Guiding Principles

Features

The main features of InferPy are listed below.

• Allows a simple definition of inference over probabilistic models containing deep neural networks.

• The models that can be defined in InferPy are those that can be defined using Edward2 (i.e., tfp.edward2), whose probability distributions are mainly inherited from the module distributions in the tensorflow-probability package.

• Edward’s drawback is that for the model definition, the user has to manage complex multi-dimensional arrays called tensors. By contrast, in InferPy all the parameters in a model can be defined using the standard Python types (compatibility with Numpy is available as well).

• InferPy directly relies on top of Edward’s inference engine and thus includes all the inference algorithms available in that package. As Edward’s inference engine relies on TensorFlow computing engine, InferPy also relies on it too.

• Unlike Edward, our package does not require a strong background in the inference methods.

Architecture

Given the previous considerations, we can summarize the InferPy architecture as follows.

Note that InferPy can be seen as an upper layer for working with probabilistic distributions defined over tensors. Most of the interaction is done with Edward: the definitions of the random variables and the inference. However, InferPy also interacts directly with TensorFlow in some operations that are hidden to the user, e.g., the manipulation of the tensors representing the parameters of the distributions.

An additional advantage of using Edward and TensorFlow as inference engine is that all the parallelization details are hidden to the user. Moreover, the same code will run either in CPUs or GPUs.