The Monk directory contains train and test splits of the MONK's problems datasets. The datasets have already been one-hot encoded. For information about the dataset itself, please refer to the UCI Machine Learning Repository.
The ML_Cup directory, instead, contains a dataset of sensor readings. The task is to predict the last 2 columns using the first 10 as features. The dataset is divided into a training set and a blind test set, meaning that target values aren't available for the latter. Nonetheless, you can evaluate the network's performances by taking a subset of the training set and use it as a validation or test set.