Skip to main content

Quick Start

Short task-oriented paths to first success. Not sure where to start? Pick your role and goal in the cookbook. Installation details: Installation. CLI flag list: CLI reference.

CSV → Parquet in 30 seconds​

pip install undatum   # or: uv tool install undatum
printf 'name,age,city\nAda,36,London\nGrace,85,NYC\n' > people.csv
undatum convert people.csv people.parquet
undatum stats people.parquet

For multi-GB inputs:

undatum convert huge.jsonl.zst huge.parquet --low-memory

Validate a dataset before publishing​

Describe your expectations in a rules file, then validate against it:

cat > rules.yml <<'EOF'
rules:
  - field: email
    name: Email format
    required: true
    type: string
    format: email
    severity: error
EOF

undatum validate data.csv --rules rules.yml
undatum analyze data.csv
undatum package create data.csv --output datapackage.json

For a one-off check without a rules file, use legacy single-rule mode:

undatum validate data.csv --fields email --rule common.email

Query JSONL with SQL​

undatum sql takes the query first, then the input file(s). A single input is available as the view data; multiple inputs are named after their file stems.

undatum sql "SELECT city, COUNT(*) AS n FROM data GROUP BY 1" events.jsonl
# or use frequency / select for simpler extractions:
undatum frequency events.jsonl --fields city
undatum select --fields id,city,ts --filter '`city` == "Berlin"' events.jsonl

Dump a database table to Parquet​

undatum db dump --db sqlite:///app.db --table users --output users.parquet --to parquet

Next steps​