching subpackage. Teaching related material can go in the examples dir of the repo or in a separate repo.

[mmcky]

I think this issue can be closed. All seem to be in favour of listing numba as a dependancy. The use of the anaconda distribution means numba comes packaged.

Outcome: Adopted

[moldovean]

import quantecon as qe no longer works and I can't install numba

[mmcky]

@moldovean Are you using the latest anaconda with Python 3.5?

If you are the current anaconda release doesn't come with numba packaged for python 3.5. It is available using conda install numba - and hopefully anaconda will be updated shortly.