Basic concepts

Polars is a data analytics library written in Rust. Polars relies on the following data structures: Series, and ChunkedArray<T>, DataFrame, and lazyframe.

A DataFrame is a series of two-dimensional data structures that can be thought of as tables of data made up of rows and columns (columns, called "fields" in data science). What you can do on a DataFrame is very similar to a query in SQL. YOU CAN GROUP, JOIN, PIVOT¹, etc. Dataframes can be thought of as an abstraction of Vec<Series> , with each column corresponding to a Series. Series only holds Arc<dyn SeriesTrait>. The ChunkedArray<T> type implements SeriesTrait. Series is a column data representation of the obscure type of Polars. Some operations that are not related to data types are provided by Series and SeriesTrait, such as indexing and renaming operations. Operations related to data types must be downgraded to ChunkedArray<T>, the underlying data structure of the Series. ChunkedArray<T>, which is a chunked array, is similar to Vec<dyn ArrowArray> at the bottom layer, and chunking is conducive to parallel operation of data. This is the underlying data structure of Polars and implements a number of operations. Most operations are defined in chunked_array::ops or implemented on a ChunkedArray structure.