Metadata Generation

The make_metadata_from_data.py utility generates baseline CSVW-EO metadata from an existing dataset.

---

Purpose

The generator infers:

• Datatypes
• Nullable proportions
• Numeric bounds
• Public keys
• Contribution assumptions
• Optional dependencies
• Optional partitions
• Optional column groups

Important

Automatically generated metadata may leak sensitive information. Manual review is always required before publication.

---

Basic Usage

python make_metadata_from_data.py \
  data.csv \
  --privacy_unit user_id

Contribution Levels

Four contribution levels are supported:

Level	Description
table	Table-level DP metadata only
table_with_keys	Table-level metadata with public keys
column	Per-column DP contribution metadata
partition	Fine-grained partition-level metadata

Higher level incurr more risk of disclosure. Always use the lowest metadata granularity sufficient for the use case.

Example: Table-Level Metadata

python make_metadata_from_data.py \
  data.csv \
  --privacy_unit user_id \
  --default_contributions_level table

Example: Continuous Partitions

python make_metadata_from_data.py \
  data.csv \
  --privacy_unit user_id \
  --continuous_partitions '{"age":[0,18,30,50,100]}'

Example: Column Groups

python make_metadata_from_data.py \
  data.csv \
  --privacy_unit user_id \
  --column_groups '[["age","income"]]'

Example: Dependency Detection

python make_metadata_from_data.py \
  data.csv \
  --privacy_unit user_id \
  --with_dependencies True

Notes

• Datetime columns are inferred automatically
• Numeric bounds are inferred for numeric columns
• Dependency detection may increase runtime
• Fine-grained metadata increases disclosure risk