1. Predicates

Predicates are used within marc21-learn to filter records and fields according to specific criteria. Filtering out records that are not needed early on prevents unnecessary computations; in particular, it avoids unnecessary memory allocations. In the context of the Polars integration, the filter expressions are used for predicate pushdown to filter out records as early as possible.

When filtering records, a distinction is made between the two basic expressions: leader matcher and field matcher. These elementary expressions can be combined into more complex expressions using the Boolean connectives conjunction (&&) and disjunction (||), as well as grouping ((...)). While the leader matcher makes statements about the leader of a record, the field matcher is used to make statements about the variable fields.

The following sections demonstrate the various ways to filter records using the read_marc21() function and the Ada Lovelace test record. After the predicate has been applied to the record, the height of the DataFrame is checked. If the height is 1, the test record meets the criterion; otherwise, it does not.