For probabilistic modeling, experimental libraries abound. Running probabilistic networks can be much slower than algorithmic (non-algorithmic) approaches, which until not long ago rendered them impractical for anything but very small datasets. In fact, most of the tutorials and examples relate to toy datasets.
However, this has changed in recent years due to faster hardware and variational inference. With TensorFlow Probability, it is often straightforward to define architectures, losses, and layers, even with probabilistic sampling with full GPU support, and state-of-the-art implementations that support fast training.
In this recipe, we'll implement an application in healthcare – we'll diagnose a disease.
Getting ready
We already have scikit-learn and TensorFlow installed from previous chapters.
For this recipe, we'll need tensorflow-probability as well:
pip install tensorflow-probability
Now that tensorflow-probability is installed, we&apos...
