Added live_traceplot function #1934

davidbrochart · 2017-03-22T16:56:00Z

This allows to have live trace plots in the notebook, which can be useful when you want to see what is going on as you run the simulation.

twiecki · 2017-03-22T17:02:07Z

This is really cool. Will try to take a look soon.

twiecki · 2017-03-22T17:02:58Z

I wonder if perhaps the API should be:
pm.sample(..., live=True)

ColCarroll

This looks super cool, and a really nice use of the iter_sample. Will take a more substantive look later.

ColCarroll · 2017-03-22T16:58:11Z

pymc3/sampling.py

@@ -16,7 +16,7 @@
 import sys
 sys.setrecursionlimit(10000)

-__all__ = ['sample', 'iter_sample', 'sample_ppc', 'init_nuts']
+__all__ = ['assign_step_methods', 'sample', 'iter_sample', 'sample_ppc', 'init_nuts']


intentional?

Yes because iter_sample needs a step function, so if you don't have one already...

ColCarroll · 2017-03-22T17:02:48Z

pymc3/plots/traceplot.py

+                    varnames = get_default_varnames(trace, plot_transformed)
+                x0 = skip_first
+            elif (it - skip_first) % refresh_every == 0:
+                for i, v in enumerate(varnames):


I haven't looked closely, but do you think it is possible for this code to be factored out and shared with the base traceplot code?

Probably yes, it's a first shot.

davidbrochart · 2017-03-22T17:04:07Z

Yes, that could be an option, I wasn't sure.

fonnesbeck · 2017-03-22T18:08:05Z

I get a TypeError when I try to run the notebook:

TypeError: ufunc 'true_divide' output (typecode 'd') could not be coerced to provided output parameter (typecode 'l') according to the casting rule ''same_kind''

I agree with @twiecki that this should be a flag in sample rather than a new sampling function (even if a separate function is ultimately called underneath). The required additional arguments could be passed as kwargs.

aseyboldt · 2017-03-22T18:55:33Z

Maybe this is also a good time to mention that iter_sample has a bit of a performance problem. In each iteration it slices the whole trace. In the ndarray backend that is bad, but not terrible (when sampling a single normal it takes about 3 times longer), but in the sqlite and the hdf5 backend that means loading the whole thing into memory from scratch at each iteration.
Edit: Wrong place for this, sorry. It's #1935. Really neat idea with the live update by the way.

twiecki · 2017-03-22T19:00:03Z

@aseyboldt I hadn't realized that. Any ideas on how this could be improved?

davidbrochart · 2017-03-22T19:45:19Z

@fonnesbeck Sorry I merged a bit quickly, I'll fix and update with the feedback I get.

davidbrochart · 2017-03-22T20:24:43Z

@fonnesbeck I checked with Python2.7 and Python3.5, no error on my side. Has anyone else run the notebook?

aseyboldt · 2017-03-22T21:01:04Z

I get the same error TypeError.

aseyboldt · 2017-03-22T21:06:51Z

It works with this change:

diff --git a/pymc3/plots/utils.py b/pymc3/plots/utils.py
index b4ff6f5d..15694817 100644
--- a/pymc3/plots/utils.py
+++ b/pymc3/plots/utils.py
@@ -99,6 +99,6 @@ def fast_kde(x):

     norm_factor = n * dx * (2 * np.pi * std_x ** 2 * scotts_factor ** 2) ** 0.5

-    grid /= norm_factor
+    grid = grid / norm_factor

     return grid, xmin, xmax

davidbrochart · 2017-03-22T21:22:13Z

Yes I get the error when upgrading Numpy from 1.11.3 to 1.12.1.

twiecki · 2017-03-23T13:02:56Z

@davidbrochart You get the error with the fix @aseyboldt posted (which should already be on master, I think @aloctavodia did that)?

davidbrochart · 2017-03-23T13:05:46Z

@twiecki No, with @aseyboldt 's fix there is no error any more.

twiecki · 2017-03-23T13:07:03Z

Are you not on pymc3 master?

davidbrochart · 2017-03-23T13:31:29Z

Yes I am now, it works fine.

davidbrochart · 2017-03-23T19:57:05Z

I changed the API to what @twiecki suggested, and took into account @ColCarroll 's suggestion on sharing the traceplot code.

fonnesbeck · 2017-03-24T23:57:59Z

How does this work with a large model? It might be worth allowing a list of the subset of parameters to monitor.

davidbrochart · 2017-03-25T08:00:36Z

You can still pass a varnames list of variables to be plotted, all the parameters of the traceplot function are accepted.

twiecki · 2017-03-27T04:37:53Z

pymc3/plots/traceplot.py

@@ -7,7 +7,7 @@

 def traceplot(trace, varnames=None, transform=identity_transform, figsize=None, lines=None,
              combined=False, plot_transformed=False, grid=False, alpha=0.35, priors=None,
-              prior_alpha=1, prior_style='--', ax=None):
+              prior_alpha=1, prior_style='--', ax=None, live_plot=False, skip_first=0, refresh_every=100, roll_over=1000):


better formatting (new-lines so we don't exceed 80 chars).

twiecki · 2017-03-27T04:39:51Z

LGTM. Seems like you need to rebase.

* Expand sampler-stats.ipynb example include model diagnose from case study example in Stan http://mc-stan.org/documentation/case-studies/divergences_and_bias.html * Sampler Diagnose for NUTS * descriptive annotation and axis labels * Fix typos * PEP8 styling * minor updates 1, add example to examples.rst 2, original content in Markdown code block

…s/data/__init__.py

) * refactor module, add histogram * add more tests * refactor some code concerning AEVB histogram * fix test for histogram * use mean as deterministic point in Histogram * remove unused import * change names of shortcuts * add names to shared params * add new line at the end of `approximations.py`

* fix some svgd problems * switch -> ifelse * except in record

* add docs * delete redundant code * add usage example * remove unused import

* use only free RVs from trace * use memoize in Histogram.histogram_logp * Change tests for histogram

…pymc3 into live_sample_plots

davidbrochart · 2017-03-27T08:46:40Z

Not sure what I did here! Let me know if that's okay.

twiecki · 2017-03-27T09:35:46Z

@davidbrochart Seems like something went wrong, maybe this is helpful: https://github.com/edx/edx-platform/wiki/How-to-Rebase-a-Pull-Request

twiecki · 2017-03-27T09:36:05Z

Oh actually this seems fine, my bad.

twiecki · 2017-03-27T09:36:21Z

This is awesome, thanks!

Added live_traceplot function

9d5b20a

Cosmetic change

616a876

ColCarroll reviewed Mar 22, 2017

View reviewed changes

aseyboldt mentioned this pull request Mar 22, 2017

iter_sample is slow #1935

Closed

Merge branch 'master' into live_sample_plots

6f54f81

Changed the API to pm.sample(..., live_plot=True)

de127b8

twiecki reviewed Mar 27, 2017

View reviewed changes

Junpeng Lao and others added 4 commits March 27, 2017 10:21

DOC Change heading names.

8af47d8

Make install scripts idempotent (pymc-devs#1879)

07b3d15

Add examples of censored data models (pymc-devs#1870)

bc62f3e

fonnesbeck and others added 24 commits March 27, 2017 10:21

Added newline to MANIFEST

ce3a5e6

Replaced package list with find_packages in setup.py; removed example…

9025d7f

…s/data/__init__.py

Updated version to rc2

75da69c

Fixed stray version string

d6bd886

Fix indexing traces with steps greater one

0dbd346

SVGD problems (pymc-devs#1916)

3a0d654

* fix some svgd problems * switch -> ifelse * except in record

Histogram docs (pymc-devs#1914)

31d514d

* add docs * delete redundant code * add usage example * remove unused import

improve aesthetics

4240ea5

Bump theano to 0.9.0rc4 (pymc-devs#1921)

f6190fc

Histogram: use only free RVs from trace (pymc-devs#1926)

96ca0ac

* use only free RVs from trace * use memoize in Histogram.histogram_logp * Change tests for histogram

small fix to prevent a TypeError with the ufunc true_divide

33406e8

Bump theano to be at least 0.9.0

6f984b3

Add LKJCholeskyCov

4d531be

Fix log jacobian in LKJCholeskyCov

6a718fa

Add documentation for LKJCholeskyCov

34b716f

Add expand_packed_triangular

d8566f9

Add tests for LKJCholeskyCov

62835ae

Fix tests for py2

69aab74

Add floatX wrappers in test_advi

8326741

Changed the API to pm.sample(..., live_plot=True)

45d0887

Merge branch 'live_sample_plots' of https://github.com/davidbrochart/…

44b12e6

…pymc3 into live_sample_plots

Better formatting

0f782cf

Merged with lastest master

9083ac8

twiecki merged commit d2bca90 into pymc-devs:master Mar 27, 2017

jake-westfall mentioned this pull request Apr 5, 2017

Stan bambinos/bambi#67

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Added live_traceplot function #1934

Added live_traceplot function #1934

davidbrochart commented Mar 22, 2017

twiecki commented Mar 22, 2017

twiecki commented Mar 22, 2017

ColCarroll left a comment

ColCarroll Mar 22, 2017

davidbrochart Mar 22, 2017

ColCarroll Mar 22, 2017

davidbrochart Mar 22, 2017

davidbrochart commented Mar 22, 2017

fonnesbeck commented Mar 22, 2017

aseyboldt commented Mar 22, 2017 •

edited

Loading

twiecki commented Mar 22, 2017

davidbrochart commented Mar 22, 2017

davidbrochart commented Mar 22, 2017

aseyboldt commented Mar 22, 2017

aseyboldt commented Mar 22, 2017

davidbrochart commented Mar 22, 2017

twiecki commented Mar 23, 2017

davidbrochart commented Mar 23, 2017

twiecki commented Mar 23, 2017

davidbrochart commented Mar 23, 2017

davidbrochart commented Mar 23, 2017

fonnesbeck commented Mar 24, 2017

davidbrochart commented Mar 25, 2017 •

edited

Loading

twiecki Mar 27, 2017

twiecki commented Mar 27, 2017

davidbrochart commented Mar 27, 2017

twiecki commented Mar 27, 2017

twiecki commented Mar 27, 2017

twiecki commented Mar 27, 2017

Added live_traceplot function #1934

Added live_traceplot function #1934

Conversation

davidbrochart commented Mar 22, 2017

twiecki commented Mar 22, 2017

twiecki commented Mar 22, 2017

ColCarroll left a comment

Choose a reason for hiding this comment

ColCarroll Mar 22, 2017

Choose a reason for hiding this comment

davidbrochart Mar 22, 2017

Choose a reason for hiding this comment

ColCarroll Mar 22, 2017

Choose a reason for hiding this comment

davidbrochart Mar 22, 2017

Choose a reason for hiding this comment

davidbrochart commented Mar 22, 2017

fonnesbeck commented Mar 22, 2017

aseyboldt commented Mar 22, 2017 • edited Loading

twiecki commented Mar 22, 2017

davidbrochart commented Mar 22, 2017

davidbrochart commented Mar 22, 2017

aseyboldt commented Mar 22, 2017

aseyboldt commented Mar 22, 2017

davidbrochart commented Mar 22, 2017

twiecki commented Mar 23, 2017

davidbrochart commented Mar 23, 2017

twiecki commented Mar 23, 2017

davidbrochart commented Mar 23, 2017

davidbrochart commented Mar 23, 2017

fonnesbeck commented Mar 24, 2017

davidbrochart commented Mar 25, 2017 • edited Loading

twiecki Mar 27, 2017

Choose a reason for hiding this comment

twiecki commented Mar 27, 2017

davidbrochart commented Mar 27, 2017

twiecki commented Mar 27, 2017

twiecki commented Mar 27, 2017

twiecki commented Mar 27, 2017

aseyboldt commented Mar 22, 2017 •

edited

Loading

davidbrochart commented Mar 25, 2017 •

edited

Loading