Add initial support for ibis.random function

docs/source/release.rst

@@ -9,6 +9,7 @@ Release Notes
 
 * :feature:`2048` Introduce a top level vectorized UDF module (experimental). Implement element-wise UDF for pandas and PySpark backend.
 * :release:`1.2.1 <pending>`
+* :feature:`2060` Add initial support for ibis.random function
 * :bug:`2061` CI: Fix CI builds related to new pandas 1.0 compatibility
 * :feature:`1976` Add DenseRank, RowNumber, MinRank, Count, PercentRank/CumeDist window operations to OmniSciDB
 * :bug:`2055` Fix "cudf" import on OmniSciDB backend

ibis/expr/api.py

@@ -187,6 +187,7 @@
     'param',
     'pi',
     'prevent_rewrite',
+    'random',
     'range_window',
     'row_number',
     'rows_with_max_lookback',
@@ -1344,6 +1345,11 @@ def clip(arg, lower=None, upper=None):
     return op.to_expr()
 
 
+def random():
+    op = ops.Random()
+    return op.to_expr()
+
+
 def quantile(arg, quantile, interpolation='linear'):
     """
     Return value at the given quantile, a la numpy.percentile.

ibis/expr/operations.py

@@ -2450,6 +2450,11 @@ def output_type(self):
         return dt.timestamp.scalar_type()
 
 
+class Random(Constant):
+    def output_type(self):
+        return dt.float64.scalar_type()
+
+
 class E(Constant):
     def output_type(self):
         return functools.partial(ir.FloatingScalar, dtype=dt.float64)

ibis/sql/mysql/compiler.py

@@ -191,6 +191,10 @@ def _literal(t, expr):
         return sa.literal(value)
 
 
+def _random(t, expr):
+    return sa.func.random()
+
+
 _operation_registry.update(
     {
         ops.Literal: _literal,
@@ -204,6 +208,7 @@ def _literal(t, expr):
         ops.Log2: unary(sa.func.log2),
         ops.Log10: unary(sa.func.log10),
         ops.Round: _round,
+        ops.Random: _random,
         # dates and times
         ops.Date: unary(sa.func.date),
         ops.DateAdd: infix_op('+'),

ibis/sql/postgres/compiler.py

@@ -607,6 +607,10 @@ def _literal(t, expr):
         return sa.literal(value)
 
 
+def _random(t, expr):
+    return sa.func.random()
+
+
 def _day_of_week_index(t, expr):
     (sa_arg,) = map(t.translate, expr.op().args)
     return sa.cast(
@@ -682,6 +686,7 @@ def _day_of_week_name(t, expr):
         ops.Max: _reduction('max'),
         ops.Variance: _variance_reduction('var'),
         ops.StandardDev: _variance_reduction('stddev'),
+        ops.Random: _random,
         # now is in the timezone of the server, but we want UTC
         ops.TimestampNow: lambda *args: sa.func.timezone('UTC', sa.func.now()),
         ops.CumulativeAll: unary(sa.func.bool_and),

ibis/tests/all/test_numeric.py

@@ -405,3 +405,17 @@ def test_sa_default_numeric_precision_and_scale(
 
     assert_equal(schema, expected)
     con.drop_table(table_name, force=True)
+
+
+@pytest.mark.only_on_backends([PostgreSQL, MySQL])
+def test_random(con):
+    expr = ibis.random()
+    result = con.execute(expr)
+    assert isinstance(result, float)
+    assert 0 <= result < 1
+
+    # Ensure a different random number of produced on subsequent calls
+    result2 = con.execute(expr)
+    assert isinstance(result2, float)
+    assert 0 <= result2 < 1
+    assert result != result2