← All notes

Short answer

dataset.features is a dict of column name to column type. It tells you, for example, that a label column is a ClassLabel with named classes stored as integers. Change types with cast() or cast_column(), which return a new dataset. [S1][S2]

Example

from datasets import ClassLabel, Value
ds = ds.cast_column("label", ClassLabel(names=["neg", "pos"]))
print(ds.features)

Use List for list-valued columns and nested dicts for structured fields. [S1]

Claims

• id: C1 claim: Features defines a dataset's internal structure. source: S1 evidence: "Features defines the internal structure of a dataset."
• id: C2 claim: The Features format is a dict of column name to type. source: S1 evidence: "It is a dictionary of column name and column type pairs."
• id: C3 claim: ClassLabel labels are stored as integers. source: S1 evidence: "Labels are stored as integers in the dataset."
• id: C4 claim: cast transforms the feature type of columns. source: S2 evidence: "function transforms the feature type of one or more columns"
• id: C5 claim: Processing methods return a new Dataset, not in-place. source: S2 evidence: "Modification is not done in-place."

Sources

• [S1] Dataset features · Hugging Face. Hugging Face (Datasets docs). https://huggingface.co/docs/datasets/about_dataset_features . License: Apache-2.0 (huggingface/datasets docs). Accessed 2026-11-04.
• [S2] Process · Hugging Face. Hugging Face (Datasets docs). https://huggingface.co/docs/datasets/process . License: Apache-2.0 (huggingface/datasets docs). Accessed 2026-11-04.

License: this note is CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Summary by iinventedlight (AI-assisted).

@misc{iil_note_wfb7hcd8,
  author = {{iinventedlight}},
  title = {{Hugging Face Datasets features: column types, ClassLabel and cast}},
  howpublished = {iinventedlight Library},
  year = {2026},
  date = {2026-11-04},
  url = {https://iinventedlight.com/library/notes/hf-datasets-features-casting/},
  note = {Permanent ID iil:note:wfb7hcd8. License: CC-BY-4.0.},
}

Markdown

Spotted a mistake? Use the MCP feedback tool or email hello@iinventedlight.com.

For language models and agents: start at /llms.txt. This page in Markdown: /library/notes/hf-datasets-features-casting.md.