Stan 2.1.0 ASAP, other improvements later?

[Bob Carpenter]

I'd like to get 2.1.0 out the door ASAP because of
the bug in variance estimation in 2.0.1.

I believe CmdStan 2.1.0 is all ready to go (Daniel, is that right?).

I'd like to get a 2.1.0 release of RStan/PyStan out ASAP,
even if it means holding off some important features until
2.1.0.1. Is there an estimate of how long this will take and
whether it's feasible?

Andrew wanted to delay announcing the bug to the users because
we don't have 2.1.0 ready. Everyone seemed to think we should announce
2.1.0 for all platforms at the same time.

This is making me very uneasy and I don't think it's fair to our
users. If we can't get PyStan and RStan 2.1.0 out soon, my own preference
is to announce the bug with an example, release CmdStan 2.1.0, and tell
users to stop using Stan until 2.1.0 comes out or to go back to 1.3.0,
which didn't have this bug.

Sorry the timing is so bad w.r.t. the holidays. This wouldn't be
so urgent if we hadn't found the critical bug in the sampler in
2.0.1.

- Bob

[Daniel Lee]

On Mon, Dec 23, 2013 at 2:29 PM, Bob Carpenter <ca...@alias-i.com> wrote:

I'd like to get 2.1.0 out the door ASAP because of
the bug in variance estimation in 2.0.1.

I believe CmdStan 2.1.0 is all ready to go (Daniel, is that right?).

Yes. I'll tag CmdStan 2.1.0 now.
 

I'd like to get a 2.1.0 release of RStan/PyStan out ASAP,
even if it means holding off some important features until
2.1.0.1.  Is there an estimate of how long this will take and
whether it's feasible?

Andrew wanted to delay announcing the bug to the users because
we don't have 2.1.0 ready.  Everyone seemed to think we should announce
2.1.0 for all platforms at the same time.

I'm happy to tag 2.1.0 in Stan/CmdStan and wait until PyStan and RStan are ready.

I'm also happy to release CmdStan now and have PyStan and RStan catch up.

This is making me very uneasy and I don't think it's fair to our
users.  If we can't get PyStan and RStan 2.1.0 out soon, my own preference
is to announce the bug with an example, release CmdStan 2.1.0, and tell
users to stop using Stan until 2.1.0 comes out or to go back to 1.3.0,
which didn't have this bug.

I think announcing that there is a bug is sufficient. No real need to tell people to stop. The language still works. Once all interfaces are ready, we can really encourage users to move to 2.1.0 and not turn back.

 

Sorry the timing is so bad w.r.t. the holidays.  This wouldn't be
so urgent if we hadn't found the critical bug in the sampler in
2.0.1.

- Bob

--
You received this message because you are subscribed to the Google Groups "stan development mailing list" group.
To unsubscribe from this group and stop receiving emails from it, send an email to stan-dev+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

[Daniel Lee]

Bob, let's work on getting the release notes ready. I'll help comb through the 2.1.0 milestone to prepare it.

Here's the place where we need to update it in GitHub:
https://github.com/stan-dev/stan/releases/tag/v2.1.0

Daniel

[Ben Goodrich]

On Monday, December 23, 2013 2:29:21 PM UTC-5, Bob Carpenter wrote:

I'd like to get 2.1.0 out the door ASAP because of
the bug in variance estimation in 2.0.1.

I believe CmdStan 2.1.0 is all ready to go (Daniel, is that right?).

I think that is fine, but bugs like this are what hotfix releases are for. You just fix the bug and release, in which case RStan and PyStan just need a rebuild.

Ben

[Bob Carpenter]

Good point.

We didn't realize it was a bug in sampling warranting
a hotfix until well after it was fixed. And I'm still not
sure if Michael isolated what the cause was --- there was a
fix to the basic NUTS tree building, but I think Michael
ruled that out as what fixed the issue.

- Bob

[Michael Betancourt]

> And I'm still not
> sure if Michael isolated what the cause was --- there was a
> fix to the basic NUTS tree building, but I think Michael
> ruled that out as what fixed the issue.

No, I'm pretty sure that's what it was. It's just hard to nail down the exact commit because of how merges also merge the commit history.

[Bob Carpenter]

I'll add the release notes.

I'd like to get rid of the non-existent diag_pre_multiply and
diag_post_multiply functions from the doc (they're held up for
lack of testing).

I'm sure we'll have to violate ye olde Gitflow yet again. Maybe
Ben can suggest a way to cause minimal damage from here?

I think one problem we're having is that we've been over-eager to
create the release version before we've finalized the release on
develop.

Not sure how to coordinate Gitflow with RStan and PyStan.

- Bob

On 12/23/13, 3:58 PM, Daniel Lee wrote:
> Bob, let's work on getting the release notes ready. I'll help comb through the 2.1.0 milestone to prepare it.
>
> Here's the place where we need to update it in GitHub:
> https://github.com/stan-dev/stan/releases/tag/v2.1.0
>
>
>
> Daniel
>
>
>

> On Mon, Dec 23, 2013 at 3:49 PM, Daniel Lee <bea...@alum.mit.edu <mailto:bea...@alum.mit.edu>> wrote:

> stan-dev+unsubscribe@__googlegroups.com <mailto:stan-dev%2Bunsu...@googlegroups.com>.
> For more options, visit https://groups.google.com/__groups/opt_out <https://groups.google.com/groups/opt_out>.

>
>
>
> --
> You received this message because you are subscribed to the Google Groups "stan development mailing list" group.

> To unsubscribe from this group and stop receiving emails from it, send an email to stan-dev+u...@googlegroups.com.

[Bob Carpenter]

Great. I must've mistook not nailing down the exact commit with
not nailing down the basic cause. That did seem like a likely
culprit :-)

- Bob

[Allen Riddell]

I think I can get PyStan 2.1.0 working by Friday. I still need to debug the
handling of the new control options.

Going forward I'll create a branch that keeps in sync with stan develop (as
RStan does) so this doesn't take so long next time.

Best,

Allen

[Bob Carpenter]

On 12/24/13, 9:21 AM, Allen Riddell wrote:
> I think I can get PyStan 2.1.0 working by Friday. I still need to debug the
> handling of the new control options.

Great news.

> Going forward I'll create a branch that keeps in sync with stan develop (as
> RStan does) so this doesn't take so long next time.

We really need to work on making the code more
modular so PyStan, RStan and CmdStan don't themselves
need to change so much when features get added to the
samplers or optimizers or diagnostics.

And as Ben said, the fix to the sampler should've been
a hotfix, not in develop, and then there would've been
no rush at all to update.

- Bob

[Allen Riddell]

PyStan 2.1.0 is up on pypi (Python's CRAN).

On 12/23, Bob Carpenter wrote:

[Bob Carpenter]

Great. Thanks so much for getting this out.

We'll announce and update the web site when RStan's
ready to go.

- Bob

[Ben Goodrich]

I pushed the Hessian fix into develop but I was having trouble cherry-picking it into the release/v2.1.0 branch. It says

error: src refspec remotes/origin/release/v2.1.0 matches more than one.
error: failed to push some refs to 'https://github.com/stan-dev/rstan.git'

The command that needs to get run on the release branch is

git cherry-pick 063abbd

Ben

[Andrew Gelman]

I just want to thank everyone for their efforts on this!

[Jiqiang Guo]

Thanks. I will take care of it.  I think it just takes a bit more than building the rstan pkg for the releasing.

--
Jiqiang

[Jiqiang Guo]

i have uploaded 2.1.0 so in fact rstan 2.1.0 is up. 

Jiqiang

[Bob Carpenter]

Awesome.

Daniel and I can put together the release announcement
and update the web pages tomorrow (Friday).

- Bob

On 12/26/13, 5:09 PM, Jiqiang Guo wrote:
> i have uploaded 2.1.0 so in fact rstan 2.1.0 is up.
>
> Jiqiang
>
>
> On Thu, Dec 26, 2013 at 2:32 PM, Jiqiang Guo <guo...@gmail.com <mailto:guo...@gmail.com>> wrote:
>
> Thanks. I will take care of it. I think it just takes a bit more than building the rstan pkg for the releasing.
>
> --
> Jiqiang
>

> stan-dev+u...@__googlegroups.com.
> >> For more options, visit https://groups.google.com/__groups/opt_out
> <https://groups.google.com/groups/opt_out>.

> >
>
> --
> You received this message because you are subscribed to the Google Groups "stan development mailing list" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to

> stan-dev+u...@googlegroups.com <mailto:stan-dev%2Bunsu...@googlegroups.com>.