Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

SMC-ABC add distance, refactor and update notebook #3996

Merged
merged 10 commits into from
Jul 16, 2020

Conversation

aloctavodia
Copy link
Member

  • adds the wasserstein and energy distance functions
  • Refactors API, the distance, sum_stats and epsilon arguments are now passed pm.Simulator istead of pm.sample_smc, which I think is a more natural place.
  • Add random method to the pm.Simulator.
  • Add option to save the simulated data, as data is generated as part of the sampler it makes sense to have the option to save it, instead of computing it later (defaults to False).
  • Improves LaTeX representation
  • Finally, this also updates the SMC-ABC notebook and clarifies a few details on how to define and run an ABC model.

@review-notebook-app
Copy link

Check out this pull request on  ReviewNB

Review Jupyter notebook visual diffs & provide feedback on notebooks.


Powered by ReviewNB

@aloctavodia aloctavodia changed the title update notebook SMC-ABC add distance, refactor and update notebook Jul 3, 2020
Copy link
Member

@junpenglao junpenglao left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

Comment on lines 110 to 119
def gaussian_kernel(self, obs_data, sim_data):
return np.sum(-0.5 * ((obs_data - sim_data) / self.epsilon) ** 2)

raise NotImplementedError("Not implemented yet")
# we are assuming obs_data and sim_data are already sorted!
def wasserstein(self, obs_data, sim_data):
return np.mean(np.abs((obs_data - sim_data) / self.epsilon))

# we are assuming obs_data and sim_data are already sorted!
def energy(self, obs_data, sim_data):
return 1.4142 * np.mean(((obs_data - sim_data) / self.epsilon) ** 2) ** 0.5
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

recommend isolating it into separate functions instead of methods of a single class

self.distance = gaussian_kernel
elif distance == "wasserstein":
self.distance = wasserstein
sum_stat = "sort"
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can we do a assertion here or a warning that sum_stat need to be "sort" for wasserstein and energy?

raise NotImplementedError("Not implemented yet")
params = draw_values([*self.params], point=point, size=size)
if size is None:
return self.function(*params)
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

if self.function is a theano function it will return a tensor - maybe we can push the draw_values call to the end and return that?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

self.function is intended to be a Python function not a Theano one.

"""
This class stores a function defined by the user in python language.

function: function
Simulation function defined by the user.
params: list
Parameters passed to function.
distance: str or callable
Distance functions. Available options are "gaussian_kernel" (default), "wasserstein",
"energy" or a user defined function
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
"energy" or a user defined function
"energy" or a user defined function that takes epsilon (a scaler), and (the summary statistics of) observed_data, and simulated_data as input.

@codecov
Copy link

codecov bot commented Jul 7, 2020

Codecov Report

Merging #3996 into master will increase coverage by 0.11%.
The diff coverage is 90.36%.

Impacted file tree graph

@@            Coverage Diff             @@
##           master    #3996      +/-   ##
==========================================
+ Coverage   86.65%   86.77%   +0.11%     
==========================================
  Files          88       88              
  Lines       14090    14137      +47     
==========================================
+ Hits        12210    12267      +57     
+ Misses       1880     1870      -10     
Impacted Files Coverage Δ
pymc3/distributions/simulator.py 86.66% <85.71%> (+28.77%) ⬆️
pymc3/smc/sample_smc.py 87.50% <88.88%> (-0.66%) ⬇️
pymc3/smc/smc.py 99.48% <100.00%> (+5.73%) ⬆️

@aloctavodia aloctavodia merged commit 9e8975f into pymc-devs:master Jul 16, 2020
@aloctavodia aloctavodia deleted the abc_ref branch July 16, 2020 13:07
gmingas added a commit to alan-turing-institute/pymc3 that referenced this pull request Jul 22, 2020
* Update GP NBs to use standard notebook style (pymc-devs#3978)

* update gp-latent nb to use arviz

* rerun, run black

* rerun after fixes from comments

* rerun black

* rewrite radon notebook using ArviZ and xarray (pymc-devs#3963)

* rewrite radon notebook using ArviZ and xarray

Roughly half notebook has been updated

* add comments on xarray usage

* rewrite 2n half of notebook

* minor fix

* rerun notebook and minor changes

* rerun notebook on pymc3.9.2 and ArviZ 0.9.0

* remove unused import

* add change to release notes

* SMC: refactor, speed-up and run multiple chains in parallel for diagnostics (pymc-devs#3981)

* first attempt to vectorize smc kernel

* add ess, remove multiprocessing

* run multiple chains

* remove unused imports

* add more info to report

* minor fix

* test log

* fix type_num error

* remove unused imports update BF notebook

* update notebook with diagnostics

* update notebooks

* update notebook

* update notebook

* Honor discard_tuned_samples during KeyboardInterrupt (pymc-devs#3785)

* Honor discard_tuned_samples during KeyboardInterrupt

* Do not compute convergence checks without samples

* Add time values as sampler stats for NUTS (pymc-devs#3986)

* Add time values as sampler stats for NUTS

* Use float time counters for nuts stats

* Add timing sampler stats to release notes

* Improve doc of time related sampler stats

Co-authored-by: Alexandre ANDORRA <[email protected]>

Co-authored-by: Alexandre ANDORRA <[email protected]>

* Drop support for py3.6 (pymc-devs#3992)

* Drop support for py3.6

* Update RELEASE-NOTES.md

Co-authored-by: Colin <[email protected]>

Co-authored-by: Colin <[email protected]>

* Fix Mixture distribution mode computation and logp dimensions

Closes pymc-devs#3994.

* Add more info to divergence warnings (pymc-devs#3990)

* Add more info to divergence warnings

* Add dataclasses as requirement for py3.6

* Fix tests for extra divergence info

* Remove py3.6 requirements

* follow-up of py36 drop (pymc-devs#3998)

* Revert "Drop support for py3.6 (pymc-devs#3992)"

This reverts commit 1bf867e.

* Update README.rst

* Update setup.py

* Update requirements.txt

* Update requirements.txt

Co-authored-by: Adrian Seyboldt <[email protected]>

* Show pickling issues in notebook on windows (pymc-devs#3991)

* Merge close remote connection

* Manually pickle step method in multiprocess sampling

* Fix tests for extra divergence info

* Add test for remote process crash

* Better formatting in test_parallel_sampling

Co-authored-by: Junpeng Lao <[email protected]>

* Use mp_ctx forkserver on MacOS

* Add test for pickle with dill

Co-authored-by: Junpeng Lao <[email protected]>

* Fix keep_size for arviz structures. (pymc-devs#4006)

* Fix posterior pred. sampling keep_size w/ arviz input.

Previously posterior predictive sampling functions did not properly
handle the `keep_size` keyword argument when getting an xarray Dataset
as parameter.

Also extended these functions to accept InferenceData object as input.

* Reformatting.

* Check type errors.

Make errors consistent across sample_posterior_predictive and fast_sample_posterior_predictive, and add 2 tests.

* Add changelog entry.

Co-authored-by: Robert P. Goldman <[email protected]>

* SMC-ABC add distance, refactor and update notebook (pymc-devs#3996)

* update notebook

* move dist functions out of simulator class

* fix docstring

* add warning and test for automatic selection of sort sum_stat when using wassertein and energy distances

* update release notes

* fix typo

* add sim_data test

* update and add tests

* update and add tests

* add docs for interpretation of length scales in periodic kernel (pymc-devs#3989)

* fix the expression of periodic kernel

* revert change and add doc

* FIXUP: add suggested doc string

* FIXUP: revertchanges in .gitignore

* Fix Matplotlib type error for tests (pymc-devs#4023)

* Fix for issue 4022.

Check for support for `warn` argument in `matplotlib.use()` call. Drop it if it causes an error.

* Alternative fix.

* Switch from pm.DensityDist to pm.Potential to describe the likelihood in MLDA notebooks and script examples. This is done because of the bug described in arviz-devs/arviz#1279. The commit also changes a few parameters in the MLDA .py example to match the ones in the equivalent notebook.

* Remove Dirichlet distribution type restrictions (pymc-devs#4000)

* Remove Dirichlet distribution type restrictions

Closes pymc-devs#3999.

* Add missing Dirichlet shape parameters to tests

* Remove Dirichlet positive concentration parameter constructor tests

This test can't be performed in the constructor if we're allowing Theano-type
distribution parameters.

* Add a hack to statically infer Dirichlet argument shapes

Co-authored-by: Brandon T. Willard <[email protected]>

Co-authored-by: Bill Engels <[email protected]>
Co-authored-by: Oriol Abril-Pla <[email protected]>
Co-authored-by: Osvaldo Martin <[email protected]>
Co-authored-by: Adrian Seyboldt <[email protected]>
Co-authored-by: Alexandre ANDORRA <[email protected]>
Co-authored-by: Colin <[email protected]>
Co-authored-by: Brandon T. Willard <[email protected]>
Co-authored-by: Junpeng Lao <[email protected]>
Co-authored-by: rpgoldman <[email protected]>
Co-authored-by: Robert P. Goldman <[email protected]>
Co-authored-by: Tirth Patel <[email protected]>
Co-authored-by: Brandon T. Willard <[email protected]>
@kyleabeauchamp kyleabeauchamp added this to the 3.9.3 milestone Jul 28, 2020
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants