---

name: data-preparation description: "Load, inspect, clean, reshape, validate, and save Orange data." disable-model-invocation: true metadata: disco-role: operating license: NOASSERTION

Data Preparation

Use this sub-skill for Orange data loading, schema inspection, column/domain edits, preprocessing, reshaping, validation, and saving.

Start here

1. Identify the input shape: local file, URL, sample dataset, in-memory Table, or SQL source.
2. Decide whether the task is only inspection or also cleaning, reshaping, validation, or export.
3. Read references/data-formats.md for the Table/Domain/Variable model and file-format rules.
4. Read references/file-and-sql-workflows.md for file, widget, reshape, save, and optional SQL flows.
5. Read references/troubleshooting.md when parsing, encoding, duplicate-name, empty-table, or save/load problems appear.
6. Optionally run scripts/data_smoke.py for a non-mutating sanity check.

Use this for

• Loading tables from .tab, .csv, .tsv, .xlsx, .pkl, sample datasets, or URLs.
• Understanding domains, attributes, class variables, metas, and variable reuse.
• Cleaning and reshaping data with OWFile, OWCSVImport, OWDataSets, OWDataInfo, OWFeatureStatistics, OWSelectColumns, OWEditDomain, OWSelectRows, OWImpute, OWContinuize, OWDiscretize, OWPurgeDomain, OWConcatenate, OWMelt, OWTranspose, OWGroupBy, OWRandomize, OWUnique, OWTransform, OWDataSampler, OWCreateClass, OWCreateInstance, and OWPreprocess.
• Exporting with OWSave or Table.save.
• Using OWSql and Orange.data.sql only when a backend is already available and configured.

Route away

• Supervised learners, predictors, scoring, and evaluation: use the supervised-modeling sibling.
• Plots, projections, clustering, and other exploratory visualization: use the exploration-visualization sibling.
• Widget framework mechanics, canvas behavior, or widget-catalog tooling: use the widget-development sibling.

Operating notes

• Table stores rows in X, Y, metas, and W; Domain controls which variables live in each part.
• Variable.make(...) is preferred when you want Orange to reuse an existing descriptor by name and type.
• Prefer Table.from_file(...) / Table.save(...) for file round trips, and Table.from_numpy(...) / Table.from_table(...) for programmatic construction or domain conversion.
• Table.transpose(...) is the main reshape helper for turning columns into rows.
• Direct in-place edits in scripts usually need with table.unlocked():.
• OWFile can rename duplicate columns and switch readers; OWCSVImport is the escape hatch for delimiter and encoding problems.
• OWSave refuses unsupported sparse formats; pickle is the safe fallback for sparse tables.
• OWSql is optional and service-bound; do not make it a minimum requirement.

Boundaries

• Do not cover supervised modeling, model evaluation, plotting, projection, or clustering here.
• Do not assume a live SQL backend in the inspection environment.
• Keep SQL guidance documented but optional.