NEWS

---

croissant 0.0.1

• read_croissant() reads Croissant 1.0 and 1.1 JSON-LD metadata from a file, URL, or JSON string into a Croissant object.
• write_croissant() writes a Croissant object as JSON-LD with the standard context of its version.
• S7 classes Croissant, FileObject, FileSet, RecordSet, Field, ParentField, Source, Extract, Transform, Person, Organization, and CreativeWork represent Croissant nodes, with the properties that 'mlcroissant' types; every other property is kept in other_properties.
• croissant_context() returns the standard JSON-LD context of a Croissant version.
• validate_croissant() reports every error and warning in Croissant metadata as a CroissantIssues object; is_valid() tests it for errors.
• read_croissant() reports every read problem of a document in one error.
• read_record_set() reads the records of a record set as a data.table, tibble, data.frame, or arrow Table, from CSV, TSV, JSON, JSON Lines, and Parquet files, local or downloaded, including files in zip and tar archives.
• croissant_from_data() writes a table to a CSV or Parquet file and returns Croissant metadata that describes it, with field data types from column classes and the file's SHA-256 checksum.
• read_croissant_hf() reads the Croissant metadata of a Hugging Face dataset, and read_croissant() and read_record_set() accept hf://datasets/ paths; read_record_set() downloads the dataset's Parquet files from the resolved commit and verifies them.
• hf_download() downloads and verifies one file of a Hugging Face dataset.
• read_record_set() gains n_max, which stops reading files once that many records are held, and token, for gated and private Hugging Face datasets.