- PYTHON_VERSION=3.5 FLOATX='float64' RUN_PYLINT="true" TESTCMD="--durations=50 --cov-append pmlearn/tests/test_base.py pmlearn/linear_model/tests/test_base.py pmlearn/linear_model/tests/test_logistic.py --ignore=pmlearn/gaussian_process/tests/test_gpr.py --ignore=pmlearn/mixture/tests/test_gaussian_mixture.py --ignore=pmlearn/mixture/tests/test_dirichlet_process.py --ignore=pmlearn/naive_bayes/tests/test_naive_bayes.py --ignore=pmlearn/neural_network/test_multilayer_perceptron.py"

- PYTHON_VERSION=3.5 FLOATX='float64' RUN_PYLINT="true" TESTCMD="--durations=50 --cov-append pmlearn/tests/test_base.py --ignore=pmlearn/linear_model/tests/test_base.py --ignore=pmlearn/linear_model/tests/test_logistic.py pmlearn/gaussian_process/tests/test_gpr.py --ignore=pmlearn/mixture/tests/test_gaussian_mixture.py --ignore=pmlearn/mixture/tests/test_dirichlet_process.py --ignore=pmlearn/naive_bayes/tests/test_naive_bayes.py --ignore=pmlearn/neural_network/test_multilayer_perceptron.py"

- PYTHON_VERSION=3.5 FLOATX='float64' RUN_PYLINT="true" TESTCMD="--durations=50 --cov-append pmlearn/tests/test_base.py pmlearn/linear_model/tests/test_base.py pmlearn/linear_model/tests/test_logistic.py --ignore=pmlearn/gaussian_process/tests/test_gpr.py --ignore=pmlearn/mixture/tests/test_gaussian_mixture.py --ignore=pmlearn/mixture/tests/test_dirichlet_process.py --ignore=pmlearn/naive_bayes/tests/test_naive_bayes.py --ignore=pmlearn/neural_network/test_multilayer_perceptron.py"

- PYTHON_VERSION=3.5 FLOATX='float64' RUN_PYLINT="true" TESTCMD="--durations=50 --cov-append pmlearn/tests/test_base.py --ignore=pmlearn/linear_model/tests/test_base.py --ignore=pmlearn/linear_model/tests/test_logistic.py pmlearn/gaussian_process/tests/test_gpr.py --ignore=pmlearn/mixture/tests/test_gaussian_mixture.py --ignore=pmlearn/mixture/tests/test_dirichlet_process.py --ignore=pmlearn/naive_bayes/tests/test_naive_bayes.py --ignore=pmlearn/neural_network/test_multilayer_perceptron.py"

- Lines should be a maximum of 80 characters long

It provides a variety of state-of-the art probabilistic models for supervised

and unsupervised machine learning. **It is inspired by**

`scikit-learn <http://scikit-learn.org>`_ **and focuses on bringing probabilistic

machine learning to non-specialists**. It uses a syntax that mimics scikit-learn.

Emphasis is put on ease of use, productivity, flexibility, performance,

documentation, and an API consistent with scikit-learn. It depends on scikit-learn

and `PyMC3 <https://docs.pymc.io/>`_ and is distributed under the new BSD-3 license,

encouraging its use in both academia and industry.

You can install ``pymc-learn`` from PyPi using pip as follows:

.. code-block:: bash

Or from source as follows:

.. CAUTION::

``pymc-learn`` is under heavy development.

PyMC3, Scikit-learn, NumPy, SciPy, and Matplotlib (see ``requirements.txt``

for version information).

@article{emaasit2018pymc,

title={Pymc-learn: Practical probabilistic machine learning in {P}ython},

author={Emaasit, Daniel and others},

journal={arXiv preprint arXiv:1810.xxxxx},

year={2018}

and how to apply them.

``pymc-learn`` leverages and extends the Base template provided by the PyMC3

Models project: https://github.com/parsing-science/pymc3_models.

- *Familiarity*: ``pymc-learn`` mimics the syntax of `scikit-learn <https://scikit-learn.org>`_ -- a popular Python library for machine learning -- which has a consistent & simple API, and is very user friendly.

- *Ease of use*: This makes ``pymc-learn`` easy to learn and use for first-time users.

- *Productivity*: For scikit-learn users, you don't have to completely rewrite your code. Your code looks almost the same. You are more productive, allowing you to try more ideas faster.

- *Flexibility*: This ease of use does not come at the cost of reduced flexibility. Given that ``pymc-learn`` integrates with `PyMC3 <https://docs.pymc.io>`_, it enables you to implement anything you could have built in the base language.

- *Performance*. The primary inference algorithm is gradient-based automatic differentiation variational inference (ADVI) (Kucukelbir et al., 2017), which estimates a divergence measure between approximate and true posterior distributions. Pymc-learn scales to complex, high-dimensional models thanks to GPU-accelerated tensor math and reverse-mode automatic differentiation via Theano (Theano Development Team, 2016), and it scales to large datasets thanks to estimates computed over mini-batches of data in ADVI.

Why do we need pymc-learn?

Currently, there is a growing need for principled machine learning approaches by

non-specialists in many fields including the pure sciences (e.g. biology, physics,

chemistry), the applied sciences (e.g. political science, biostatistics),

engineering (e.g. transportation, mechanical), medicine (e.g. medical imaging),

the arts (e.g visual art), and software industries.

This has lead to increased adoption of probabilistic modeling. This trend is

attributed in part to three major factors:

(1) the need for transparent models with calibrated quantities of uncertainty, i.e. "models should know when they don't know",

(2) the ever-increasing number of promising results achieved on a variety of fundamental problems in AI (Ghahramani, 2015), and

(3) the emergency of probabilistic programming languages (PPLs) that provide a fexible framework to build richly structured probabilistic models that incorporate domain knowledge.

However, usage of PPLs requires a specialized understanding of probability

theory, probabilistic graphical modeling, and probabilistic inference. Some PPLs

also require a good command of software coding. These requirements make it

difficult for non-specialists to adopt and apply probabilistic machine learning

to their domain problems.

``Pymc-learn`` seeks to address these challenges by providing state-of-the art

implementations of several popular probabilistic machine learning models.

**It is inspired by scikit-learn** (Pedregosa et al., 2011) **and focuses on

bringing probabilistic machine learning to non-specialists**. It puts emphasis

on:

(1) ease of use,

(2) productivity,

(3) fexibility,

(4) performance,

(5) documentation, and

(6) an API consistent with scikit-learn.

The underlying probabilistic models are built using pymc3 (Salvatier et al., 2016).

5. Salvatier, J., Wiecki, T. V., & Fonnesbeck, C. (2016). Probabilistic programming in Python using PyMC3. PeerJ Computer Science, 2, e55.

6. Alp Kucukelbir, Dustin Tran, Rajesh Ranganath, Andrew Gelman, and David M Blei. Automatic differentiation variational inference. The Journal of Machine Learning Research, 18(1):430{474, 2017.

7. Fabian Pedregosa, Gael Varoquaux, Alexandre Gramfort, Vincent Michel, Bertrand Thirion, Olivier Grisel, Mathieu Blondel, Peter Prettenhofer, Ron Weiss, Vincent Dubourg, et al. Scikit-learn: Machine learning in python. Journal of machine learning research, 12(Oct): 2825-2830, 2011.

8. Theano Development Team. Theano: A Python framework for fast computation of mathematical expressions. arXiv e-prints, abs/1605.02688, May 2016. URL http://arxiv.org/abs/1605.02688.