Link type: Linking, Deduping or Both

Splink allows data to be linked, deduplicated or both.

Linking refers to finding links between datasets, whereas deduplication finding links within datasets.

Data linking is therefore only meaningful when more than one dataset is provided.

This guide shows how to specify settings and initialise the linker for the three link types. The examples assume you already have input data in df, or in df_1, df_2 and df_n. Register each input with the same database API before passing it to Linker.

Deduplication

The dedupe_only link type expects the user to provide a single input table, and is specified as follows

from splink import DuckDBAPI, Linker, SettingsCreator

settings = SettingsCreator(
    link_type= "dedupe_only",
)

db_api = DuckDBAPI()
df_splink = db_api.register(df, dataset_display_name="people")
linker = Linker(df_splink, settings)

Link only

The link_only link type expects the user to provide a list of input tables, and is specified as follows:

from splink import DuckDBAPI, Linker, SettingsCreator

settings = SettingsCreator(
    link_type= "link_only",
)

db_api = DuckDBAPI()
input_tables = [
    db_api.register(df_1, dataset_display_name="dataset_1"),
    db_api.register(df_2, dataset_display_name="dataset_2"),
    db_api.register(df_n, dataset_display_name="dataset_n"),
]
linker = Linker(input_tables, settings)

Dataset labels used in the outputs (the source_dataset column) are set via the dataset_display_name argument when registering each table with the database API (e.g. db_api.register(df_1, dataset_display_name="name1")). If not provided at registration, defaults will be automatically chosen by Splink.

Link and dedupe

The link_and_dedupe link type expects the user to provide a list of input tables, and is specified as follows:

from splink import DuckDBAPI, Linker, SettingsCreator

settings = SettingsCreator(
    link_type= "link_and_dedupe",
)

db_api = DuckDBAPI()
input_tables = [
    db_api.register(df_1, dataset_display_name="dataset_1"),
    db_api.register(df_2, dataset_display_name="dataset_2"),
    db_api.register(df_n, dataset_display_name="dataset_n"),
]
linker = Linker(input_tables, settings)

Dataset labels used in the outputs (the source_dataset column) are set via the dataset_display_name argument when registering each table with the database API (e.g. db_api.register(df_1, dataset_display_name="name1")). If not provided at registration, defaults will be automatically chosen by Splink.