Hands-On Unsupervised Learning with Python

Example of an unsupervised DBN

In this example, we want to employ a DBN in order to find a low-dimensional representation of the MNIST dataset. As the complexity of these models can easily grow, we are going to limit the process to 500 random samples. The implementation is based on the deep-belief-network package (https://github.com/albertbup/deep-belief-network), which supports both NumPy and TensorFlow. In the former case, the classes (whose names remain unchanged) must be imported from the dbn package, while in the latter, the package is dbn.tensorflow. In this example, we are going to use the NumPy version, which has fewer requirements, but the reader is invited to check the TensorFlow version, too.

Let's start by loading and normalizing ...