Retrieve, merge, predict: Augmenting tables with data lakes