def test_tensor_from_numpy():
with pytest.raises(RuntimeError) as e:
_ts.MatrixF(np.array([1, 2, 3])) # trying to assign a int array
assert str(e.value) == "Incompatible buffer format!"
array = np.array([[1, 2, 3], [4, 5, 6]]).astype(np.float32)
tensor = _ts.MatrixF(array)
for i in range(tensor.shape()[0]):
for j in range(tensor.shape()[1]):
assert tensor[i, j] == array[i, j]
def test_cross_entropy_loss():
loss = _ts.CrossEntropyLoss()
logits = _ts.MatrixF(np.random.randn(300, 3).astype(np.float32))
labels = _ts.VectorI(np.random.randint(0, 2, (300,)).astype(np.int32))
loss_value = loss.forward(logits, labels)
loss.backward()
assert loss_value != 0.0
def test_numpy_from_tensor():
tensor = _ts.MatrixF(5, 4)
assert memoryview(tensor).shape == (5, 4)
assert tensor[2, 3] == 0.
tensor[2, 3] = 11.
tensor[3, 2] = 7.
assert tensor[2, 3] == 11.
assert tensor[3, 2] == 7.
array = np.array(tensor)
assert array.shape == (5, 4)
assert array[2, 3] == 11.
assert array[3, 2] == 7.
assert abs(array).sum() == 11 + 7
array[2, 3] = 5.
assert array[2, 3] == 5.
def test_matrix_f_init():
tensor = _ts.MatrixF(2, 2)
assert tensor.shape() == [2, 2]
assert tensor.data_size() == 4
def test_matrix_f_setitem():
tensor = _ts.MatrixF(2, 2)
assert tensor[1, 1] == 0.
tensor[1, 1] = 1337.
assert tensor[1, 1] == 1337.
def test_matrix_f_getitem():
tensor = _ts.MatrixF(2, 2)
assert tensor[0, 0] == 0.
assert tensor[0, 1] == 0.
assert tensor[1, 0] == 0.
assert tensor[1, 1] == 0.