feat(api): add levenshtein edit distance API

ci/schema/postgres.sql

@@ -3,6 +3,7 @@ CREATE EXTENSION IF NOT EXISTS postgis;
 CREATE EXTENSION IF NOT EXISTS plpython3u;
 CREATE EXTENSION IF NOT EXISTS vector;
 CREATE EXTENSION IF NOT EXISTS first_last_agg;
+CREATE EXTENSION IF NOT EXISTS fuzzystrmatch;
 
 DROP TABLE IF EXISTS diamonds CASCADE;
 

ibis/backends/oracle/registry.py

@@ -90,6 +90,7 @@ def _string_join(t, op):
         # Generic
         ops.Hash: unary(sa.func.ora_hash),
         ops.Literal: _literal,
+        ops.Levenshtein: fixed_arity(sa.func.utl_match.edit_distance, 2),
     }
 )
 

ibis/backends/postgres/registry.py

@@ -715,5 +715,6 @@ def _unnest(t, op):
         ops.ExtractMicrosecond: fixed_arity(
             lambda arg: sa.extract("microsecond", arg) % 1_000_000, 1
         ),
+        ops.Levenshtein: fixed_arity(sa.func.levenshtein, 2),
     }
 )

ibis/backends/pyspark/compiler.py

@@ -2059,3 +2059,10 @@ def compile_hash_column(t, op, **kwargs):
 @compiles(ops.ArrayZip)
 def compile_zip(t, op, **kwargs):
     return F.arrays_zip(*map(partial(t.translate, **kwargs), op.arg))
+
+
+@compiles(ops.Levenshtein)
+def compile_levenshtein(t, op, **kwargs):
+    left = t.translate(op.left, **kwargs)
+    right = t.translate(op.right, **kwargs)
+    return F.levenshtein(left, right)

ibis/backends/snowflake/registry.py

@@ -458,6 +458,7 @@ def _map_get(t, op):
         ops.ApproxMedian: reduction(lambda x: sa.func.approx_percentile(x, 0.5)),
         ops.Median: reduction(sa.func.median),
         ops.TableColumn: _table_column,
+        ops.Levenshtein: fixed_arity(sa.func.editdistance, 2),
     }
 )
 

ibis/backends/tests/test_string.py

@@ -1023,3 +1023,28 @@ def test_multiple_subs(con):
     expr = ibis.literal("foo").substitute(m)
     result = con.execute(expr)
     assert result == "FOO"
+
+
+@pytest.mark.notimpl(
+    [
+        "clickhouse",
+        "dask",
+        "datafusion",
+        "druid",
+        "impala",
+        "mssql",
+        "mysql",
+        "pandas",
+        "polars",
+        "sqlite",
+    ],
+    raises=com.OperationNotDefinedError,
+)
+@pytest.mark.parametrize(
+    "right", ["sitting", ibis.literal("sitting")], ids=["python", "ibis"]
+)
+def test_levenshtein(con, right):
+    left = ibis.literal("kitten")
+    expr = left.levenshtein(right)
+    result = con.execute(expr)
+    assert result == 3

ibis/backends/trino/registry.py

@@ -473,6 +473,7 @@ def _try_cast(t, op):
             lambda arg: sa.cast(sa.func.date_format(arg, "%f"), sa.INTEGER()),
             1,
         ),
+        ops.Levenshtein: fixed_arity(sa.func.levenshtein_distance, 2),
     }
 )
 

ibis/expr/operations/strings.py

@@ -298,3 +298,12 @@ class StringContains(Value):
 
     output_shape = rlz.shape_like("args")
     output_dtype = dt.bool
+
+
+@public
+class Levenshtein(Value):
+    left = rlz.string
+    right = rlz.string
+
+    output_dtype = dt.int64
+    output_shape = rlz.shape_like("args")

ibis/expr/types/strings.py

@@ -1515,6 +1515,29 @@ def __mul__(self, n: int | ir.IntegerValue) -> StringValue | NotImplemented:
 
     __rmul__ = __mul__
 
+    def levenshtein(self, other: StringValue) -> ir.IntegerValue:
+        """Return the Levenshtein distance between two strings.
+
+        Parameters
+        ----------
+        other
+            String to compare to
+
+        Returns
+        -------
+        IntegerValue
+            The edit distance between the two strings
+
+        Examples
+        --------
+        >>> import ibis
+        >>> ibis.options.interactive = True
+        >>> s = ibis.literal("kitten")
+        >>> s.levenshtein("sitting")
+        3
+        """
+        return ops.Levenshtein(self, other).to_expr()
+
 
 @public
 class StringScalar(Scalar, StringValue):