Autoencoders using H2O

An autoencoder is an ANN used for learning without efficient coding control. The purpose of an autoencoder is to learn coding for a set of data, typically to reduce dimensionality. Architecturally, the simplest form of autoencoder is an advanced and non-recurring neural network very similar to the MLP, with an input level, an output layer, and one or more hidden layers that connect them, but with the layer outputs having the same number of input level nodes for rebuilding their inputs.

In the following is proposed an example of autoencoder using h2o on a movie dataset.

Note

The dataset used in this example is a set of movies and genre taken from https://grouplens.org/datasets/movielens.

We use the movies.csv file, which has three columns:

• movieId
• title
• genres

There are 164,979 rows of data for clustering. We will use h2o.deeplearning to have the autoencoder parameter fix the clusters. The objective of the exercise is to cluster the movies based on genre, which can then be...