goldenmatch-duckdb

GoldenMatch entity resolution functions for DuckDB.

pip install goldenmatch-duckdb

Usage

import duckdb
import goldenmatch_duckdb

con = duckdb.connect()
goldenmatch_duckdb.register(con)

# Score two strings
con.sql("SELECT goldenmatch_score('John Smith', 'Jon Smyth', 'jaro_winkler')").show()

# Deduplicate a table
con.sql("""
    CREATE TABLE customers AS SELECT * FROM (VALUES
        ('John', 'john@x.com'),
        ('JOHN', 'john@x.com'),
        ('Jane', 'jane@y.com')
    ) AS t(name, email)
""")
con.sql("SELECT goldenmatch_dedupe_table('customers', '{\"exact\": [\"email\"]}')").show()

# Match two tables
con.sql("SELECT goldenmatch_match_tables('prospects', 'reference', '{\"fuzzy\": {\"name\": 0.85}}')").show()

Functions

Core-API functions

Thin wrappers over goldenmatch's public core APIs. All return JSON strings (scalar functions noted otherwise); table-input functions read the named DuckDB table directly.

# Otsu threshold suggestion
con.sql("SELECT goldenmatch_suggest_threshold('[0.1,0.12,0.9,0.92]')").show()

# Detect domain from columns
con.sql("SELECT goldenmatch_detect_domain('[\"product_title\",\"brand\",\"sku\"]')").show()

# Profile / validate / auto-fix a table
con.sql("SELECT goldenmatch_profile_table('customers')").show()

# Fellegi-Sunter: train, then score
con.sql("""
    SELECT goldenmatch_score_probabilistic(
        :rows, :mk,
        goldenmatch_train_em(:rows, :mk, '{}')
    )
""")

GoldenFlow transforms

8 scalar VARCHAR -> VARCHAR UDFs wrapping GoldenFlow's transform registry -- normalize / canonicalize a column before matching. Byte-equivalent to the Postgres goldenflow_* functions, so a transform written for one backend ports to the other. They fail open (pass the input through unchanged) when goldenflow isn't installed -- pip install goldenflow to enable.

# Normalize columns inline before matching
con.sql("""
    SELECT
        goldenflow_normalize_email(email)  AS email_norm,
        goldenflow_normalize_phone(phone)  AS phone_e164
    FROM customers
""").show()

Requirements

• Python 3.11+
• DuckDB 1.0+
• goldenmatch >= 1.1.0