本文整理汇总了Python中mxnet.nd.empty方法的典型用法代码示例。如果您正苦于以下问题:Python nd.empty方法的具体用法?Python nd.empty怎么用?Python nd.empty使用的例子?那么, 这里精选的方法代码示例或许可以为您提供帮助。您也可以进一步了解该方法所在类mxnet.nd
的用法示例。
在下文中一共展示了nd.empty方法的4个代码示例,这些例子默认根据受欢迎程度排序。您可以为喜欢或者感觉有用的代码点赞,您的评价将有助于系统推荐出更棒的Python代码示例。
示例1: default_mp_pad_batchify_fn
# 需要导入模块: from mxnet import nd [as 别名]
# 或者: from mxnet.nd import empty [as 别名]
def default_mp_pad_batchify_fn(data):
"""Use shared memory for collating data into batch, labels are padded to same shape"""
if isinstance(data[0], nd.NDArray):
out = nd.empty((len(data),) + data[0].shape, dtype=data[0].dtype,
ctx=context.Context('cpu_shared', 0))
return nd.stack(*data, out=out)
elif isinstance(data[0], tuple):
data = zip(*data)
return [default_mp_pad_batchify_fn(i) for i in data]
else:
data = np.asarray(data)
batch_size = len(data)
pad = max([l.shape[0] for l in data] + [1,])
buf = np.full((batch_size, pad, data[0].shape[-1]), -1, dtype=data[0].dtype)
for i, l in enumerate(data):
buf[i][:l.shape[0], :] = l
return nd.array(buf, dtype=data[0].dtype, ctx=context.Context('cpu_shared', 0))
示例2: __next__
# 需要导入模块: from mxnet import nd [as 别名]
# 或者: from mxnet.nd import empty [as 别名]
def __next__(self) -> DataEntry:
# if the buffer is empty, fill the buffer first.
# (should only executed in the first round)
if not self.shuffle_buffer:
self.shuffle_buffer = list(
itertools.islice(
self.base_iterator, self.shuffle_buffer_length
)
)
# if buffer still empty, means all elements used,
# return a signal of end of iterator
if not self.shuffle_buffer:
raise StopIteration
# choose an element at a random index and yield it
# and fill it with the next element in the sequential generator
idx = random.randint(0, len(self.shuffle_buffer) - 1)
next_sample = self.shuffle_buffer[idx]
# replace the index with the next element in the iterator if the iterator has not finished.
# delete the index otherwise.
try:
self.shuffle_buffer[idx] = next(self.base_iterator)
except StopIteration:
del self.shuffle_buffer[idx]
return next_sample
示例3: clip_grad
# 需要导入模块: from mxnet import nd [as 别名]
# 或者: from mxnet.nd import empty [as 别名]
def clip_grad(
grads: Union[Generator[NDArray, NDArray, NDArray], List[NDArray], Tuple[NDArray]],
clip_method: GradientClippingMethod,
clip_val: float,
inplace=True) -> List[NDArray]:
"""
Clip gradient values inplace
:param grads: gradients to be clipped
:param clip_method: clipping method
:param clip_val: clipping value. Interpreted differently depending on clipping method.
:param inplace: modify grads if True, otherwise create NDArrays
:return: clipped gradients
"""
output = list(grads) if inplace else list(nd.empty(g.shape) for g in grads)
if clip_method == GradientClippingMethod.ClipByGlobalNorm:
norm_unclipped_grads = global_norm(grads)
scale = clip_val / (norm_unclipped_grads.asscalar() + 1e-8) # todo: use branching operators?
if scale < 1.0:
for g, o in zip(grads, output):
nd.broadcast_mul(g, nd.array([scale]), out=o)
elif clip_method == GradientClippingMethod.ClipByValue:
for g, o in zip(grads, output):
g.clip(-clip_val, clip_val, out=o)
elif clip_method == GradientClippingMethod.ClipByNorm:
for g, o in zip(grads, output):
nd.broadcast_mul(g, nd.minimum(1.0, clip_val / (g.norm() + 1e-8)), out=o)
else:
raise KeyError('Unsupported gradient clipping method')
return output
示例4: _worker_fn
# 需要导入模块: from mxnet import nd [as 别名]
# 或者: from mxnet.nd import empty [as 别名]
def _worker_fn(
batch_size: int,
batchify_fn: Callable,
dtype: DType,
is_train: bool,
cyclic: bool,
cycle_num: int,
shuffle_buffer_length: int,
):
"""Function for processing data in worker process."""
# initialize, or reset the iterator at each cycle
if (_WorkerData.iterator_latest_reset_cycle < cycle_num) and (
_WorkerData.iterator_latest_reset_cycle == 0 or not cyclic
):
_worker_reset_iterator(
is_train, cyclic, cycle_num, shuffle_buffer_length
)
# retrieve the samples that will be batched
batch_samples = list(
itertools.islice(_WorkerData.dataset_iterator, batch_size)
)
# batch the samples, if there were any
if batch_samples:
success = True
batch = batchify_fn(
data=batch_samples, dtype=dtype, multi_processing=True
)
else:
# the second time without being able to provide a batch we want to delay calling them again
# on fist exhaustion they should not be delayed, since they need to indicate depletion
# dont make the penalty to high, since that delays rescheduling of non empty iterators
if _WorkerData.iterator_exhausted_indicator:
time.sleep(0.05)
else:
_WorkerData.iterator_exhausted_indicator = True
success = False
batch = None
buf = io.BytesIO()
ForkingPickler(buf, pickle.HIGHEST_PROTOCOL).dump(
(success, MPWorkerInfo.worker_id, batch)
)
return buf.getvalue()