|Version Badge| |Build Badge| |Docs Badge| |Coverage Badge|
.. |Version Badge| image:: https://img.shields.io/pypi/v/probflow :target: https://pypi.org/project/probflow/
.. |Build Badge| image:: https://github.com/brendanhasz/probflow/workflows/tests/badge.svg :target: https://github.com/brendanhasz/probflow/actions?query=branch%3Amaster
.. |Docs Badge| image:: https://readthedocs.org/projects/probflow/badge/ :target: http://probflow.readthedocs.io
.. |Coverage Badge| image:: https://codecov.io/gh/brendanhasz/probflow/branch/master/graph/badge.svg :target: https://codecov.io/gh/brendanhasz/probflow
ProbFlow is a Python package for building probabilistic Bayesian models with TensorFlow 2.0 <http://www.tensorflow.org/beta>
_ or PyTorch <http://pytorch.org>
_, performing stochastic variational inference with those models, and evaluating the models' inferences. It provides both high-level modules for building Bayesian neural networks, as well as low-level parameters and distributions for constructing custom Bayesian models.
It's very much still a work in progress.
ProbFlow allows you to quickly and less painfully build, fit, and evaluate custom Bayesian models (or ready-made <http://probflow.readthedocs.io/en/latest/api/applications.html>
_ ones!) which run on top of either TensorFlow 2.0 <http://www.tensorflow.org/beta>
_ and TensorFlow Probability <http://www.tensorflow.org/probability>
_ or PyTorch <http://pytorch.org>
_.
With ProbFlow, the core building blocks of a Bayesian model are parameters and probability distributions (and, of course, the input data). Parameters define how the independent variables (the features) predict the probability distribution of the dependent variables (the target).
For example, a simple Bayesian linear regression
.. image:: https://raw.githubusercontent.com/brendanhasz/probflow/master/docs/img/regression_equation.svg?sanitize=true :width: 30 % :align: center
can be built by creating a ProbFlow Model. This is just a class which inherits pf.Model
(or pf.ContinuousModel
or pf.CategoricalModel
depending on the target type). The __init__
method sets up the parameters, and the __call__
method performs a forward pass of the model, returning the predicted probability distribution of the target:
.. code-block:: python
import probflow as pf
import tensorflow as tf
class LinearRegression(pf.ContinuousModel):
def __init__(self):
self.weight = pf.Parameter(name='weight')
self.bias = pf.Parameter(name='bias')
self.std = pf.ScaleParameter(name='sigma')
def __call__(self, x):
return pf.Normal(x*self.weight()+self.bias(), self.std())
model = LinearRegression()
Then, the model can be fit using stochastic variational inference, in one line:
.. code-block:: python
# x and y are Numpy arrays or pandas DataFrame/Series
model.fit(x, y)
You can generate predictions for new data:
.. code-block:: pycon
# x_test is a Numpy array or pandas DataFrame
>>> model.predict(x_test)
[0.983]
Compute probabilistic predictions for new data, with 95% confidence intervals:
.. code-block:: python
model.pred_dist_plot(x_test, ci=0.95)
.. image:: https://raw.githubusercontent.com/brendanhasz/probflow/master/docs/img/pred_dist_light.svg?sanitize=true :width: 90 % :align: center
Evaluate your model's performance using metrics:
.. code-block:: pycon
>>> model.metric('mse', x_test, y_test)
0.217
Inspect the posterior distributions of your fit model's parameters, with 95% confidence intervals:
.. code-block:: python
model.posterior_plot(ci=0.95)
.. image:: https://raw.githubusercontent.com/brendanhasz/probflow/master/docs/img/posteriors_light.svg?sanitize=true :width: 90 % :align: center
Investigate how well your model is capturing uncertainty by examining how accurate its predictive intervals are:
.. code-block:: pycon
>>> model.pred_dist_coverage(ci=0.95)
0.903
and diagnose where your model is having problems capturing uncertainty:
.. code-block:: python
model.coverage_by(ci=0.95)
.. image:: https://raw.githubusercontent.com/brendanhasz/probflow/master/docs/img/coverage_light.svg?sanitize=true :width: 90 % :align: center
ProbFlow also provides more complex modules, such as those required for building Bayesian neural networks. Also, you can mix ProbFlow with TensorFlow (or PyTorch!) code. For example, even a somewhat complex multi-layer Bayesian neural network like this:
.. image:: https://raw.githubusercontent.com/brendanhasz/probflow/master/docs/img/dual_headed_net_light.svg?sanitize=true :width: 99 % :align: center
Can be built and fit with ProbFlow in only a few lines:
.. code-block:: python
class DensityNetwork(pf.ContinuousModel):
def __init__(self, units, head_units):
self.core = pf.DenseNetwork(units)
self.mean = pf.DenseNetwork(head_units)
self.std = pf.DenseNetwork(head_units)
def __call__(self, x):
z = tf.nn.relu(self.core(x))
return pf.Normal(self.mean(z), tf.exp(self.std(z)))
# Create the model
model = DensityNetwork([x.shape[1], 256, 128], [128, 64, 32, 1])
# Fit it!
model.fit(x, y)
For convenience, ProbFlow also includes several pre-built models <http://probflow.readthedocs.io/en/latest/api/applications.html>
_ for standard tasks (such as linear regressions, logistic regressions, and multi-layer dense neural networks). For example, the above linear regression example could have been done with much less work by using ProbFlow's ready-made LinearRegression model:
.. code-block:: python
model = pf.LinearRegression(x.shape[1])
model.fit(x, y)
And a multi-layer Bayesian neural net can be made easily using ProbFlow's ready-made DenseRegression model:
.. code-block:: python
model = pf.DenseRegression([x.shape[1], 128, 64, 1])
model.fit(x, y)
Using parameters and distributions as simple building blocks, ProbFlow allows
for the painless creation of more complicated Bayesian models like generalized linear models <http://probflow.readthedocs.io/en/latest/examples/glm.html>
,
deep time-to-event models <http://probflow.readthedocs.io/en/latest/examples/time_to_event.html>
,
neural matrix factorization <http://probflow.readthedocs.io/en/latest/examples/nmf.html>
_ models, and
Gaussian mixture models <http://probflow.readthedocs.io/en/latest/examples/gmm.html>
. You can even
mix probabilistic and non-probabilistic models <http://probflow.readthedocs.io/en/latest/examples/neural_linear.html>
! Take
a look at the examples <http://probflow.readthedocs.io/en/latest/examples/examples.html>
_
and the user guide <http://probflow.readthedocs.io/en/latest/user_guide/user_guide.html>
_
for more!
If you already have your desired backend installed (i.e. Tensorflow/TFP or PyTorch), then you can just do:
.. code-block:: bash
pip install probflow
Or, to install both ProbFlow and the CPU version of TensorFlow + TensorFlow Probability,
.. code-block:: bash
pip install probflow[tensorflow]
Or, to install ProbFlow and the GPU version of TensorFlow + TensorFlow Probability,
.. code-block:: bash
pip install probflow[tensorflow_gpu]
Or, to install ProbFlow and PyTorch,
.. code-block:: bash
pip install probflow[pytorch]
Post bug reports, feature requests, and tutorial requests in GitHub issues <http://github.com/brendanhasz/probflow/issues>
_.
Pull requests <http://github.com/brendanhasz/probflow/pulls>
_ are totally welcome! Any contribution would be appreciated, from things as minor as pointing out typos to things as major as writing new applications and distributions.
Because it's a package for probabilistic modeling, and it was built on TensorFlow. ¯\(ツ)/¯