Analysts need an unbiased evaluation of the quality of their machine learning models. To get this, they partition the available data into two parts. They use one part to build the machine learning model and retain the remaining data as hold out data. After building the model, they evaluate the model's performance on the hold out data. This recipe shows you how to partition data. It separately addresses the situations when the target variable is numeric and when it is categorical. It also covers the process of creating two partitions or three.