def __init__(self, num_feats, in_feats, num_hops, sample_size):
super(PartialWeightedAggregator, self).__init__()
self.weight_store = []
self.agg_feats = nn.ParameterList()
self.discounts = nn.ParameterList()
self.num_hops = num_hops
for _ in range(num_hops):
self.weight_store.append(paddle.Tensor(num_feats, in_feats))
# self.agg_feats.append(nn.Parameter(torch.Tensor(sample_size, in_feats)))
# self.discounts.append(nn.Parameter(torch.Tensor(in_feats)))
# nn.init.xavier_uniform_(self.weight_store[-1])
self.agg_feats.append(paddle.create_parameter(shape=paddle.Tensor(sample_size, in_feats),dtype='float32',attr=paddle.framework.ParamAttr(name="linear_weight", initializer=paddle.nn.initializer.XavierNormal(self.agg_feats[-1]))))
self.discounts.append(paddle.create_parameter(shape=paddle.Tensor(in_feats),dtype='float32',attr=paddle.framework.ParamAttr(name="linear_weight", initializer=paddle.nn.initializer.XavierNormal(self.agg_feats[-1]))))
self.reset_parameters()
def test_io_struct():
n = 16
x1 = ti.Struct.field({"a": ti.i32, "b": ti.f32}, shape=(n, ))
p1 = {
"a": paddle.Tensor(2 * np.ones(n, dtype=np.int32)),
"b": paddle.Tensor(3 * np.ones(n, dtype=np.float32)),
}
x1.from_paddle(p1)
for i in range(n):
assert x1[i].a == 2
assert x1[i].b == 3
p2 = x1.to_paddle()
for k in p1:
assert (p1[k] == p2[k]).all()
def tokenize(texts: Union[str, List[str]], context_length: int = 77):
"""
Returns the tokenized representation of given input string(s)
Parameters
----------
texts : Union[str, List[str]]
An input string or a list of input strings to tokenize
context_length : int
The context length to use; all CLIP models use 77 as the context length
Returns
-------
A two-dimensional tensor containing the resulting tokens, shape = [number of input strings, context_length]
"""
if isinstance(texts, str):
texts = [texts]
sot_token = _tokenizer.encoder["<|startoftext|>"]
eot_token = _tokenizer.encoder["<|endoftext|>"]
all_tokens = [[sot_token] + _tokenizer.encode(text) + [eot_token]
for text in texts]
result = paddle.zeros((len(all_tokens), context_length), dtype='int64')
for i, tokens in enumerate(all_tokens):
if len(tokens) > context_length:
raise RuntimeError(
f"Input {texts[i]} is too long for context length {context_length}"
)
result[i, :len(tokens)] = paddle.Tensor(np.array(tokens))
return result
def pbtPredict_Callback(self):
__img, img_array = [], [
] # 将图像统一从qimage->pil image -> np.array [1, 1, 28, 28]
# 获取qimage格式图像
if self.mode == MODE_MNIST:
__img = self.lbDataArea.pixmap() # label内若无图像返回None
if __img == None: # 无图像则用纯黑代替
# __img = QImage(224, 224, QImage.Format_Grayscale8)
__img = ImageQt.ImageQt(
Image.fromarray(np.uint8(np.zeros([224, 224]))))
else:
__img = __img.toImage()
elif self.mode == MODE_WRITE:
__img = self.paintBoard.getContentAsQImage()
# 转换成pil image类型处理
pil_img = ImageQt.fromqimage(__img)
pil_img = pil_img.resize((28, 28), Image.ANTIALIAS)
img_array = np.array(pil_img.convert('L')).reshape(1, 1, 28, 28)
img = normalize(img_array)
img = paddle.Tensor(img)
__result = network(img)
argmax__result = paddle.argmax(__result).numpy()
self.result[0] = argmax__result[0] # 置信度
m = F.sigmoid(__result).numpy()
self.result[1] = m[0][self.result[0]]
self.lbResult.setText("%d" % (self.result[0]))
self.lbCofidence.setText("%.8f" % (self.result[1]))
def test_type_Tensor(self):
paddle.disable_static()
inx = np.array([1, 2])
tensorx = paddle.Tensor(inx)
typex_str = str(type(tensorx))
expectx = "<class 'paddle.Tensor'>"
self.assertEqual((typex_str == expectx), True)
tensorx = paddle.tensor.logic.Tensor(inx)
typex_str = str(type(tensorx))
expectx = "<class 'paddle.Tensor'>"
self.assertEqual((typex_str == expectx), True)
def to_tensor(string_values, name="text"):
"""
Create the tensor that the value holds the list of string.
NOTICE: The value will be holded in the cpu place.
Parameters:
string_values(list[string]): The value will be setted to the tensor.
name(string): The name of the tensor.
"""
tensor = paddle.Tensor(core.VarDesc.VarType.STRING, [], name,
core.VarDesc.VarType.STRINGS, False)
tensor.value().set_string_list(string_values)
return tensor
def to_map_tensor(string_dict, name):
"""
Create the tensor that the value holds the map, the type of key is the string
and the value is the int.
NOTICE: The value will be holded in the cpu place.
Args:
string_dict(dict): The value will be setted to the tensor.
name(string): The name of the tensor.
"""
tensor = paddle.Tensor(core.VarDesc.VarType.RAW, [], name,
core.VarDesc.VarType.VOCAB, True)
tensor.value().set_vocab(string_dict)
return tensor
def test_io_simple():
n = 32
x1 = ti.field(ti.f32, shape=(n, n))
p1 = paddle.Tensor(3 * np.ones((n, n), dtype=np.float32))
x2 = ti.Matrix.field(2, 3, ti.f32, shape=(n, n))
p2 = paddle.Tensor(3 * np.ones((n, n, 2, 3), dtype=np.float32))
x1.from_paddle(p1)
for i in range(n):
for j in range(n):
assert x1[i, j] == 3
x2.from_paddle(p2)
for i in range(n):
for j in range(n):
for k in range(2):
for l in range(3):
assert x2[i, j][k, l] == 3
p3 = x2.to_paddle()
assert (p2 == p3).all()
def random_crop2d(s,
crop_len,
tempo_axis=0): # random crop according to temporal direction
assert tempo_axis < s.ndim, 'axis out of range'
n = s.shape[tempo_axis]
idx = randint(high=n - crop_len)
if type(s) == np.ndarray:
sli = [slice(None) for i in range(s.ndim)]
sli[tempo_axis] = slice(idx, idx + crop_len)
out = s[tuple(sli)]
else:
out = paddle.index_select(
s,
paddle.Tensor(np.array([i for i in range(idx, idx + crop_len)])),
axis=tempo_axis)
return out
def from_dlpack(dlpack):
tensor_from_dlpack = fluid.core.from_dlpack(dlpack)
place = tensor_from_dlpack._place()
if True: # "win" in platform: # CPU env
if "int32" in str(tensor_from_dlpack):
return paddorch.convertTensor(
paddle.to_tensor(np.array(tensor_from_dlpack), dtype="int32"))
else:
return paddorch.Tensor(
paddle.to_tensor(np.array(tensor_from_dlpack)))
else:
with paddle.fluid.dygraph.guard(place=place):
tensor_from_dlpack.__class__ = paddle.fluid.LoDTensor
ret = paddle.Tensor(tensor_from_dlpack)
if "int32" in str(tensor_from_dlpack):
ret = paddle.to_tensor(ret, dtype="int32")
tensor_from_dlpack.__class__ = paddle.fluid.core_avx.Tensor
return ret
def load_and_extract_feature(file):
s, r = pa.load(file, sr=c['sample_rate'])
x = pa.features.mel_spect(s,
sample_rate=c['sample_rate'],
window_size=c['window_size'],
hop_length=c['hop_size'],
mel_bins=c['mel_bins'],
fmin=c['fmin'],
fmax=c['fmax'],
window='hann',
center=True,
pad_mode='reflect',
ref=1.0,
amin=1e-10,
top_db=None)
x = x.T #!!
x = paddle.Tensor(x).unsqueeze((0, 1))
return x
RN50_TEXT_MEAN = 0.20489279925823212
VIT_TEXT_MEAN = 0.2554764747619629
#ViT-B/32 0.2554764747619629
#RN50 0.20489279925823212
#RN101 0.1773824542760849
print('Start testing to make sure all models are aligned with official impl')
print('======testing RN101======')
model = build_rn101_model()
sd = paddle.load('./assets/RN101.pdparams')
model.load_dict(sd)
image_input = np.load('./assets/image.npy')
image_input = paddle.Tensor(image_input)
model.eval()
text = tokenize(["a diagram", "a dog", "a cat"])
text_feature = model.encode_text(text)
if abs(float(paddle.mean(text_feature**2)) - RN101_TEXT_MEAN) < EPS:
print('Testing passed for text transformer')
else:
print('Testing failed for text transformer')
out = model.encode_image(image_input)
if abs(float(paddle.mean(out**2)) - RN101_OUT_MEAN) < EPS:
print('Testing passed for RN101 image encoder')
else:
print('Testing failed for RN101 image encoder')
paddle.Tensor.view = paddorch.Tensor.view paddle.Tensor.repeat = paddorch.Tensor.repeat paddle.Tensor.add = paddorch.Tensor.add paddle.Tensor.item = paddorch.Tensor.item paddle.Tensor.t = paddorch.Tensor.t paddle.Tensor.reshape = paddorch.Tensor.reshape paddle.Tensor.__setitem__ = paddorch.Tensor.__setitem__ paddle.Tensor.__getitem__ = paddorch.Tensor.__getitem__ paddle.Tensor.index_copy_ = paddorch.Tensor.index_copy_ paddle.Tensor.index_copy = paddorch.Tensor.index_copy paddle.Tensor.new_empty = paddorch.Tensor.new_empty paddle.Tensor.view_as = paddorch.Tensor.view_as paddle.Tensor.clamp = paddorch.Tensor.clamp paddle.Tensor.requires_grad_ = paddorch.Tensor.requires_grad_ paddle.Tensor.set_gradient = paddorch.Tensor.set_gradient paddle.Tensor.backward = paddorch.Tensor.backward paddle.Tensor.new_zeros = paddorch.Tensor.new_zeros paddle.Tensor.new_ones = paddorch.Tensor.new_ones paddle.Tensor.sort = paddorch.Tensor.sort paddle.Tensor.index_select = paddorch.Tensor.index_select paddle.Tensor.masked_fill_ = paddorch.Tensor.masked_fill_ paddle.Tensor.argmax = paddorch.Tensor.argmax paddle.Tensor.tolist = paddorch.Tensor.tolist paddle.Tensor.uniform_ = paddorch.Tensor.uniform_ paddle.Tensor.__getstate__ = paddorch.Tensor.__getstate__ paddle.Tensor.__setstate__ = paddorch.Tensor.__setstate__ import numpy as np a = paddle.Tensor(np.random.rand(3, 4)) print(a.view(-1))
out=np.array([[[0.9427, 0.0364, 0.2218],
[0.5494, 0.0364, 0.2218],
[0.4433, 0.3639, 0.2218],
[0.4433, 0.0364, 0.2587],
[0.5494, 0.0364, 0.2218]],
[[0.2322, 0.3581, 0.3620],
[0.6472, 0.0879, 0.7137],
[0.2322, 0.3581, 0.8765],
[0.6472, 0.9749, 0.7137],
[0.6472, 0.9749, 0.7137]]])
print(out.shape)
torch_out=torch.gather(torch.FloatTensor(x),dim=-1,index=torch.LongTensor(i))
print("torch out",torch_out)
ind=paddle.Tensor(i.astype("int64"))
ind=ind.flatten()
row=paddle.expand( paddle.reshape(paddle.arange(x.shape[0]), (i.shape[0],1,1)), (i.shape)).flatten()
print(row)
col=paddle.expand( paddle.reshape(paddle.arange(x.shape[1]), (1,i.shape[1],1)), (i.shape)).flatten()
print(col)
ind2=paddle.stack([row,col,ind]).transpose([1,0])
print(ind2.shape)
paddle_out=paddle.gather_nd(paddle.Tensor(x), ind2).reshape(i.shape)
print("paddle out",paddle_out)
batch_size=64,
shuffle=False)
# net definition
net = Net(reid=True)
assert os.path.isfile(
"./checkpoint/ckpt.t7"), "Error: no checkpoint file found!"
print('Loading from checkpoint/ckpt.t7')
checkpoint = torch.load("./checkpoint/ckpt.t7")
net_dict = checkpoint['net_dict']
net.load_state_dict(net_dict, strict=False)
net.eval()
net.to(device)
# compute features
query_features = torch.Tensor([]).float()
query_labels = torch.Tensor([]).long()
gallery_features = torch.Tensor([]).float()
gallery_labels = torch.Tensor([]).long()
with torch.no_grad():
for idx, (inputs, labels) in enumerate(queryloader):
inputs = inputs.to(device)
features = net(inputs).cpu()
query_features = cat((query_features, features), axis=0)
query_labels = cat((query_labels, labels))
for idx, (inputs, labels) in enumerate(galleryloader):
inputs = inputs.to(device)
features = net(inputs).cpu()
gallery_features = cat((gallery_features, features), axis=0)
def fmod(x, y):
if isinstance(y, int):
y = paddle.Tensor(np.array([y], dtype="float32"))
return convertTensor(paddle.floor_mod(x, y))
def test_sqeuclidean(get_a_b):
a, b = get_a_b
r_torch = sqe_paddle(paddle.Tensor(a), paddle.Tensor(b))
r_numpy = sqe_numpy(a, b)
np.testing.assert_almost_equal(r_torch, r_numpy)
def to_tensor(data, dtype=None, place=None, stop_gradient=True):
r"""
Constructs a ``paddle.Tensor`` from ``data`` ,
which can be scalar, tuple, list, numpy\.ndarray, paddle\.Tensor.
If the ``data`` is already a tensor, and ``dtype`` or ``place`` does't change, no copy
will be performed and return origin tensor, otherwise a new tensor will be constructed
and returned.
Args:
data(scalar|tuple|list|ndarray|Tensor): Initial data for the tensor.
Can be a scalar, list, tuple, numpy\.ndarray, paddle\.Tensor.
dtype(str|np.dtype, optional): The desired data type of returned tensor. Can be 'bool' , 'float16' ,
'float32' , 'float64' , 'int8' , 'int16' , 'int32' , 'int64' , 'uint8',
'complex64' , 'complex128'. Default: None, infers dtype from ``data``
except for python float number which gets dtype from ``get_default_type`` .
place(CPUPlace|CUDAPinnedPlace|CUDAPlace, optional): The place to allocate Tensor. Can be
CPUPlace, CUDAPinnedPlace, CUDAPlace. Default: None, means global place.
stop_gradient(bool, optional): Whether to block the gradient propagation of Autograd. Default: True.
Returns:
Tensor: A Tensor constructed from ``data`` .
Raises:
TypeError: If the data type of ``data`` is not scalar, list, tuple, numpy.ndarray, paddle.Tensor
ValueError: If ``data`` is tuple|list, it can't contain nested tuple|list with different lengths , such as: [[1, 2], [3, 4, 5]]
TypeError: If ``dtype`` is not bool, float16, float32, float64, int8, int16, int32, int64, uint8, complex64, complex128
ValueError: If ``place`` is not paddle.CPUPlace, paddle.CUDAPinnedPlace, paddle.CUDAPlace
Examples:
.. code-block:: python
import paddle
type(paddle.to_tensor(1))
# <class 'paddle.Tensor'>
paddle.to_tensor(1)
# Tensor(shape=[1], dtype=int64, place=CUDAPlace(0), stop_gradient=True,
# [1])
x = paddle.to_tensor(1)
paddle.to_tensor(x, dtype='int32', place=paddle.CPUPlace()) # A new tensor will be constructed due to different dtype or place
# Tensor(shape=[1], dtype=int32, place=CPUPlace, stop_gradient=True,
# [1])
paddle.to_tensor((1.1, 2.2), place=paddle.CUDAPinnedPlace())
# Tensor(shape=[1], dtype=float32, place=CUDAPinnedPlace, stop_gradient=True,
# [1])
paddle.to_tensor([[0.1, 0.2], [0.3, 0.4]], place=paddle.CUDAPlace(0), stop_gradient=False)
# Tensor(shape=[2, 2], dtype=float32, place=CUDAPlace(0), stop_gradient=False,
# [[0.10000000, 0.20000000],
# [0.30000001, 0.40000001]])
type(paddle.to_tensor([[1+1j, 2], [3+2j, 4]], dtype='complex64'))
# <class 'paddle.VarBase'>
paddle.to_tensor([[1+1j, 2], [3+2j, 4]], dtype='complex64')
# Tensor(shape=[2, 2], dtype=complex64, place=CUDAPlace(0), stop_gradient=True,
# [[(1+1j), (2+0j)],
# [(3+2j), (4+0j)]])
"""
if place is None:
place = _current_expected_place()
elif not isinstance(
place,
(core.Place, core.CPUPlace, core.CUDAPinnedPlace, core.CUDAPlace)):
raise ValueError(
"'place' must be any of paddle.Place, paddle.CPUPlace, paddle.CUDAPinnedPlace, paddle.CUDAPlace"
)
#Todo(zhouwei): Support allocate tensor on any other specified card
if isinstance(place, core.CUDAPlace) and isinstance(
_current_expected_place(), core.CUDAPlace
) and place._get_device_id() != _current_expected_place()._get_device_id():
place = _current_expected_place()
if not isinstance(data, np.ndarray):
if np.isscalar(data) and not isinstance(data, str):
data = np.array([data])
elif isinstance(data, (list, tuple)):
data = np.array(data)
if data.dtype == np.object:
raise ValueError(
"\n\tFaild to convert input data to a regular ndarray :\n\t - Usually "
"this means the input data contains nested lists with different lengths. "
)
elif isinstance(data, paddle.Tensor):
data.stop_gradient = stop_gradient
if not data.place._equals(place):
data = data._copy_to(place, False)
if dtype:
if convert_dtype(dtype) != convert_dtype(data.dtype):
return data.astype(convert_dtype(dtype))
return data
else:
raise TypeError(
"Can't constructs a 'paddle.Tensor' with data type {}, data type must be scalar|list|tuple|numpy.ndarray|paddle.Tensor"
.format(type(data)))
if not dtype and data.dtype in [
'float16', 'float32', 'float64', 'complex64', 'complex128'
]:
default_type = paddle.get_default_dtype()
if np.iscomplexobj(data):
default_type = 'complex64' if default_type in [
'float16', 'float32'
] else 'complex128'
data = data.astype(default_type)
if dtype and convert_dtype(dtype) != data.dtype:
data = data.astype(dtype)
return paddle.Tensor(value=data,
place=place,
persistable=False,
zero_copy=False,
stop_gradient=stop_gradient)
import paddle tensor = paddle.randn((3, 4)) dlpack = tensor.value().get_tensor()._to_dlpack() tensor_from_dlpack = paddle.fluid.core.from_dlpack(dlpack) tensor_from_dlpack.__class__ = paddle.fluid.LoDTensor bb = paddle.Tensor(tensor_from_dlpack) bb = bb.cpu() print(bb) paddle.set_device() tensor_from_dlpack.__class__ = paddle.fluid.core_avx.Tensor # paddle.fluid.dygraph.to_variable( tensor_from_dlpack)
def to_tensor(data, dtype=None, place=None, stop_gradient=True):
"""
Constructs a ``paddle.Tensor`` or ``paddle.ComplexTensor`` from ``data`` ,
which can be scalar, tuple, list, numpy\.ndarray, paddle\.Tensor, paddle\.ComplexTensor.
If the ``data`` is already a tensor, and ``dtype`` or ``place`` does't change, no copy
will be performed and return origin tensor, otherwise a new tensor will be constructed
and returned. Similarly, if the data is an numpy\.ndarray of with the same ``dtype``
and the current place is cpu, no copy will be performed.
The ``ComplexTensor`` is a unique type of paddle. If x is ``ComplexTensor``, then
``x.real`` is the real part, and ``x.imag`` is the imaginary part.
Args:
data(scalar|tuple|list|ndarray|Tensor|ComplexTensor): Initial data for the tensor.
Can be a scalar, list, tuple, numpy\.ndarray, paddle\.Tensor, paddle\.ComplexTensor.
dtype(str|np.dtype, optional): The desired data type of returned tensor. Can be 'bool' , 'float16' ,
'float32' , 'float64' , 'int8' , 'int16' , 'int32' , 'int64' , 'uint8'. And
'complex64' , 'complex128' only for ComplexTensor. Default: None, infers dtype from ``data``
except for python float number which gets dtype from ``get_default_type`` .
place(CPUPlace|CUDAPinnedPlace|CUDAPlace, optional): The place to allocate Tensor. Can be
CPUPlace, CUDAPinnedPlace, CUDAPlace. Default: None, means global place.
stop_gradient(bool, optional): Whether to block the gradient propagation of Autograd. Default: True.
Returns:
Tensor: A Tensor or ComplexTensor constructed from ``data`` .
Raises:
TypeError: If the data type of ``data`` is not scalar, list, tuple, numpy.ndarray, paddle.Tensor, paddle.ComplexTensor
ValueError: If ``data`` is tuple|list, it can't contain nested tuple|list with different lengths , such as: [[1, 2], [3, 4, 5]]
TypeError: If ``dtype`` is not bool, float16, float32, float64, int8, int16, int32, int64, uint8, complex64, complex128
ValueError: If ``place`` is not paddle.CPUPlace, paddle.CUDAPinnedPlace, paddle.CUDAPlace
Examples:
.. code-block:: python
import paddle
import numpy as np
paddle.disable_static()
type(paddle.to_tensor(1))
# <class 'paddle.Tensor'>
paddle.to_tensor(1)
# Tensor: generated_tensor_0
# - place: CUDAPlace(0) # allocate on global default place CPU:0
# - shape: [1]
# - layout: NCHW
# - dtype: int64_t
# - data: [1]
x = paddle.to_tensor(1)
paddle.to_tensor(x, dtype='int32', place=paddle.CPUPlace()) # A new tensor will be constructed due to different dtype or place
# Tensor: generated_tensor_01
# - place: CPUPlace
# - shape: [1]
# - layout: NCHW
# - dtype: int
# - data: [1]
paddle.to_tensor((1.1, 2.2), place=paddle.CUDAPinnedPlace())
# Tensor: generated_tensor_1
# - place: CUDAPinnedPlace
# - shape: [2]
# - layout: NCHW
# - dtype: double
# - data: [1.1 2.2]
paddle.to_tensor([[0.1, 0.2], [0.3, 0.4]], place=paddle.CUDAPlace(0), stop_gradient=False)
# Tensor: generated_tensor_2
# - place: CUDAPlace(0)
# - shape: [2, 2]
# - layout: NCHW
# - dtype: double
# - data: [0.1 0.2 0.3 0.4]
type(paddle.to_tensor([[1+1j, 2], [3+2j, 4]]), dtype='complex64')
# <class 'paddle.ComplexTensor'>
paddle.to_tensor([[1+1j, 2], [3+2j, 4]], dtype='complex64')
# ComplexTensor[real]: generated_tensor_0.real
# - place: CUDAPlace(0)
# - shape: [2, 2]
# - layout: NCHW
# - dtype: float
# - data: [1 2 3 4]
# ComplexTensor[imag]: generated_tensor_0.imag
# - place: CUDAPlace(0)
# - shape: [2, 2]
# - layout: NCHW
# - dtype: float
# - data: [1 0 2 0]
"""
if place is None:
place = _current_expected_place()
elif not isinstance(place,
(core.CPUPlace, core.CUDAPinnedPlace, core.CUDAPlace)):
raise ValueError(
"'place' must be any of paddle.Place, paddle.CUDAPinnedPlace, paddle.CUDAPlace"
)
#Todo(zhouwei): Support allocate tensor on any other specified card
if isinstance(place, core.CUDAPlace) and isinstance(
_current_expected_place(), core.CUDAPlace) and place._get_device_id(
) != _current_expected_place()._get_device_id():
place = _current_expected_place()
if not isinstance(data, np.ndarray):
if np.isscalar(data) and not isinstance(data, str):
data = np.array([data])
elif isinstance(data, (list, tuple)):
data = np.array(data)
if data.dtype == np.object:
raise ValueError(
"\n\tFaild to convert input data to a regular ndarray :\n\t - Usually "
"this means the input data contains nested lists with different lengths. "
)
elif isinstance(data, paddle.Tensor):
data.stop_gradient = stop_gradient
if not data.place._equals(place):
data = data._copy_to(place, False)
if dtype:
if convert_dtype(dtype) != convert_dtype(data.dtype):
return data.astype(convert_dtype(dtype))
return data
elif isinstance(data, paddle.ComplexTensor):
return data
else:
raise TypeError(
"Can't constructs a 'paddle.Tensor' with data type {}, data type must be scalar|list|tuple|numpy.ndarray|paddle.Tensor|paddle.ComplexTensor".
format(type(data)))
if not dtype and data.dtype in [
'float16', 'float32', 'float64', 'complex64', 'complex128'
]:
default_type = paddle.get_default_dtype()
if np.iscomplexobj(data):
default_type = 'complex64' if default_type in [
'float16', 'float32'
] else 'complex128'
data = data.astype(default_type)
if dtype and convert_dtype(dtype) != data.dtype:
data = data.astype(dtype)
if not np.iscomplexobj(data):
if dtype and convert_dtype(dtype) != data.dtype:
data = data.astype(dtype)
return paddle.Tensor(
value=data,
place=place,
persistable=False,
zero_copy=True,
stop_gradient=stop_gradient)
else:
name = unique_name.generate('generated_tensor')
real_tensor = paddle.Tensor(
value=data.real,
place=place,
zero_copy=True,
name=name + ".real",
stop_gradient=stop_gradient)
imag_tensor = paddle.Tensor(
value=data.imag,
place=place,
zero_copy=True,
name=name + ".imag",
stop_gradient=stop_gradient)
return paddle.ComplexTensor(real_tensor, imag_tensor)
from paddle import fluid
import paddorch as torch
import paddle
from scipy.sparse import csr_matrix
from paddorch.nn.init import xavier_uniform_
from paddorch.sparse import FloatTensor
import numpy as np
place = fluid.CPUPlace()
with fluid.dygraph.guard(place=place):
i = torch.from_numpy(np.array([[0, 2], [1, 0], [1, 2]]) ).astype("int32")
v = paddle.Tensor(np.array([3, 4, 5])).astype("float32")
x=FloatTensor(i,v,(2,3))
print(x)
A = csr_matrix([[1, 2, 0], [0, 0, 3], [4, 0, 5]]).todense().astype("float32")
x= paddle.Tensor(A)
# x = torch.randn((4, 23, 16))
print(xavier_uniform_(x))
def test_cosine(get_a_b):
a, b = get_a_b
r_torch = cosine_paddle(paddle.Tensor(a), paddle.Tensor(b))
r_numpy = cosine_numpy(a, b)
np.testing.assert_almost_equal(r_torch, r_numpy)
