def test_shapes(self):
# test shape tuple
shape1 = dc.vectorSizeT([2, 3, 4, 5])
shape2 = dc.vectorSizeT([5, 4, 3, 2])
a = dc.random(2, 3, 4, 5).asTypeInt()
assert a.rank() == 4
assert a.shape() == (2, 3, 4, 5)
# reshape to new dimensions
a.reshape(shape2)
assert a.shape() == (5, 4, 3, 2)
# return a new tensor with flattened dimensions.
b = a.flatten()
assert a.shape() == (5, 4, 3, 2)
assert b.shape() == (120, )
# flatten the same tensor
a.flatteninplace()
assert a.shape() == (120, )
shape3 = dc.vectorSizeT([8, 15, 1, 1])
# new shape
a.reshape(shape3)
# confirm new shape
assert a.shape() == (8, 15, 1, 1)
# dnnc method to reshape.
a = dc.random(2, 3, 4, 5)
dc.reshape(a, (8, 15, 1, 1))
assert a.shape() == (8, 15, 1, 1)
dc.reshape(a, (120, ))
assert a.shape() == (120, )
def test_dtypes(self):
a = dc.random(2, 3)
assert a.dtype() == 'float'
# transform datatype to int.
aint = a.asTypeInt()
assert aint.dtype() == 'int32_t'
# transform datatype to double.
adbl = a.asTypeDouble()
assert adbl.dtype() == 'double'
# transform datatype to double.
abool = a.asTypeBool()
assert abool.dtype() == 'bool'
def test_non_binary():
t1 = dc.array(2, 3)
t2 = dc.array(3, 2)
mul = dc.matmul(t1, t2)
#print ("multiplication : " , mul.to_string())
t3 = dc.array(2, 3, 4)
#print("old shape", t1.shape())
new_shape = dc.vectorSizeT([2, 12])
t3.reshape(new_shape)
t3.reshape(4, 6)
t3.reshape((4, 6))
t3.reshape([4, 6])
#print("new shape", t1.shape())
py_list = list(t3)
# convert tensor to python list
py_tuple = tuple(t3)
# convert tensor to python tuple
np_ary = t3.numpy()
# convert to numpy array
#t4 = dc.thresholded_relu(t1);
#print("relu", t4.to_string())
#replace first few values in tensor with new values.
data = dc.vectorFloat([1.0, 2.0, 3.0, 4.0])
t3.load(data)
#print(t3.to_string())
arr = dc.array([1, 2])
#print(arr)
arr2D = dc.array([[1, 2], [10, 20]]).asTypeInt()
#print(arr2D)
arrRand = dc.random(2, 3)
#print(arrRand)
empty = dc.empty(3, 2)
#print(empty)
zeros = dc.zeros(2, 2)
#print(zeros);
ones = dc.ones(2, 2)
#print(ones)
ranges = dc.arange(15, 3, 2)
#print(ranges)
dc.reshape(arr2D, (1, 4))
#3D MatMul Test1
a = dc.array(2, 2, 2)
b = dc.array(2, 2, 2)
adata = dc.vectorFloat([1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0])
bdata = dc.vectorFloat([8.0, 7.0, 6.0, 5.0, 4.0, 3.0, 2.0, 1.0])
a.load(adata)
b.load(bdata)
test_multiply(a, b)
#3D MatMul Test2
a = dc.array(2, 2, 3)
b = dc.array(2, 3, 2)
adata = dc.vectorFloat(
[1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0])
bdata = dc.vectorFloat(
[12.0, 11.0, 10.0, 9.0, 8.0, 7.0, 6.0, 5.0, 4.0, 3.0, 2.0, 1.0])
a.load(adata)
b.load(bdata)
test_multiply(a, b)
def test_create(self):
# null tensor test
a = dc.array(0)
assert a.isnull() == True
assert a.empty() == True
# test assignment is shallow copy of memory
b = a
assert a.sameas(b) == True
assert a.identifier() == b.identifier()
# tensor without initiliaztion
a = dc.array(2, 3, 4, 5)
assert a.length() == 120
# tensor random initiliaztion
a = dc.random(2, 3, 4, 5)
assert a.length() == 120
# tensor without initiliaztion
a = dc.empty(2, 3, 4, 5)
assert a.length() == 120
# zero tensor
a = dc.zeros(2, 3, 4, 5)
assert np.array(list(a.data())).sum().astype(np.int) == 0
# one tensor
a = dc.ones(2, 3, 4, 5)
assert np.array(list(a.data())).sum().astype(np.int) == 120
# tensor from python list
l1D = [1, 3, 5]
a = dc.array(l1D).asTypeInt()
np.testing.assert_equal(np.array(l1D), np.array(list(a.data())))
# tensor from python list of lists
l2D = [[1, 3, 5], [2, 4, 6]]
a = dc.array(l2D).asTypeInt()
assert a.rank() == 2
assert a.shape() == (2, 3)
np.testing.assert_equal(np.array(l2D).flatten(), \
np.array(list(a.data())))
# copy tensor
b = a.copy()
assert a.sameas(b) == False
assert a.identifier() != b.identifier()
# arange
a = dc.arange(10)
assert a.length() == 10
# add start and step
a = dc.arange(10, 5, 3).asTypeInt()
assert a.data() == (5, 8)
# swap start and stop.
a = dc.arange(5, 10, 3).asTypeInt()
assert a.data() == (5, 8)
def __init__(self, n):
self.N = n
self.dc_a = dc.random(n,n);
self.dc_b = dc.random(n,n);