How to Split and Sample a Dataset in BigQuery Using SQL

Splitting data means that we will divide it into subsets. For data science models, datasets are usually partitioned into two or three subsets: training, validation, and test. Each subset of data has a purpose, from creating a model to ensuring its performance. To decide on the size of each subset, we often see standard rules and ratios. There have been some discussions about what an optimal split might be, but in general, I would recommend keeping in mind that not having enough data, either on the training or validation set, will result in a model that is difficult to learn/train, or you will have difficulty determining whether this model actually performs well or not. It's worth noting that you don't always have to make three segments.