Machine Learning Algorithms - Second Edition

Data collection

The first step is always the most generic because it depends on every single context. However, before working with any dataset, it's necessary to collect the data from all the sources where it is stored. The ideal situation is to have a Comma-Separated Values (CSV) (or another suitable format) dump that can be immediately loaded, but, more often, the engineer has to look for all the database tables, define the right SQL query to collect all the pieces of information, and manage data type conversion and encoding.

We're not going to discuss this topic (which is more related to data engineering), but it's important not to under-evaluate this stage, because it can be much more difficult than expected. I suggest, whenever possible, ...