A library that allows for inference on probabilistic models

Meta Research

Last update: Dec 29, 2022

Related tags

Deep Learning beanmachine

Overview

Bean Machine

Overview

Bean Machine is a probabilistic programming language for inference over statistical models written in the Python language using a declarative syntax. Bean Machine is built on top of PyTorch and Bean Machine Graph, a custom C++ backend. Check out our tutorials and Quick Start to get started!

Installation

Bean Machine supports Python 3.7, 3.8 and PyTorch 1.10.

Install the Latest Release with Pip

pip install beanmachine

Install from Source

To download the latest Bean Machine source code from GitHub:

git clone https://github.com/facebookresearch/beanmachine.git
cd beanmachine

Then, you can choose from any of the following installation options.

Anaconda

We recommend using conda to manage the virtual environment and install the necessary build dependencies.

conda create -n {env name} python=3.7; conda activate {env name}
conda install boost eigen
pip install .

Docker

docker build -t beanmachine .
docker run -it beanmachine:latest bash

Validate Installation

If you would like to run the builtin unit tests:

# install pytest 7.0 from GitHub
pip install git+https://github.com/pytest-dev/[email protected]
pytest .

License

Bean Machine is MIT licensed, as found in the LICENSE file.

Comments

Better arviz support in base inference
Motivation

When I was doing bug fixing I noticed my arviz changes were lost in the shuffle. These commits add the functionality while keeping the APIs unchanged as much as possible.

Changes proposed

Changes the

Test Plan

Please provide clear instructions on how the changes were verified. Attach screenshots if applicable.

Types of changes

[ ] Docs change / refactoring / dependency upgrade

[ ] Bug fix (non-breaking change which fixes an issue)

[x] New feature (non-breaking change which adds functionality)

[ ] Breaking change (fix or feature that would cause existing functionality to change)

Checklist

[x] My code follows the code style of this project.

[ ] My change requires a change to the documentation.

[ ] I have updated the documentation accordingly.

[x] I have read the CONTRIBUTING document.

[x] I have added tests to cover my changes.

[x] All new and existing tests passed.

[x] The title of my pull request is a short description of the requested changes.

CLA Signed
opened by zaxtax 55
Adds Hierarchical Modeling tutorial notebook

Motivation

This PR presents the first beanmachine tutorial notebook from the OpenTeams/Quansight team.

Changes proposed

This embeds a new notebook within the tutorials directory under the subdirectory openteams (can be moved/renamed if necessary).
CLA Signed Merged

opened by dhavide 28
Multiple random seed runs introduced to single_site_random_walk_conjugate_test_nightly.py

Summary: To help defend against potential flakiness and differences in the behavior of the random number generators across platform, this diff is adding multiple random seed runs to the test file mentioned above.

Differential Revision: D28516365
CLA Signed fb-exported

opened by wtaha 27
Having error messages report random_seed

Summary: We have been having some trouble with deterministic tests producing different results on phabricator than on our machines. This diff extends the messages of the relevant assert statements so that we also have information about the random_seed at the time of the test failure.

Differential Revision: D28517313
CLA Signed fb-exported

opened by wtaha 26
handling plot category

Summary: Add support for trace plot and autocorrelation plots for sampled variables . User can override these functions or add new plot functionalities by registering these new methods via plot_fn(...)

Differential Revision: D17911632
CLA Signed Merged

opened by torabin 26
Extending tests to be better localize problem in single_site_ancestral_mh_conjugate_test_nightly.py

Summary: This diff introduces a set of ten seeds to run the various tests on. The distant_normal case fails on all ten values. This further supports the idea that the issue is with distant_normal_normal case and not necessarily the algorithm itself. In all cases it fails with the error indicated in the comment on the test (low ESS, and exactly the same value on both seeds). This suggests that the algorithm is consistently getting stuck on this case.

Differential Revision: D28519753
CLA Signed fb-exported

opened by wtaha 25
Factoring out Normal log density and Poisson log probability

Summary: In preparation for reuse of Normal log density and Poisson log probability, they are factored out to util.h.

Differential Revision: D39178764
CLA Signed fb-exported

opened by rodrigodesalvobraz 24
GP tutorial NumPy error
The Gaussian_Process_Gpytorch tutorial has been updated for conversion to MDX. See #1392 for more info about the updates. Only minor edits were made including:

link updates

markdown display of dataframes

using ArviZ for summary statistics

spelling and grammar checks

In cells 12 and 13, we attempt to show the summary statistics from ArviZ for the model. The following error occurs when converting the Bean Machine MonteCarloSamples object to either an xarray or ArviZ InferenceData object.

RuntimeError: Can't call numpy() on Tensor that requires grad. Use tensor.detach().numpy() instead.

This error is coming from the to_inference_data method, and can be seen in the stack trace in the notebook.

Types of changes

[ ] Docs change / refactoring / dependency upgrade

[ ] Bug fix (non-breaking change which fixes an issue)

[ ] New feature (non-breaking change which adds functionality)

[ ] Breaking change (fix or feature that would cause existing functionality to change)

Checklist

[x] My code follows the code style of this project.

[ ] My change requires a change to the documentation.

[ ] I have updated the documentation accordingly.

[x] I have read the CONTRIBUTING document.

[ ] I have added tests to cover my changes.

[ ] All new and existing tests passed.

[x] The title of my pull request is a short description of the requested changes.

bug CLA Signed
opened by ndmlny-qs 24
Marginal diagnostic tool
This commit includes the marginal 1D diagnostic tool with JavaScript callbacks.

Motivation

This PR completes one tool that uses Bokeh and JavaScript callbacks in order to create an interactive tool that can be viewed in Jupyter. This refactors the code in PR #1631 heavily, since pure Python callbacks were found to not function properly with internal tools.

Changes proposed

A new tool folder in the diagnostics folder contains the proposed changes. In this folder there is a js folder that contains all the JavaScript callbacks needed for the Bokeh tool. The tool creates plots of marginal distributions for each random variable of the model. The output is a self-contained HTML object that can be rendered in Jupyter without any external CDN calls for JS resources.

Test Plan

Unit tests for the Python and JavaScript will be done at a later commit. Right now the testing was to run the tool in the Coin Flipping tutorial, and to inspect the output and ensure only static resources were used.

Types of changes

[ ] Docs change / refactoring / dependency upgrade

[ ] Bug fix (non-breaking change which fixes an issue)

[x] New feature (non-breaking change which adds functionality)

[ ] Breaking change (fix or feature that would cause existing functionality to change)

Checklist

[x] My code follows the code style of this project.

[ ] My change requires a change to the documentation.

[ ] I have updated the documentation accordingly.

[x] I have read the CONTRIBUTING document.

[ ] I have added tests to cover my changes.

[ ] All new and existing tests passed.

[x] The title of my pull request is a short description of the requested changes.

TODO

[ ] Python unit tests

[ ] JavaScript unit tests

[ ] Figure out if the build should run npm run build for the tools, or if we should just have the minified code for the JS callbacks in the code base.

enhancement CLA Signed
opened by ndmlny-qs 23
ipynb to mdx
This commit contains the following changes.

Updates to the tutorials plotting module. Changes here were propagated to the tutorials using them. They include baseball.py and nba.py.

Tutorials were modified so they can accommodate conversion to mdx.

The following tutorials have been updated to include interactive figures:

Bayesian_Logistic_Regression

Robust_Linear_Regression

Sparse_Logistic_Regression

The website/tutorials.json file was updated to reflect the tutorials that can be successfully transformed into mdx files.

Resolves #1346

Types of changes

[x] Docs change / refactoring / dependency upgrade

[ ] Bug fix (non-breaking change which fixes an issue)

[x] New feature (non-breaking change which adds functionality)

[ ] Breaking change (fix or feature that would cause existing functionality to change)

Checklist

[x] My code follows the code style of this project.

[x] My change requires a change to the documentation.

[x] I have updated the documentation accordingly.

[x] I have read the CONTRIBUTING document.

[ ] I have added tests to cover my changes.

[ ] All new and existing tests passed.

[x] The title of my pull request is a short description of the requested changes.

Bugs

General

Remove tqdm progress bars in the mdx output

Tutorials

Coin flipping

:heavy_check_mark: runs in Colab

plots visible before execution in Colab

:heavy_check_mark: matplotlib

:x: plotly

Create a ticket to update with interactive figures.

Linear regression

Contains errors, create a ticket.

Robust linear regression tutorial

:x: runs in Colab (needs updated plots.py module)

plots visible before execution in Colab

:heavy_check_mark: Bokeh

Logistic regression tutorial

:x: runs in Colab (needs updated plots.py module)

plots visible before execution in Colab

:heavy_check_mark: Bokeh

Sparse logistic regression tutorial

:x: runs in Colab (needs updated plots.py module)

plots visible before execution in Colab

:heavy_check_mark: Bokeh

Hierarchical regression

:heavy_check_mark: runs in Colab

plots visible before execution in Colab

:heavy_check_mark: Bokeh

Create a ticket to ensure all figures are interactive.

Hierarchical modeling

:heavy_check_mark: runs in Colab

plots visible before execution in Colab

:heavy_check_mark: Bokeh

Create a ticket to ensure all figures are interactive.

IRT tutorial

:heavy_check_mark: runs in Colab

plots visible before execution in Colab

:heavy_check_mark: Bokeh

Create ticket to update displays of Markdown dataframes.

Zero inflated count data

:heavy_check_mark: runs in Colab

plots visible before execution in Colab

:heavy_check_mark: Bokeh

Create a ticket to ensure all figures are interactive.

Hidden Markov model

:heavy_check_mark: runs in Colab

plots visible before execution in Colab

:heavy_check_mark: matplotlib

Create a ticket to update with interactive figures.

Gaussian mixture model

:heavy_check_mark: runs in Colab

plots visible before execution in Colab

:x: Plotly

Create a ticket to update with interactive figures.

Neal's funnel

:heavy_check_mark: runs in Colab

plots visible before execution in Colab

:heavy_check_mark: matplotlib

:x: plotly

Create a ticket to update with interactive figures.

Gaussian process using gpytorch

Contains errors, create a ticket.

documentation CLA Signed
opened by ndmlny-qs 23
Adding Item Response Theory tutorial
Motivation

Please describe your motivation for the changes. Provide link to any related issues.

Changes proposed

Outline the proposed changes and alternatives considered.

Test Plan

Please provide clear instructions on how the changes were verified. Attach screenshots if applicable.

Types of changes

[ ] Docs change / refactoring / dependency upgrade

[ ] Bug fix (non-breaking change which fixes an issue)

[x] New feature (non-breaking change which adds functionality)

[ ] Breaking change (fix or feature that would cause existing functionality to change)

Checklist

[x] My code follows the code style of this project.

[ ] My change requires a change to the documentation.

[ ] I have updated the documentation accordingly.

[x] I have read the CONTRIBUTING document.

[ ] I have added tests to cover my changes.

[ ] All new and existing tests passed.

[x] The title of my pull request is a short description of the requested changes.

CLA Signed Merged
opened by zaxtax 23
Fix GP unit test

Summary: Turning off fast computations in BoTorch broke this test. The MVN covar had 1 more batch dimension than loc, which the approximate log_prob was happy with but the exact log_prob did not like.

Reviewed By: Balandat

Differential Revision: D42194245
CLA Signed fb-exported

opened by saitcakmak 1
Remove arviz as a dependency
Motivation

This removes arviz as an explicit dependency of Bean Machine. This will also prevent a circular dependency with arviz.

Changes proposed

Updates setup.py and moves import into to_inference_data

Types of changes

[ ] Docs change / refactoring / dependency upgrade

[x] Bug fix (non-breaking change which fixes an issue)

[ ] New feature (non-breaking change which adds functionality)

[ ] Breaking change (fix or feature that would cause existing functionality to change)

Checklist

[x] My code follows the code style of this project.

[ ] My change requires a change to the documentation.

[ ] I have updated the documentation accordingly.

[x] I have read the CONTRIBUTING document.

[ ] I have added tests to cover my changes.

[x] All new and existing tests passed.

[x] The title of my pull request is a short description of the requested changes.

CLA Signed
opened by zaxtax 1
Connect marginalization to Graph API

Summary: We introduce two new values to InferenceType enum for automatic discrete marginalization (AMD) so AMD can be ran from the normal Graph inference API.

Differential Revision: D41060415
CLA Signed fb-exported

opened by rodrigodesalvobraz 1
Point stat choice
Resolves #1817

Motivation

Allow a user to select which point statistic they want displayed in the Marginal 1D tool: mean or median.

Changes proposed

Added method to calculate the median of an array in JavaScript.

Updated the Marginal 1D tool to use the new method.

Updated the Marginal 1D interfaces to include a new widget component.

Updated the Python side of the tool to render the new widget.

Updated Python TypedDict objects to include docstrings for inclusion in the documentation in the future.

Updated the coin flipping tutorial to use the new point statistic button.

Test Plan

Updated the coin flipping tutorial, and visually inspected the tool worked.

Built the documentation locally for testing the new docstrings in Sphinx.

Built the Docusaurus documentation locally to ensure the new widgets worked.

new-widget-check.webm

Types of changes

[ ] Docs change / refactoring / dependency upgrade

[ ] Bug fix (non-breaking change which fixes an issue)

[x] New feature (non-breaking change which adds functionality)

[ ] Breaking change (fix or feature that would cause existing functionality to change)

Checklist

[x] My code follows the code style of this project.

[ ] My change requires a change to the documentation.

[ ] I have updated the documentation accordingly.

[x] I have read the CONTRIBUTING document.

[ ] I have added tests to cover my changes.

[ ] All new and existing tests passed.

[x] The title of my pull request is a short description of the requested changes.

enhancement CLA Signed
opened by ndmlny-qs 2

Releases(v0.2.0)

v0.2.0(Sep 6, 2022)
Full Changelog: https://github.com/facebookresearch/beanmachine/compare/v0.1.2...v0.2.0

New Features

Graduated VI from experimental (#1609)

Added ADVI and MAP inference as support for variational inference methods

Graduated NNC from experimental (#1618)

Allows the use of Functorch’s AOT compiler by default for HMC and NUTS

If working with a non-static model or unexpected errors are encountered, you may need to manually disable the nnc_compile flag.

Changes

VerboseLevel in infer is deprecated. Users should use the new boolean argument show_progress_bar to control whether the tqdm progress bar is displayed or not ( #1603).

Fixes

HMC/ NUTS throws an exception when the step size becomes zero (#1606)

Random variables warn users against using torch tensors as arguments (#1639)

Documentations

Added VI static documentation (#1613)

Added NNC static documentation (#1619)

Added VI PPCA tutorial (#1617)

Added VI tutorial demonstrating ADVI against Gaussian (perfect) and Gamma (approximation gap) targets (#1621)

Added VI tutorial replicating the Tensorflow probability GLMM tutorial (#1622)

Addded VI tutorial demonstrating MAP on Bayesian linear regression and how it coincides with Tikhonov regularization (with Gaussian prior) and LASSO (with Laplace prior) (#1623)

Source code(tar.gz)
Source code(zip)
v0.1.2(Jul 6, 2022)
Full Changelog: https://github.com/facebookresearch/beanmachine/compare/v0.1.1...v0.1.2

New Features

Supports accelerated inference on HMC and NUTS with functorch’s Neural Network Compiler (NNC), which can be controlled setting nnc_compile flag when initializing an inference method (#1385) (Docs)

Supports parallel sampling when number of chains > 1, which can be controlled by setting run_in_parallel flag when calling infer (#1369)

Added progress bar to BMGInference (#1321)

MonteCarloSamples object returned from an inference will contain log likelihood and observations now (#1269)

Reworked bm.simulate, which accepts a dictionary of posterior as inputs as well (#1474)

Binary wheels for M1 Apple Silicon and Python 3.10 are included in the release (#1419, #1507)

Changes

The default number of adaptive samples will be algorithm-specific now. For most of the algorithms, the default number of adaptive samples is still 0. For HMC and NUTS, the default is changed to half of number of samples (i.e. num_samples // 2 ) (#1353)

In CompositionalInference, the default algorithm for continuous latent variables is changed to NUTS (GlobalNoUTurnSampler) (#1407).

Fixes

Resolved deprecation warnings to support PyTorch 1.11 (#1378) (Note: PyTorch 1.12 is also supported now)

Documentations

Added a Bayesian structural time series tutorial (#1376) (link to tutorial)

Used the experimental NNC compile feature in supported tutorials (#1408)

Added MiniBM, a minimal and standalone implementation of Bean Machine in around a hundred lines of code (excluding comments) (#1415) (minibm.py)

Source code(tar.gz)
Source code(zip)
v0.1.1(Jan 28, 2022)
Full Changelog: https://github.com/facebookresearch/beanmachine/compare/v0.1.0...v0.1.1

Highlights

Bean Machine now supports Python 3.9 (#1302)

Adds the missing cpp header files to the source distribution in PyPI (#1309)

In case of invalid initialization, BM will attempt to re-initialize an inference and throw a ValueError if the model is misspecified (#1313)

Source code(tar.gz)
Source code(zip)
v0.1.0.post1(Dec 13, 2021)
Full Changelog: https://github.com/facebookresearch/beanmachine/compare/v0.1.0...v0.1.0.post1

Highlights

Fixed the order of samples returning from MonteCarloSamples.get_variables (#1253)

Source code(tar.gz)
Source code(zip)
v0.1.0(Dec 11, 2021)

Source code(tar.gz)
Source code(zip)