def to_torch(x: Union[Batch, dict, list, tuple, np.ndarray, torch.Tensor],
dtype: Optional[torch.dtype] = None,
device: Union[str, int, torch.device] = 'cpu'
) -> Union[Batch, dict, list, tuple, np.ndarray, torch.Tensor]:
"""Return an object without np.ndarray."""
if isinstance(x, torch.Tensor):
if dtype is not None:
x = x.type(dtype)
x = x.to(device)
elif isinstance(x, dict):
for k, v in x.items():
x[k] = to_torch(v, dtype, device)
elif isinstance(x, Batch):
x.to_torch(dtype, device)
elif isinstance(x, (np.number, np.bool_, Number)):
x = to_torch(np.asanyarray(x), dtype, device)
elif isinstance(x, (list, tuple)):
try:
x = to_torch(_parse_value(x), dtype, device)
except TypeError:
x = [to_torch(e, dtype, device) for e in x]
else: # fallback
x = np.asanyarray(x)
if issubclass(x.dtype.type, (np.bool_, np.number)):
x = torch.from_numpy(x).to(device)
if dtype is not None:
x = x.type(dtype)
else:
raise TypeError(f"object {x} cannot be converted to torch.")
return x
def to_torch(
x: Any,
dtype: Optional[torch.dtype] = None,
device: Union[str, int, torch.device] = "cpu",
) -> Union[Batch, torch.Tensor]:
"""Return an object without np.ndarray."""
if isinstance(x, np.ndarray) and issubclass(
x.dtype.type, (np.bool_, np.number)): # most often case
x = torch.from_numpy(x).to(device) # type: ignore
if dtype is not None:
x = x.type(dtype)
return x
elif isinstance(x, torch.Tensor): # second often case
if dtype is not None:
x = x.type(dtype)
return x.to(device) # type: ignore
elif isinstance(x, (np.number, np.bool_, Number)):
return to_torch(np.asanyarray(x), dtype, device)
elif isinstance(x, (dict, Batch)):
x = Batch(x, copy=True) if isinstance(x, dict) else deepcopy(x)
x.to_torch(dtype, device)
return x
elif isinstance(x, (list, tuple)):
return to_torch(_parse_value(x), dtype, device)
else: # fallback
raise TypeError(f"object {x} cannot be converted to torch.")
def to_numpy(
x: Union[Batch, dict, list, tuple, np.ndarray, torch.Tensor]
) -> Union[Batch, dict, list, tuple, np.ndarray, torch.Tensor]:
"""Return an object without torch.Tensor."""
if isinstance(x, torch.Tensor): # most often case
x = x.detach().cpu().numpy()
elif isinstance(x, np.ndarray): # second often case
pass
elif isinstance(x, (np.number, np.bool_, Number)):
x = np.asanyarray(x)
elif x is None:
x = np.array(None, dtype=np.object)
elif isinstance(x, Batch):
x.to_numpy()
elif isinstance(x, dict):
for k, v in x.items():
x[k] = to_numpy(v)
elif isinstance(x, (list, tuple)):
try:
x = to_numpy(_parse_value(x))
except TypeError:
x = [to_numpy(e) for e in x]
else: # fallback
x = np.asanyarray(x)
return x
def to_numpy(x: Any) -> Union[Batch, np.ndarray]:
"""Return an object without torch.Tensor."""
if isinstance(x, torch.Tensor): # most often case
return x.detach().cpu().numpy()
elif isinstance(x, np.ndarray): # second often case
return x
elif isinstance(x, (np.number, np.bool_, Number)):
return np.asanyarray(x)
elif x is None:
return np.array(None, dtype=object)
elif isinstance(x, (dict, Batch)):
x = Batch(x) if isinstance(x, dict) else deepcopy(x)
x.to_numpy()
return x
elif isinstance(x, (list, tuple)):
return to_numpy(_parse_value(x))
else: # fallback
return np.asanyarray(x)
def to_numpy(x: Union[
Batch, dict, list, tuple, np.ndarray, torch.Tensor]) -> Union[
Batch, dict, list, tuple, np.ndarray, torch.Tensor]:
"""Return an object without torch.Tensor."""
if isinstance(x, torch.Tensor):
x = x.detach().cpu().numpy()
elif isinstance(x, dict):
for k, v in x.items():
x[k] = to_numpy(v)
elif isinstance(x, Batch):
x.to_numpy()
elif isinstance(x, (list, tuple)):
try:
x = to_numpy(_parse_value(x))
except TypeError:
x = [to_numpy(e) for e in x]
else: # fallback
x = np.asanyarray(x)
return x