def choice(a, size=None, replace=True, p=None):
a = to_backend(a)
size = size if size is not None else 1
if replace:
size = size if isinstance(size, tuple) else (size,)
indices = to_backend(torch.randint(len(a), size))
samples = a[indices]
else:
indices = to_backend(torch.randperm(len(a)))[:size]
samples = a[indices]
return to_backend(samples)
def randint(low, high, size=None, dtype=None):
size = size if size is not None else (1,)
size = size if isinstance(size, tuple) else (size,)
data = torch.randint(low, high, size)
if dtype is not None:
data = data.to(dtype)
return to_backend(data)
def uniform(
low=0.0, high=1.0, size=None, dtype=None,
):
uniform = torch.distributions.uniform.Uniform(low, high)
if size is not None:
size = size if isinstance(size, tuple) else (size,)
sample = uniform.sample(size)
else:
sample = uniform.sample()
if dtype is not None:
sample = sample.to(dtype)
return to_backend(sample)
def normal(loc=0, scale=1.0, size=None):
size = size if size is not None else (1,)
size = size if isinstance(size, tuple) else (size,)
return to_backend(torch.normal(mean=loc, std=scale, size=size))
def random_sample(*args, **kwargs):
sample = torch.rand(*args, **kwargs)
return to_backend(sample)
def permutation(x):
idx = torch.randperm(x.shape[0])
sample = x[idx].to(Backend.get_device()).detach()
return to_backend(sample)