WIP tidy notebooks #1535

fonnesbeck · 2016-11-18T21:53:15Z

Making sure all the notebooks run on current master before release, and cleaning up (e.g. deleting stray cells and correcting typos).

Removed discrete_find_MAP.ipynb for now; we can replace it after 3.0 release if we can make it work.

GLM-logistic does not sample properly
GLM-model-selection takes over an hour to run. Perhaps not worth bundling with the examples?
survival_analysis.ipynb and pmf_pymc.ipynb both unacceptably slow as written. Need to revised or drop.
NUTS_scaling_using_ADVI hangs on the NUTS initialised using ADVI mean section and does not produce a trace for NUTS initialise MAP at test_point section
LKJ notebook mysteriously stops mixing.

Replaces #1465

ColCarroll · 2016-11-18T22:15:11Z

Easy check on a lot of these is whether they use the tqdm-style
progressbars in the ADVI tasks

On Fri, Nov 18, 2016 at 5:14 PM Peadar Coyle notifications@github.com
wrote:

@springcoil approved this pull request.

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
#1535 (review),
or mute the thread
https://github.com/notifications/unsubscribe-auth/ACMHEIGmWif1xoyhqWC_ZkaNhGe9StiYks5q_iM5gaJpZM4K29si
.

springcoil · 2016-11-19T13:34:35Z

======================================================================
ERROR: test_normal (pymc3.tests.test_examples.TestHierarchical)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/travis/build/pymc-devs/pymc3/pymc3/tests/test_examples.py", line 242, in test_normal
    start = pm.find_MAP()
  File "/home/travis/build/pymc-devs/pymc3/pymc3/tuning/starting.py", line 77, in find_MAP
    dlogp = bij.mapf(model.fastdlogp(vars))
  File "/home/travis/build/pymc-devs/pymc3/pymc3/model.py", line 159, in fastdlogp
    return self.model.fastfn(gradient(self.logpt, vars))
  File "/home/travis/build/pymc-devs/pymc3/pymc3/memoize.py", line 15, in memoizer
    cache[key] = obj(*args, **kwargs)
  File "/home/travis/build/pymc-devs/pymc3/pymc3/theanof.py", line 60, in gradient
    return tt.concatenate([gradient1(f, v) for v in vars], axis=0)
  File "/home/travis/build/pymc-devs/pymc3/pymc3/theanof.py", line 60, in <listcomp>
    return tt.concatenate([gradient1(f, v) for v in vars], axis=0)
  File "/home/travis/build/pymc-devs/pymc3/pymc3/theanof.py", line 48, in gradient1
    return tt.flatten(tt.grad(f, v, disconnected_inputs='warn'))
  File "/home/travis/minconda3/envs/testenv/lib/python3.5/site-packages/theano/gradient.py", line 553, in grad
    grad_dict, wrt, cost_name)
  File "/home/travis/minconda3/envs/testenv/lib/python3.5/site-packages/theano/gradient.py", line 1317, in _populate_grad_dict
    rval = [access_grad_cache(elem) for elem in wrt]
  File "/home/travis/minconda3/envs/testenv/lib/python3.5/site-packages/theano/gradient.py", line 1317, in <listcomp>
    rval = [access_grad_cache(elem) for elem in wrt]
  File "/home/travis/minconda3/envs/testenv/lib/python3.5/site-packages/theano/gradient.py", line 1272, in access_grad_cache
    term = access_term_cache(node)[idx]
  File "/home/travis/minconda3/envs/testenv/lib/python3.5/site-packages/theano/gradient.py", line 1106, in access_term_cache
    new_output_grads)
  File "/home/travis/minconda3/envs/testenv/lib/python3.5/site-packages/theano/gof/op.py", line 700, in L_op
    return self.grad(inputs, output_grads)
  File "/home/travis/minconda3/envs/testenv/lib/python3.5/site-packages/theano/tensor/elemwise.py", line 717, in grad
    sr = sr.dimshuffle(shuffle)
  File "/home/travis/minconda3/envs/testenv/lib/python3.5/site-packages/theano/tensor/var.py", line 363, in dimshuffle
    pattern)
  File "/home/travis/minconda3/envs/testenv/lib/python3.5/site-packages/theano/tensor/elemwise.py", line 175, in __init__
    (input_broadcastable, new_order))
ValueError: ('You cannot drop a non-broadcastable dimension.', ((True, False), ('x', 'x', 0)))
-------------------- >> begin captured logging << --------------------
pymc3: DEBUG: Applied interval-transform to sg and added transformed sg_interval_ to model.
pymc3: DEBUG: Applied interval-transform to s and added transformed s_interval_ to model.
--------------------- >> end captured logging << ---------------------
======================================================================
ERROR: test_sqlite (pymc3.tests.test_examples.TestHierarchical)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/travis/build/pymc-devs/pymc3/pymc3/tests/test_examples.py", line 248, in test_sqlite
    start = pm.find_MAP()
  File "/home/travis/build/pymc-devs/pymc3/pymc3/tuning/starting.py", line 77, in find_MAP
    dlogp = bij.mapf(model.fastdlogp(vars))
  File "/home/travis/build/pymc-devs/pymc3/pymc3/model.py", line 159, in fastdlogp
    return self.model.fastfn(gradient(self.logpt, vars))
  File "/home/travis/build/pymc-devs/pymc3/pymc3/memoize.py", line 15, in memoizer
    cache[key] = obj(*args, **kwargs)
  File "/home/travis/build/pymc-devs/pymc3/pymc3/theanof.py", line 60, in gradient
    return tt.concatenate([gradient1(f, v) for v in vars], axis=0)
  File "/home/travis/build/pymc-devs/pymc3/pymc3/theanof.py", line 60, in <listcomp>
    return tt.concatenate([gradient1(f, v) for v in vars], axis=0)
  File "/home/travis/build/pymc-devs/pymc3/pymc3/theanof.py", line 48, in gradient1
    return tt.flatten(tt.grad(f, v, disconnected_inputs='warn'))
  File "/home/travis/minconda3/envs/testenv/lib/python3.5/site-packages/theano/gradient.py", line 553, in grad
    grad_dict, wrt, cost_name)
  File "/home/travis/minconda3/envs/testenv/lib/python3.5/site-packages/theano/gradient.py", line 1317, in _populate_grad_dict
    rval = [access_grad_cache(elem) for elem in wrt]
  File "/home/travis/minconda3/envs/testenv/lib/python3.5/site-packages/theano/gradient.py", line 1317, in <listcomp>
    rval = [access_grad_cache(elem) for elem in wrt]
  File "/home/travis/minconda3/envs/testenv/lib/python3.5/site-packages/theano/gradient.py", line 1272, in access_grad_cache
    term = access_term_cache(node)[idx]
  File "/home/travis/minconda3/envs/testenv/lib/python3.5/site-packages/theano/gradient.py", line 1106, in access_term_cache
    new_output_grads)
  File "/home/travis/minconda3/envs/testenv/lib/python3.5/site-packages/theano/gof/op.py", line 700, in L_op
    return self.grad(inputs, output_grads)
  File "/home/travis/minconda3/envs/testenv/lib/python3.5/site-packages/theano/tensor/elemwise.py", line 717, in grad
    sr = sr.dimshuffle(shuffle)
  File "/home/travis/minconda3/envs/testenv/lib/python3.5/site-packages/theano/tensor/var.py", line 363, in dimshuffle
    pattern)
  File "/home/travis/minconda3/envs/testenv/lib/python3.5/site-packages/theano/tensor/elemwise.py", line 175, in __init__
    (input_broadcastable, new_order))
ValueError: ('You cannot drop a non-broadcastable dimension.', ((True, False), ('x', 'x', 0)))

Are these related to the other Theano errors we ran into before?

fonnesbeck · 2016-11-19T15:01:20Z

Yeah I only changed ipynb files so it's hard to see how any changes could have resulted in these errors.

ColCarroll · 2016-11-19T15:23:22Z

Can confirm that the same tests fail on master in the docker image.

springcoil · 2016-11-20T11:41:11Z

So it's unrelated to this converge, we've got some tests coming back. It looks to me like a theano error - what do you think @ColCarroll

ColCarroll · 2016-11-20T16:14:02Z

Yeah -- a rebase off master should fix the failure.

springcoil · 2016-11-20T16:28:23Z

The model selection stuff is useful, however it's included in the documentation right?

Cleanup of several notebooks Additional cleanup of notebooks Cleanup of additional notebooks Removed discrete_find_MAP.ipynb Cleanup of several notebooks Additional cleanup of notebooks Cleanup of additional notebooks WIP: Notebooks updated rugby analytics

springcoil · 2016-11-20T16:42:10Z

I rebased and reran the Rugby Analytics notebook.

springcoil · 2016-11-20T17:29:07Z

Ok the test errors are resolved. @fonnesbeck are you ready to merge yet?

fonnesbeck · 2016-11-20T18:27:20Z

Not yet, no. Will remove the WIP tag when done.

I'm tallying problems with the current set of notebooks, with respect to how they run locally from the current master. Several take hours to run, which probably means we should exclude them from the examples unless they can be revised.

Cleanup of several notebooks Additional cleanup of notebooks Cleanup of additional notebooks Removed discrete_find_MAP.ipynb Cleanup of several notebooks Additional cleanup of notebooks Cleanup of additional notebooks WIP: Notebooks updated rugby analytics

…_notebooks

fonnesbeck · 2016-11-23T19:53:25Z

The LKJ notebook appears broken. See the divergent traceplot in the current commit.

fonnesbeck · 2016-11-23T20:39:22Z

Other than the list at the top of the page, the notebooks look good in this PR. I'm happy to drop these notebooks for the release, unless an easy fix can be found (everything does not have to be a bundled example, after all). Please have a look.

springcoil · 2016-11-23T20:42:39Z

That's all fine by me. Happy that these notebooks are dropped for the
release. We can add them back in later.

On 23 Nov 2016 8:39 PM, "Chris Fonnesbeck" notifications@github.com wrote:

Other than the list at the top of the page, the notebooks look good in
this PR. I'm happy to drop these notebooks for the release, unless an easy
fix can be found (everything does not have to be a bundled example, after
all). Please have a look.

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
#1535 (comment), or mute
the thread
https://github.com/notifications/unsubscribe-auth/AA8DiH2VShNSQSW_6MWQhkR8IZvw6RZEks5rBKR7gaJpZM4K29si
.

twiecki · 2016-11-23T21:28:57Z

LKJ used Metropolis initially, maybe we should switch it back.

twiecki · 2016-11-23T21:29:43Z

pmf-pymc is slow but it's a big model and a great, comprehensive example. we shouldn't drop it.

springcoil · 2016-11-23T22:38:26Z

Actually that's a good point, since Probabilistic Matrix Factorization is a good use case for Bayesian Stats - maybe just adding a warning 'this will take a while'.

Cleanup of several notebooks Additional cleanup of notebooks Cleanup of additional notebooks Removed discrete_find_MAP.ipynb Cleanup of several notebooks Additional cleanup of notebooks Cleanup of additional notebooks WIP: Notebooks updated rugby analytics

twiecki · 2016-11-25T12:10:17Z

Rebased.

…gular errors.

… now that we auto-init in this manner.

twiecki · 2016-11-25T12:42:09Z

OK, I fixed the issues in the NBs and removed 2 (logistic and nuts scaling). I think we should keep pmf and survival in their current form -- they are slow but complex models so a longer running time is acceptable. Long term we should work on improving the inference to make those faster.

…amplers.

twiecki · 2016-11-25T13:05:50Z

I also converted the getting started NB to use auto-assigned step methods and NUTS auto-initialization. Makes the code and text much simpler.

fonnesbeck · 2016-11-25T16:00:36Z

I'm happy with this. Thanks for lending a hand.

fonnesbeck · 2016-11-25T16:07:17Z

In the future, we should probably avoid growing the bundled notebooks set any larger. Not everything has to be included in the repository. Might be worth setting up a site (repo/wiki?) to house a wider array of examples. There is a fair amount of overlap among the current set of notebooks, and several are not well-annotated, so new notebooks should probably displace an existing one, rather than expanding the set.

…_notebooks

This was referenced Nov 18, 2016

WIP notebooks cleanup #1465

Closed

Bump version to 3.0.rc3 #1458

Closed

springcoil approved these changes Nov 18, 2016

View reviewed changes

twiecki added the WIP label Nov 19, 2016

springcoil force-pushed the tidy_notebooks branch from 918932e to 4b8ac49 Compare November 20, 2016 16:38

fonnesbeck added 7 commits November 23, 2016 10:55

Removed discrete_find_MAP.ipynb

4a5a119

Cleanup of several notebooks

b7351e7

Additional cleanup of notebooks

36591a6

Cleanup of additional notebooks

b252441

More cleaned up notebooks (including divergent LKJ)

08a4a41

Merge branch 'tidy_notebooks' of github.com:pymc-devs/pymc3 into tidy…

65ffd6b

…_notebooks

fonnesbeck and others added 3 commits November 25, 2016 13:06

Cleanup of several notebooks

781a0aa

More cleaned up notebooks (including divergent LKJ)

cfadd07

DOC Rename pmf example.

e7d201b

twiecki force-pushed the tidy_notebooks branch from 65ffd6b to e7d201b Compare November 25, 2016 12:07

twiecki added 4 commits November 25, 2016 13:14

DOC Remove glm-logistic as it timeouts on reading data.

2c60807

DOC Revert LKJ model use Metropolis sampler as NUTS runs into non-sin…

8d42ce7

…gular errors.

DOC Remove NUTS scaling NB as it stops running and is mostly obsolete…

96dbd51

… now that we auto-init in this manner.

DOC Change some samplers in model selection NB to make it faster.

e8f49e7

twiecki removed the WIP label Nov 25, 2016

twiecki added 2 commits November 25, 2016 14:04

DOC Change getting started guide to use auto-init and auto-assigned s…

00171aa

…amplers.

DOC Remove link to paper as it's out-of-date.

1a11121

Merge branch 'tidy_notebooks' of github.com:pymc-devs/pymc3 into tidy…

746a296

…_notebooks

twiecki merged commit 8a59d87 into master Nov 28, 2016

junpenglao deleted the tidy_notebooks branch June 26, 2017 12:26

WIP tidy notebooks #1535

WIP tidy notebooks #1535

Uh oh!

Conversation

fonnesbeck commented Nov 18, 2016 • edited by twiecki Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ColCarroll commented Nov 18, 2016

Uh oh!

springcoil commented Nov 19, 2016

Uh oh!

fonnesbeck commented Nov 19, 2016

Uh oh!

ColCarroll commented Nov 19, 2016

Uh oh!

springcoil commented Nov 20, 2016

Uh oh!

ColCarroll commented Nov 20, 2016

Uh oh!

springcoil commented Nov 20, 2016

Uh oh!

springcoil commented Nov 20, 2016

Uh oh!

springcoil commented Nov 20, 2016

Uh oh!

fonnesbeck commented Nov 20, 2016 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

fonnesbeck commented Nov 23, 2016

Uh oh!

fonnesbeck commented Nov 23, 2016

Uh oh!

springcoil commented Nov 23, 2016

Uh oh!

twiecki commented Nov 23, 2016

Uh oh!

twiecki commented Nov 23, 2016

Uh oh!

springcoil commented Nov 23, 2016

Uh oh!

twiecki commented Nov 25, 2016

Uh oh!

twiecki commented Nov 25, 2016

Uh oh!

twiecki commented Nov 25, 2016

Uh oh!

fonnesbeck commented Nov 25, 2016

Uh oh!

fonnesbeck commented Nov 25, 2016

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

fonnesbeck commented Nov 18, 2016 •

edited by twiecki

Loading

fonnesbeck commented Nov 20, 2016 •

edited

Loading