1.0 Release

[WardBrian]

This is intended as the final item in the 1.0 release project

Oct. 19

• Run release scripts for version 1.0.0rc1
• 1.0.0rc1 on Conda

Early November

After #436

• Run release scripts for version 1.0.0rc2
• 1.0.0rc2 on Conda

One week later

• Run release scripts for version 1.0.0
• Verify doc updated on mc-stan.org and readthedocs.io
• (for Brian) update Conda after builds have propagated. Include tqdm as a dependency
• (optional) Post on forums/columbia stats blog
• (later) We should let Arviz/fbprophet teams know about stabilization of the API and changes. Arviz seemed more affected when I checked. There is additionally this list of dependent packages

[serban-nicusor-toptal]

Hey @WardBrian can I do this now or should I wait for the others (math, stan, cmdstan ) release first ?

[WardBrian]

IMO we can do it whenever

[WardBrian]

Did we decide whether or not to do a RC, @mitzimorris ?

[serban-nicusor-toptal]

Will 1.0.0rc1 be read fine in cmdstanpy/_version.py ?
On the RTD part we should be good with that naming

Edit: reading your post from another issue, pypi will do the right thing. Not sure about conda

[WardBrian]

Conda will be up to me to get right, or we just won't release the rc on conda and wait for 1.0 there

[serban-nicusor-toptal]

I think conda will be fine too, while reading this doc they point to the pypi versioning scheme

[WardBrian]

Yes, the conda-forge KB also says this https://conda-forge.org/docs/maintainer/knowledge_base.html#pre-release-builds

Should be possible, just have to make sure I get all the settings right when things propagate

[WardBrian]

I say we might as well run a 1.0.0rc1 build today. Not sure how long we will want to wait, a week maybe?

[serban-nicusor-toptal]

Usually we do the rc then code freeze for one week

[mitzimorris]

excellent plan - let's do it!

[serban-nicusor-toptal]

1.0.0rc1 started release pipeline

[serban-nicusor-toptal]

Release is now complete, RTD seems fine: https://readthedocs.org/projects/cmdstanpy/versions/

[WardBrian]

Seems like the actual site doesn't work properly: https://cmdstanpy.readthedocs.io/

[serban-nicusor-toptal]

After each release, they actually build the docs on their side so I would assume we need to wait a few minutes.

[serban-nicusor-toptal]

rtd build is done: https://cmdstanpy.readthedocs.io/en/v1.0.0rc1/

[mitzimorris]

looks good on my machine - many thanks!

[WardBrian]

RC1 is on conda-forge now

[mitzimorris]

downloaded from conda-forge, ran a notebook and exported results to arviz and no problems.

[mitzimorris]

looking at the dependencies - gamestonk terminal has detailed install instructions - we should do something similar.
step zero is obnoxious, but that's the domain, I guess -
https://github.com/GamestonkTerminal/GamestonkTerminal#local-install---anaconda-and-python

[WardBrian]

Weirdly it seems like they don't actually depend on us, at least not any more.

I wouldn't say their instructions are any more detailed than ours, they just require a lot of extra steps for their use case. I think we should link to the conda installation instructions on our page, but the rest of their instructions are unnecessary for us

[WardBrian]

Just to confirm, we're good to go for 1.0 release tomorrow? @mitzimorris

[mitzimorris]

I was slow to get the announcment up on the forums. right now I'm seeing only 16 downloads on conda-forge so far.
maybe we should wait until that number is closer to ??? 30? 40? 50?

[WardBrian]

Hard to say. PyPi (pip) is easier to install rcs through, but harder to get download stats out of, so it's unclear how many people are actually using it. Happy to wait another week as well, so that the release comes after our meeting with the CmdStanR devs

[mitzimorris]

I drew up an overview of everything deprecated that was removed - https://discourse.mc-stan.org/t/cmdstanpy-1-0-0rc1/24973/2?u=mitzimorris

next up - filing FYI issues on Prophet and Arviz et al.

[mitzimorris]

https://pepy.tech/project/CmdStanPy?versions=1.0.0rc1

[WardBrian]

Do we think 1.2 million downloads in the last month is correct? On the main CRAN mirror, rstan only got 126,755 downloads in the past 30 days, so this would imply our priors about where our userbase is are potentially completely wrong

Edit: matplotlib last 30 days: 33,046,236. ggplot2 last 30 days: 3,294,429

Maybe Python is just way more popular across the board?

[WardBrian]

Prophet does pin us, but it isn't downloaded that much:
https://pepy.tech/project/prophet

Really odd. I think the numbers are probably correct (Google BigData seems like they'd know how to count) but I have no clue where they'd be coming from. According to pepy (and any other site) we got more downloads this month than prophet and arviz combined by a factor of two

[mitzimorris]

no clue, but those are the stats. CmdStanPy and PyStan are both way up there. lotta Python users? lotta AWS instances running production pipelines?

[ahartikainen]

https://pepy.tech/project/fbprophet

[mitzimorris]

OK, after some digging, SO and pypi say that internal mirrors might explain what we're seeing -

python docs: https://packaging.python.org/guides/analyzing-pypi-package-downloads/

Highly inaccurate: A number of things prevent the download counts from being accurate, some of which include:

• Internal or unofficial mirrors (can both raise or lower download counts)

old SO post: https://stackoverflow.com/questions/9648015/pypi-download-counts-seem-unrealistic

FWIW, Google Colab notebooks now include, CmdStanPy (0.9.5), PyStan (2.19), and fbprophet .

[WardBrian]

I was looking at the package prophet, which looks like another clone of fbprophet? Strange

so yes, that probably is our number of downloads, just not the number of users most likely

[mitzimorris]

let's release 1.0 next Tuesday - Nov 2nd.
I'll put together a post for Andrew's blog together over the weekend and run it by you guys (including @ahartikainen!) beforehand.

[WardBrian]

Do we want to patch #486 first? It should be a very simple change

[mitzimorris]

sure - could you do this?

[ahartikainen]

Sounds good.

For arviz do we need to check anything special?

[WardBrian]

Note: We will now be doing a rc2 and delaying the release until #436

[serban-nicusor-toptal]

Please let me know if/when I need to run the rc2 or if I can help with anything here.

[WardBrian]

Thanks Nic, will do. It will probably be at least a couple days or so to get #436 good to go

[WardBrian]

So @mitzimorris how do you feel about cutting a RC2? If we mostly have doc left to write we can do that during the release period, but if we think there are more changes could be good to wait?

[mitzimorris]

Yes!!!!
please, let's cut a release and announce on Discourse

[mitzimorris]

@ahartikainen - noticed Arviz issue w/ cmdstanpy - is there anything needed from our side? arviz-devs/arviz#1646 (comment)

[WardBrian]

rc2 is out on pypi, conda will follow (probably tomorrow)

[mitzimorris]

slouching towards Beta - https://discourse.mc-stan.org/t/cmdstanpy-1-0-0rc2/25337

[WardBrian]

On conda now

[WardBrian]

#499 is simple but important enough that I think we should cut rc3 when it's patched. It won't need a conda version, as it only effects non-conda installs.

[WardBrian]

Do we want to release 1.0 this week? Latest rc will have been out for 2 weeks on Thursday

[mitzimorris]

yes!!!

[mitzimorris]

@rok-cesnovar @jgabry - OK if we do a 1.0 CmdStanPy release tomorrow? (Dec 1)
where is CmdStanR at?

[WardBrian]

I think we should do it tomorrow (Dec 2) so that the RC had a full 2 weeks.

We should also note in the release forum/blog post that 1.0.0 is supported on Python 3.6 through 3.10, but future versions will require 3.7 (I think it's pretty unlikely we fit in a 1.1 in the next 20 days before 3.6 hits EOL)

[WardBrian]

Here goes nothing https://github.com/stan-dev/cmdstanpy/actions/runs/1531682007

[mitzimorris]

here's a blurb - it's a bit dry:

Announcing version 1.0 of CmdStanPy!

CmdStanPy is now available via Conda Forge: https://anaconda.org/conda-forge/cmdstanpy
and PyPi: https://pypi.org/project/cmdstanpy/

Reasons to use CmdStanPy:

1. Runs with latest Stan release

2. Easy to install: CmdStanPy can automatically CmdStan and the underlying C++ toolchains

3. Scalable: can fit big models and big dataset - if CmdStan can fit a model and dataset, so can CmdStanPy.

4. Cross-platform: runs on Linux, macOS, and Windows.

5. Non-restrictive lisence: BSD-3 license: https://github.com/stan-dev/cmdstanpy/blob/develop/LICENSE.md

CmdStanPy provdes access to the Stan compiler, inference algorithms, and diagnostics.
It supports both development and production workflows and is well-suited for teaching and learning Stan.

[WardBrian]

I think the blurb is good! I'd also add a link to the repo and docs homepage