catalog-config

Catalog Configuration

Dataset type naming

Use the current module path:

kedro_datasets.<library>.<Type>Dataset

Short form also works — Kedro resolves pandas.CSVDataset automatically.

Three things agents get wrong:

• Wrong module: kedro.extras.datasets.* is deprecated and removed — never generate it.

• Wrong casing: since kedro-datasets 2.0, use lowercase Dataset (e.g. CSVDataset, not CSVDataSet).

• Top-level layer:: deprecated since Kedro 0.19 and silently ignored. Use metadata.kedro-viz.layer instead:

  companies:
    type: pandas.CSVDataset
    filepath: data/01_raw/companies.csv
    metadata:
      kedro-viz:
        layer: raw
  

Check the docs before writing an entry

Do not guess dataset types or constructor arguments from training data — they change across versions and your knowledge may be outdated. You MUST verify the dataset type exists and look up its docs before writing the catalog entry.

CRITICAL: Never suggest a dataset type without verifying it exists in the installed version. If you cannot confirm it exists via the steps below, say so explicitly rather than guessing.

Step 1 — Get the installed version:

Your shell does not necessarily use the environment the user has activated in their own terminal — it often starts in the base or system interpreter. Never report a version without first confirming which interpreter produced it.

1a. Report the version and the interpreter path together, so you can judge whether to trust it:

python -c "import sys, kedro_datasets as k; print(k.__version__, sys.executable)"

Trust the result only if the path sits inside an environment — it contains /envs/, /.venv/, or /virtualenvs/. A path like /usr/bin/python, .../miniconda3/bin/python, or .../anaconda3/bin/python is the base interpreter: treat it as unusable even if the import succeeded, because the version it reports is probably not the project's.

1b. If the import failed or the path is a base interpreter, list the candidate environments: