Stan's refactor #282

syclik · 2016-12-08T19:46:46Z

Summary:

This goes along with Stan's refactor. This updates PyStan so it's compatible with the changes to Stan.

Do not merge until stan-dev/stan#1751 is fixed.

Intended Effect:

Updates PyStan.

How to Verify:

rm pystan/*.so

python3 setup.py build_ext --inplace
python3
#exec(open("foo.py").read())


import pystan

schools_code = """
data {
    int<lower=0> J; // number of schools
    real y[J]; // estimated treatment effects
    real<lower=0> sigma[J]; // s.e. of effect estimates
}
parameters {
    real mu;
    real<lower=0> tau;
    real eta[J];
}
transformed parameters {
    real theta[J];
    for (j in 1:J)
    theta[j] = mu + tau * eta[j];
}
model {
    eta ~ normal(0, 1);
    y ~ normal(theta, sigma);
}
"""

schools_dat = {'J': 8,
               'y': [28,  8, -3,  7, -1,  1, 18, 12],
               'sigma': [15, 10, 16, 11,  9, 11, 10, 18]}



fit = pystan.stan(model_code=schools_code, data=schools_dat, verbose=True, seed=1, chains=1, algorithm = "NUTS", control = dict(adapt_engaged = True, metric = "dense_e"), sample_file = "abc.csv")

Side Effects:

None.

Documentation:

None.

Reviewer Suggestions:

@ariddell

Copyright and Licensing

Please list the copyright holder for the work you are submitting (this will be you or your assignee, such as a university or company):

Columbia University

By submitting this pull request, the copyright holder is agreeing to license the submitted work under the following licenses:

Code: GPLv3 (http://opensource.org/licenses/GPL-3.0)
Documentation: CC-BY 4.0 (https://creativecommons.org/licenses/by/4.0/)

syclik · 2016-12-09T16:34:09Z

pystan/misc.py

@@ -139,7 +139,7 @@ def _format_number(num, n_signif_figures, max_width):
    """
    if max_width < 6:
        raise NotImplementedError("Guaranteed formatting in fewer than 6 characters not supported.")
-    if math.isnan(num):
+    if math.isnan(num) or math.isinf(num):


This was failing when num was infinite. I don't know if this is the way to deal with this in python.

syclik · 2016-12-09T16:34:52Z

pystan/misc.py

@@ -637,7 +637,12 @@ def _get_valid_stan_args(base_args=None):

    args['init_radius'] = args.get('init_r', 2.0)
    if (args['init_radius'] <= 0):
-        args['init'] = "0".encode('ascii')
+        args['init'] = b"0"


This bit correctly initializes to 0. This was slightly different behavior in Stan for the same args, but doing this will make it the same from PyStan.

syclik · 2016-12-09T16:36:09Z

pystan/misc.py

@@ -941,6 +946,7 @@ def _writable_sample_file(file, warn=True, wfun=None):
    if wfun is None:
        wfun = lambda x, y: '"{}" is not writable; use "{}" instead'.format(x, y)
    dir = os.path.dirname(file)
+    dir = os.getcwd() if dir == '' else dir


When file is just the name of a file, os.path.dirname(file) returns '', which was failing on the next call os.access(dir, os.W_OK). By checking if it's '' and returning the current working directory, this now allows the files to get written to the working directory. Is that ok?

syclik · 2016-12-09T16:39:32Z

pystan/tests/test_rstan_stanfit.py

@@ -21,8 +21,8 @@ def test_init_zero_exception_inf_grad(self):
        """
        sm = StanModel(model_code=code)
        assertRaisesRegex = self.assertRaisesRegexp if PY2 else self.assertRaisesRegex
-        with assertRaisesRegex(RuntimeError, 'Gradient evaluated at the initial value is not finite.'):
-            sm.sampling(init='0', iter=1)
+        with assertRaisesRegex(RuntimeError, ''):


Now, the error message from Stan is written as a message and not in the exception message.

does the message get to the user somehow? What's the new exception message? (If there's an exception message, that could be checked here.)

I'm not sure -- I think the error is handled slightly differently now. Wow... I don't even remember changing these tests. Do you want me to hunt down the difference?

syclik · 2016-12-30T22:51:25Z

@ariddell, the code all runs, but I don't think I'm getting the output correctly propagated. Some of the output looks like all zeros.

But everything works. Can you review and let's talk about how we can get everything working?

syclik · 2016-12-30T22:51:39Z

by works, I mean that it runs and I get output back out to python.

ariddell · 2016-12-31T19:54:28Z

I'll keep an eye out. If there's something specific you saw, please open an issue and I'll take care of it.

…

On 12/30/2016 05:51 PM, Daniel Lee wrote: @ariddell <https://github.com/ariddell>, the code all runs, but I don't think I'm getting the output correctly propagated. Some of the output looks like all zeros. But everything works. Can you review and let's talk about how we can get everything working? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <stan-dev/pystan#282 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AABN7ge5tz2iuqgEgVGBe-tFGP42Ymm-ks5rNYrtgaJpZM4LINbQ>.

syclik · 2016-12-31T20:01:08Z

Thanks! I saw there was a test error on travis, so I'll try to fix that first. On Sat, Dec 31, 2016 at 2:54 PM, Allen Riddell <notifications@github.com> wrote:

…

I'll keep an eye out. If there's something specific you saw, please open an issue and I'll take care of it. On 12/30/2016 05:51 PM, Daniel Lee wrote: > @ariddell <https://github.com/ariddell>, the code all runs, but I don't > think I'm getting the output correctly propagated. Some of the output > looks like all zeros. > > But everything works. Can you review and let's talk about how we can get > everything working? > > — > You are receiving this because you were mentioned. > Reply to this email directly, view it on GitHub > <stan-dev/pystan#282 (comment)>, or > mute the thread > <https://github.com/notifications/unsubscribe- auth/AABN7ge5tz2iuqgEgVGBe-tFGP42Ymm-ks5rNYrtgaJpZM4LINbQ>. > — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <stan-dev/pystan#282 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAZ_Fzz1IxA8zRd6prIuzAkk2MEOUr_Fks5rNrL1gaJpZM4LINbQ> .

ariddell · 2017-01-09T02:38:26Z

Is this ready to review and merge?

syclik · 2017-01-09T02:41:08Z

Please review. As soon as stan-dev/stan#2004 has been merged, feel free to merge!

…

On Sun, Jan 8, 2017 at 9:38 PM, Allen Riddell ***@***.***> wrote: Is this ready to review and merge? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <stan-dev/pystan#282 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAZ_F0Modq8hl85Y3nceW47HJWAo73Ytks5rQZ2igaJpZM4LINbQ> .

ariddell · 2017-01-09T02:48:37Z

pystan/tests/test_user_inits.py

            pystan.stan(model_code=model_code, iter=10, chains=1, seed=2,
                        data=data, init=[dict(mu=4)], warmup=0)
+        except RuntimeError:


this might be redundant. a RuntimeError will yield a failing test.

Sure. Please remove.

ariddell · 2017-01-10T12:12:17Z

Great. I'll fix the tests and merge today.

syclik · 2017-01-10T14:15:59Z

Thanks. Let me know if you need more from me. I stepped away for a couple weeks and it all looks foreign.

…

On Tue, Jan 10, 2017 at 7:12 AM, Allen Riddell ***@***.***> wrote: Great. I'll fix the tests and merge today. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <stan-dev/pystan#282 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAZ_F-Ozh6-lMbcKWru6EIZsralLxvxIks5rQ3WhgaJpZM4LINbQ> .

ariddell · 2017-01-12T14:29:25Z

Hey @syclik, can we add a message to the throw domain_error here: https://github.com/stan-dev/stan/blob/1c2075bf8d0fe563b6907e40220aca016864e816/src/stan/services/util/initialize.hpp#L177

Python only knows that an exception has been encountered -- it doesn't know the exception type, so the "" isn't helpful. How about

throw std::domain_error("Initialization failed.");

edit: simplified things

ariddell · 2017-01-12T14:57:52Z

I made a pull request, stan-dev/stan#2203

syclik · 2017-01-12T16:02:39Z

Yes and thanks.

…

On Jan 12, 2017, at 9:57 AM, Allen Riddell ***@***.***> wrote: I made a pull request, #2203 — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.

ariddell · 2017-01-14T21:31:03Z

(Test failures were due to matplotlib not yet being available for Python 3.6)

syclik commented Dec 9, 2016

View reviewed changes

ariddell mentioned this pull request Dec 19, 2016

Variational method #163

Closed

syclik force-pushed the feature/refactor-services branch from 11eb2bb to 790afe4 Compare January 8, 2017 20:13

ariddell reviewed Jan 9, 2017

View reviewed changes

syclik added 15 commits January 11, 2017 18:10

Updating setup.py and pystan/stan

044e8aa

Updating mode.py

a792356

Getting the test_grad stuff working

37ced8d

Getting optimization working

ba47815

Getting all of pystan working

bc3708f

Getting all of pystan working

48ab612

Getting all of pystan working

71b78b9

Got most of it working

76ce862

Fixing inits

2801dba

Updating comments

28d36ba

adding in writing of comments

2e12247

updating misc

e533190

Updating test

5fe967e

Fixing exception handling

4c541b2

Updating test

861e388

syclik and others added 8 commits January 11, 2017 18:13

Fixing last test

3fc2346

Updating number of iterations

c62d420

Fixing test_grad

cb12b05

Fixing issue with writing files

fb9856f

Updating pystan

97124e4

Updating

cf1586f

Updates

01b2ef4

Fix cvodes path

adf02d8

ariddell force-pushed the feature/refactor-services branch from fad7bb3 to c400b59 Compare January 11, 2017 23:33

ariddell force-pushed the feature/refactor-services branch from c400b59 to adf02d8 Compare January 14, 2017 13:32

ariddell merged commit e9fc733 into develop Jan 14, 2017

ariddell deleted the feature/refactor-services branch January 14, 2017 21:30

ariddell mentioned this pull request Jan 14, 2017

Variational Bayes needs at least one test #295

Closed

ariddell mentioned this pull request Mar 13, 2017

Variational Bayes output #313

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Stan's refactor #282

Stan's refactor #282

syclik commented Dec 8, 2016

syclik Dec 9, 2016

syclik Dec 9, 2016

syclik Dec 9, 2016

syclik Dec 9, 2016

ariddell Jan 9, 2017

syclik Jan 10, 2017

syclik commented Dec 30, 2016

syclik commented Dec 30, 2016

ariddell commented Dec 31, 2016 via email

syclik commented Dec 31, 2016 via email

ariddell commented Jan 9, 2017

syclik commented Jan 9, 2017 via email

ariddell Jan 9, 2017

syclik Jan 10, 2017

ariddell commented Jan 10, 2017

syclik commented Jan 10, 2017 via email

ariddell commented Jan 12, 2017 •

edited

ariddell commented Jan 12, 2017 •

edited

syclik commented Jan 12, 2017 via email

ariddell commented Jan 14, 2017

Stan's refactor #282

Stan's refactor #282

Conversation

syclik commented Dec 8, 2016

Summary:

Intended Effect:

How to Verify:

Side Effects:

Documentation:

Reviewer Suggestions:

Copyright and Licensing

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

syclik commented Dec 30, 2016

syclik commented Dec 30, 2016

ariddell commented Dec 31, 2016 via email

syclik commented Dec 31, 2016 via email

ariddell commented Jan 9, 2017

syclik commented Jan 9, 2017 via email

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ariddell commented Jan 10, 2017

syclik commented Jan 10, 2017 via email

ariddell commented Jan 12, 2017 • edited

ariddell commented Jan 12, 2017 • edited

syclik commented Jan 12, 2017 via email

ariddell commented Jan 14, 2017

ariddell commented Jan 12, 2017 •

edited

ariddell commented Jan 12, 2017 •

edited