compare#

versus.compare( table_a: DuckInput | 'DataFrame', table_b: DuckInput | 'DataFrame', *, by: Sequence[str], allow_both_na: bool = True, coerce: bool = True, table_id: Tuple[str, str] = ('a', 'b'), con: duckdb.DuckDBPyConnection | None = None, spark: 'SparkSession' | None = None, materialize: Literal['all', 'summary', 'none'] = 'all', ) → BaseComparison#

Compare two tables by key columns using DuckDB or PySpark.

Parameters:

table_a, table_bDuckDBPyRelation, pandas.DataFrame, polars.DataFrame, or pyspark.sql.DataFrame: Tabular inputs to compare. If either input is a PySpark DataFrame, or if spark= is supplied, the comparison runs on Spark and returns Spark-backed outputs. Otherwise DuckDB powers the comparison.
bysequence of str: Column names that uniquely identify rows.
allow_both_nabool, default True: Whether to treat NULL/NA values as equal when both sides are missing.
coercebool, default True: If True, allow DuckDB to coerce compatible types. If False, require exact type matches for shared columns.
table_idtuple[str, str], default (“a”, “b”): Labels used in outputs for the two tables.
conduckdb.DuckDBPyConnection, optional: DuckDB connection used to register non-Spark inputs and run queries.
sparkpyspark.sql.SparkSession, optional: Spark session used when the comparison runs on Spark.
materialize{“all”, “summary”, “none”}, default “all”: Controls which helper tables are materialized upfront.

Returns:

Comparison: Backend-specific comparison object with summary tables and diff helpers.

Examples

>>> from versus import compare, examples
>>> comparison = compare(
...     examples.example_cars_a(),
...     examples.example_cars_b(),
...     by=["car"],
... )
>>> comparison.summary()
┌────────────────┬─────────┐
│   difference   │  found  │
│    varchar     │ boolean │
├────────────────┼─────────┤
│ value_diffs    │ true    │
│ unmatched_cols │ true    │
│ unmatched_rows │ true    │
│ type_diffs     │ false   │
└────────────────┴─────────┘