Library of composable generative population models which serve as the modeling and inference backend of BayesDB.
The aim of this project is to provide a unified probabilistic programming
framework to express different models and techniques from statistics, machine
learning and non-parametric Bayes. It serves as the primary modeling and
inference runtime system for bayeslite,
an open-source implementation of BayesDB.
Composable generative population models (CGPM) are a computational abstraction
for probabilistic objects. They provide an interface that explicitly
differentiates between the sampler of a random variable from its conditional
distribution and the assessor of its conditional density. By encapsulating
models as probabilistic programs that implement CGPMs, complex models can be
built as compositions of sub-CGPMs, and queried in a model-independent way
using the Bayesian Query Language.
The easiest way to install cgpm is to use the
package on Anaconda Cloud.
Please follow these instructions.
cgpm targets Ubuntu 14.04 and 16.04. The package can be installed by cloning
this repository and following these instructions. It is highly recommended to
install cgpm inside of a virtualenv which was created using the--system-site-packages flag.
Install dependencies from apt, listed here.
Retrieve and build the source.
% git clone git@github.com:probcomp/cgpm% cd cgpm% pip install --no-deps .
Verify the installation.
% python -c 'import cgpm'% cd cgpm && ./check.sh
CGPMs, and their integration as a runtime system for
BayesDB, are described in the following
technical report:
Applications of using cgpm and bayeslite for data analysis tasks can be further
found in:
Probabilistic Search for Structured Data via Probabilistic Programming and Nonparametric Bayes.
Saad, F. Casarsa, L., and Mansinghka, V. arXiv preprint, arXiv:1704.01087, 2017.
Detecting Dependencies in Sparse, Multivariate Databases Using Probabilistic Programming and Non-parametric Bayes.
Saad, F., and Mansinghka, V. Artificial Intelligence and Statistics (AISTATS), 2017.
A Probabilistic Programming Approach to Probabilistic Data Analysis.
Saad, F., and Mansinghka, V. Advances in Neural Information Processing Systems (NIPS), 2016.
Running ./check.sh will run a subset of the tests that are considered complete
and stable. To launch the full test suite, including continuous integration
tests, run py.test in the root directory. There are more tests in the tests/
directory, but those that do not start with test_ or do start with disabled_
are not considered ready. The tip of every branch merged into master must
pass ./check.sh, and be consistent with the code conventions outlined in
HACKING.
To run the full test suite, use ./check.sh --integration tests/. Note that the
full integration test suite requires installing the C++
crosscat backend.
Copyright (c) 2015-2016 MIT Probabilistic Computing Project
Licensed under the Apache License, Version 2.0 (the “License”);
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an “AS IS” BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.