Exploring H2O

H2O has been around longer than Keras and MXNet and is still used widely. It makes use of Java and MapReduce in-memory compression to handle big datasets. H2O is used for many machine learning tasks and also supports deep learning. In particular, H2O provides native support for feedforward artificial neural networks (multilayer perceptrons). H2O performs automatic data preparation and missing value handling. Loading data requires the use of a special data type: H2OFrame.

Available functions

H2O only natively supports feedforward neural networks. Compared with the other main packages for deep learning, this creates an obvious limitation for this library. However, this is a very common deep learning implementation...