def test_integer(self):
a = sail.random.uniform(0, 11, (10, 20)).astype(sail.int32)
b = sail.random.uniform(0, 11, (20, 10)).astype(sail.int32)
c = sail.matmul(a, b)
a = a.numpy()
b = b.numpy()
c2 = np.matmul(a, b)
self.assert_eq_np_sail(c2, c, eps=1e-7)
def test_base(self, rq):
choices = [(3, 3), (12, 18), (2, 33), (32, 64)]
choices_2 = [[(3, 3), (3, 1), (3, 10)], [(18, 12), (18, 2)],
[(33, 1), (33, 33)], [(64, 12)]]
times = []
for ca, cbs in zip(choices, choices_2):
for cb in cbs:
verif_shape = (ca[0], cb[1])
arr1 = np.random.uniform(0, 1, (ca))
arr2 = np.random.uniform(0, 1, (cb))
x1 = sail.Tensor(arr1, requires_grad=rq)
x2 = sail.Tensor(arr2, requires_grad=rq)
x3 = sail.matmul(x1, x2)
arr3 = np.matmul(arr1, arr2)
self.assert_eq(x3.shape, verif_shape)
self.assert_eq_np_sail(arr3, x3, eps=1e-7)
self.assert_eq(x3.requires_grad, rq)
return
tensor (Tensor): Input data
axis (int or tuple of ints, optional): If provided, then `axis` represents the axis the min will be computed over. If `axis` is a tuple, then the axes provided will be used to compute the min
keepdims (boolean, optional): If True, then the axes that are reduced will be replaced with 1, otherwise, those axes will be removed
Examples:
>>> x = sail.random.uniform(0, 1, (12, 32, 4, 5))
>>> y = sail.min(x, 1, True)
>>> y.shape
(12, 1, 4, 5)
>>> z = sail.min(x, -2, False)
>>> z.shape
(12, 32, 5)
"""
add_docstring(sail.min, descr)
descr = r"""
sail.matmul(x1, x2) -> Tensor
Returns the matrix multiplication of `x1` and `x2`
.. math::
\text{out} = \text{x1} \cdot \text{x1}
.. note::
Both tensor `x1` and `x2` must be 2D, and their inner shapes must match.
Args:
x1 (Tensor): A 2D Tensor
x2 (Tensor): A 2D Tensor
Examples:
>>> a = sail.random.uniform(0, 1, (5, 6))
def forward(a, b):
c = sail.matmul(a, b)
d = sail.sum(c)
return d