def _div(x, y):
r"""Computes the division of x by y for each element, scalar and broadcast promotation are supported.
The formula is:
.. math::
out = \frac{X}{Y}
Args:
x (Union[int, float, flow.Tensor]): X.
y (Union[int, float, flow.Tensor]): Y.
For example:
.. code-block:: python
import oneflow as flow
# element-wise divide
x = flow.Tensor(np.random.randn(2,3))
y = flow.Tensor(np.random.randn(2,3))
out = flow.div(x,y).numpy()
print(out.shape) # (2,3)
# scalar divide
x = 5
y = flow.Tensor(np.random.randn(2,3))
out = flow.div(x,y).numpy()
print(out.shape) # (2,3)
# broadcast divide
x = flow.Tensor(np.random.randn(1,1))
y = flow.Tensor(np.random.randn(2,3))
out = flow.div(x,y).numpy()
print(out.shape) # (2,3)
"""
if isinstance(x, (int, float)):
return ScalarMul(x)(flow.reciprocal(y))
elif isinstance(y, (int, float)):
if y == 0 or y == 0.0:
y = 0.0
else:
y = 1.0 / (float(y))
return ScalarMul(y)(x)
elif x.shape == y.shape:
return BroadcastDiv()(x, y)
elif y.shape == (1, ):
return ScalarDivByTensor()(x, y)
else:
return BroadcastDiv()(x, y)
def test_reciprocal(test_case):
x = flow.Tensor(np.random.randn(2, 3))
of_out = flow.reciprocal(x)
np_out = np.reciprocal(x.numpy())
test_case.assertTrue(np.allclose(of_out.numpy(), np_out, 1e-4, 1e-4))
def _reciprocal(self):
return flow.reciprocal(self)
def forward(self, x):
return flow.reciprocal(x)
