Raw data and features – what are the differences?

ML systems are data-hungry. They rely on the data to be trained and to make inferences. However, not all data is equally important. Before the era of deep learning (DL), the data was supposed to be processed in order to be used in ML. Before DL, the algorithms were limited in the amount of data that could be used for training. The storage and memory limitations were also limited, and therefore, ML engineers had to prepare the data much more than for DL. For example, ML engineers needed to spend more effort to find a small but still representative sample of data for training. After the introduction of DL, ML models can find complex patterns in much larger datasets. Therefore, the work of ML engineers is now focused on finding sufficiently large, and representative, datasets.

Classical ML systems – that is, non-DL systems – require data in a tabular form in order to make inferences, and therefore it is important...