statsmodels : 0.09, when???

[Palmer Ristevski]

Hi there,

I see in these emails that 0.09 is out, supposedly,

BUT the website still says version : 0.08.

Does anyone know when it will be formally out?

P

On Mon, May 7, 2018 at 2:25 PM, Kevin Sheppard <kevin.k....@gmail.com> wrote:

I'm a big fan of release from master, and then, if you really want a bug fix branch, you can create a branch and backport as necessary.

Versioneer is pretty widely used including by pandas.

The point of release engineering is to make releasing as pain-free as possible, as well as to reduce errors (like tagging at the wrong time, etc). I agree that anything more than 6 months would be asking a lot, but it would be good to get contributions out more frequently. 

On Mon, May 7, 2018 at 2:47 PM Tom Augspurger <tom.augs...@gmail.com> wrote:

On Mon, May 7, 2018 at 8:33 AM, <josef...@gmail.com> wrote:

On Mon, May 7, 2018 at 8:25 AM, Kevin Sheppard <kevin.k....@gmail.com> wrote:

It would help releasing a lot if some time was given to doing some release engineering.  With some care -- setting up doctr, setting up a statsmodels-wheels repo, switching to versioneer -- it is possible to effectively automate the entire release as a few steps

1. Tag the release

2. Travel will build docs on a tag and upload to the right place on github

2. Push a commit to statsmodels-wheels with the tag that will build the wheels and upload to PyPI.

I've done this for randomgen and producing ~ 12 wheels across 3 OSes + docs is nearly painless.  

Creating docs and wheels automatically would for sure be helpful.

Is versioneer now reliable and robust enough? This would be more critical because it changes the setup.py.

About release policy in general:

I think we cannot release more often than semi-annually because it allows us some time to fix new merges in master before a release.

If doing a release itself requires less work, then it would be possible to make bug-fix point releases which would especially help with compatibility fixes.

We still need maintenance branches for the releases to be able to go from an rc to the final release.

Josef

Agreed with respect to release engineering.

For maintenance, there are tools like https://github.com/MeeseeksBox/MeeseeksDev (developed and used by Jupyter) to backport changes to release branches.

 

 

On Fri, May 4, 2018 at 2:16 AM Brock Mendel <jbrock...@gmail.com> wrote:

I'm largely indifferent to the actual minimum versions, but _really_ would like for them to specified in one and only one place.  At the moment they are specified (differently) in setup.py and INSTALL.txt.

[josef...@gmail.com]

On Mon, May 7, 2018 at 5:30 PM, Palmer Ristevski <palm...@gmail.com> wrote:

Hi there,

I see in these emails that 0.09 is out, supposedly,

BUT the website still says version : 0.08.

Does anyone know when it will be formally out?

I released the rc a week ago, https://pypi.org/project/statsmodels/0.9.0rc1/

I'm planning to release the final 0.9.0 at the end of this week.

There hasn't been any feedback on problems with the rc1, but there are a few bugfixes and one numpy compatibility fix that will be included in 0.9 final.

The website hasn't been updated yet.

Josef

[Palmer Ristevski]

OK,

Thanks Josef.

SO basically it will be at the end of the week.

Thats what I wanted to know.

Regards,

P