Is there a way to materialize the table with the pairs for evaluation (and/or get the underlying query)?

[dcaseykc]

Is there a way to extract the sql query that represents the (implied?) table of pairs to evaluate for matchy-ness?

From error tracebacks (when I've broken stuff) I know it gets computed as part of a CTE during the prediction step, but my ability to reverse engineer things further is limited.

Mostly, I'm interested in studying the resulting query since splink overall seems cleverly optimized and I'm hoping it'll teach me something.

[RobinL]

yes - there are a few options:

import logging
logging.getLogger("splink").setLevel(1)

If you set that after you instantiate the linker, all SQL will be logged

In addition, each Splink dataframe has a sql_used_to_create method and in addition the cache has an executed_queries list self.executed_queries = []