def cat(tensors, dim=0, out=None):
x = paddle.concat(tensors, axis=dim)
if out is None:
return varbase_to_tensor(x)
else:
paddle.assign(x, out)
return out
def cat(tensors, dim=0, out=None):
x = fluid.layers.concat(tensors, axis=dim)
if out is None:
return varbase_to_tensor(x)
else:
fluid.layers.assign(x, out)
return out
def randn(*shape, requires_grad=True):
if isinstance(shape[0], Iterable):
shape = shape[0]
X = varbase_to_tensor(paddle.randn(shape))
if not requires_grad:
X.stop_gradient = True
return X
def stack(inputs, dim=0, out=None):
x = paddle.stack(inputs, axis=dim)
if out is None:
return varbase_to_tensor(x)
else:
paddle.assign(x, out)
return out
def stack(inputs, dim=0, out=None):
x = fluid.layers.stack(inputs, axis=dim)
if out is None:
return varbase_to_tensor(x)
else:
fluid.layers.assign(x, out)
return out
def zeros(*size, out=None, dtype="float32", device=None, requires_grad=True):
if isinstance(size[0], Iterable):
size = size[0]
if isinstance(size[0], Iterable):
size = size[0]
X = varbase_to_tensor(paddle.zeros(size, dtype))
if not requires_grad:
X.stop_gradient = True
return X
def mean(input, dim=None, keepdim=False, out=None):
if isinstance(dim, tuple):
dim = list(dim)
x = fluid.layers.reduce_mean(input, dim, keepdim)
if out is None:
return varbase_to_tensor(x)
else:
fluid.layers.assign(x, out)
return out
def chunk(self, chunks, dim):
slices = fluid.layers.unstack(self, axis=dim, num=None)
out_list = []
step = int(np.ceil(len(slices) / chunks))
for st in range(0, len(slices), step):
out_list.append(
varbase_to_tensor(
fluid.layers.concat([
paddle.fluid.layers.unsqueeze(x, dim, name=None)
for x in slices[st:(st + step)]
],
axis=dim,
name=None)))
return out_list
def zeros_like(x, out=None, device=None):
return varbase_to_tensor(paddle.zeros_like(x, out))
def pow(x, y):
return varbase_to_tensor(fluid.layers.pow(x, y))
def sqrt(x):
return varbase_to_tensor(fluid.layers.sqrt(x))
def sum(x, dim=None, keepdim=False):
return varbase_to_tensor(fluid.layers.reduce_sum(x, dim, keepdim))
def rsqrt(x):
return varbase_to_tensor(paddle.fluid.layers.rsqrt(x, name=None))
def flatten(x, dim=1):
x = fluid.layers.flatten(x, axis=dim)
return varbase_to_tensor(x)
def ones(*size, out=None, dtype="float32", device=None):
if isinstance(size[0], Iterable):
size = size[0]
return varbase_to_tensor(paddle.ones(size, dtype))
def zeros(*size, out=None, dtype="float32", device=None, requires_grad=True):
X = varbase_to_tensor(fluid.layers.zeros(size, dtype))
if not requires_grad:
X.stop_gradient = True
return X
def randn(*shape, requires_grad=True):
X = varbase_to_tensor(fluid.layers.randn(shape))
if not requires_grad:
X.stop_gradient = True
return X
def flatten(x, start_dim=0, end_dim=-1):
x = paddle.flatten(x, start_axis=start_dim, stop_axis=end_dim)
return varbase_to_tensor(x)
def ones(*size, out=None, dtype="float32", device=None):
return varbase_to_tensor(paddle.ones(size, dtype))
def rsqrt(x):
return varbase_to_tensor(paddle.rsqrt(x, name=None))
def sqrt(x):
return varbase_to_tensor(paddle.sqrt(x))
def pow(x, y):
return varbase_to_tensor(paddle.pow(x, y))
def ones(*size, out=None, dtype="float32", device=None):
return varbase_to_tensor(fluid.layers.ones(size, dtype))
def randn(*shape, requires_grad=True):
X = varbase_to_tensor(paddle.randn(*shape))
if not requires_grad:
X.stop_gradient = True
return X
def zeros_like(x, out=None, device=None):
return varbase_to_tensor(fluid.layers.zeros_like(x, out))
def flatten(x, dim=1):
x = paddle.flatten(x, axis=dim)
return varbase_to_tensor(x)
def clamp(input, min, max, out=None):
return varbase_to_tensor(paddle.clip(input, min, max))
def min(x, dim=None, keepdim=False):
return varbase_to_tensor(fluid.layers.reduce_min(x, dim, keep_dim=keepdim))
def clamp(input, min, max, out=None):
return varbase_to_tensor(fluid.layers.clip(input, min, max))
def min(x, dim=None, keepdim=False):
return varbase_to_tensor(paddle.min(x, dim, keepdim=keepdim))
