本文整理汇总了Python中tensorflow.python.framework.tensor_util.TensorShapeProtoToList方法的典型用法代码示例。如果您正苦于以下问题:Python tensor_util.TensorShapeProtoToList方法的具体用法?Python tensor_util.TensorShapeProtoToList怎么用?Python tensor_util.TensorShapeProtoToList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tensorflow.python.framework.tensor_util
的用法示例。
在下文中一共展示了tensor_util.TensorShapeProtoToList方法的6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: quantize_weight_rounded
# 需要导入模块: from tensorflow.python.framework import tensor_util [as 别名]
# 或者: from tensorflow.python.framework.tensor_util import TensorShapeProtoToList [as 别名]
def quantize_weight_rounded(input_node):
"""Returns a replacement node for input_node containing bucketed floats."""
input_tensor = input_node.attr["value"].tensor
tensor_value = tensor_util.MakeNdarray(input_tensor)
shape = input_tensor.tensor_shape
# Currently, the parameter FLAGS.bitdepth is used to compute the
# number of buckets as 1 << FLAGS.bitdepth, meaning the number of
# buckets can only be a power of 2.
# This could be fixed by introducing a new parameter, num_buckets,
# which would allow for more flexibility in chosing the right model
# size/accuracy tradeoff. But I didn't want to add more parameters
# to this script than absolutely necessary.
num_buckets = 1 << FLAGS.bitdepth
tensor_value_rounded = quantize_array(tensor_value, num_buckets)
tensor_shape_list = tensor_util.TensorShapeProtoToList(shape)
return [create_constant_node(input_node.name, tensor_value_rounded,
tf.float32, shape=tensor_shape_list)]
示例2: quantize_weight_rounded
# 需要导入模块: from tensorflow.python.framework import tensor_util [as 别名]
# 或者: from tensorflow.python.framework.tensor_util import TensorShapeProtoToList [as 别名]
def quantize_weight_rounded(input_node):
"""Returns a replacement node for input_node containing bucketed floats."""
input_tensor = input_node.attr["value"].tensor
tensor_value = tensor_util.MakeNdarray(input_tensor)
shape = input_tensor.tensor_shape
# Currently, the parameter FLAGS.bitdepth is used to compute the
# number of buckets as 1 << FLAGS.bitdepth, meaning the number of
# buckets can only be a power of 2.
# This could be fixed by introducing a new parameter, num_buckets,
# which would allow for more flexibility in chosing the right model
# size/accuracy tradeoff. But I didn't want to add more parameters
# to this script than absolutely necessary.
num_buckets = 1 << FLAGS.bitdepth
tensor_value_rounded = quantize_array(tensor_value, num_buckets)
tensor_shape_list = tensor_util.TensorShapeProtoToList(shape)
return [
create_constant_node(
input_node.name,
tensor_value_rounded,
dtypes.float32,
shape=tensor_shape_list)
]
示例3: _parse_param
# 需要导入模块: from tensorflow.python.framework import tensor_util [as 别名]
# 或者: from tensorflow.python.framework.tensor_util import TensorShapeProtoToList [as 别名]
def _parse_param(self, key, value, name, shape):
try:
from tensorflow.python.framework import tensor_util
except ImportError as e:
raise ImportError(
"Unable to import tensorflow which is required {}".format(e))
if key == 'value':
np_array = tensor_util.MakeNdarray(value.tensor)
if np_array.dtype == np.dtype(object):
# Object types are generally tensorflow DT_STRING (DecodeJpeg op).
# Just leave it as placeholder.
if shape and name in shape:
var_shape = shape[name]
else:
var_shape = tensor_util.TensorShapeProtoToList(value.tensor.tensor_shape)
self._nodes[name] = [_expr.var(name, shape=var_shape, dtype='uint8')]
return
array_ndim = len(np_array.shape)
if array_ndim == 0:
self._nodes[name] = [tvm.relay.const(np_array)]
else:
self._params[name] = tvm.nd.array(np_array)
self._nodes[name] = [_expr.var(name,
shape=self._params[name].shape,
dtype=self._params[name].dtype)]
else:
if key not in ('dtype', '_output_shapes', '_class'):
raise NotImplementedError \
("Other attributes for a Const(param) Node {} ? .".format(key))
示例4: load_tensor_from_event
# 需要导入模块: from tensorflow.python.framework import tensor_util [as 别名]
# 或者: from tensorflow.python.framework.tensor_util import TensorShapeProtoToList [as 别名]
def load_tensor_from_event(event):
"""Load a tensor from an Event proto.
Args:
event: The Event proto, assumed to hold a tensor value in its
summary.value[0] field.
Returns:
The tensor value loaded from the event file, as a `numpy.ndarray`, if
representation of the tensor value by a `numpy.ndarray` is possible.
For uninitialized Tensors, returns `None`. For Tensors of data types that
cannot be represented as `numpy.ndarray` (e.g., `tf.resource`), return
the `TensorProto` protobuf object without converting it to a
`numpy.ndarray`.
"""
tensor_proto = event.summary.value[0].tensor
shape = tensor_util.TensorShapeProtoToList(tensor_proto.tensor_shape)
num_elements = 1
for shape_dim in shape:
num_elements *= shape_dim
if tensor_proto.tensor_content or tensor_proto.string_val or not num_elements:
# Initialized tensor or empty tensor.
if tensor_proto.dtype == types_pb2.DT_RESOURCE:
tensor_value = InconvertibleTensorProto(tensor_proto)
else:
try:
tensor_value = tensor_util.MakeNdarray(tensor_proto)
except KeyError:
tensor_value = InconvertibleTensorProto(tensor_proto)
else:
# Uninitialized tensor or tensor of unconvertible data type.
tensor_value = InconvertibleTensorProto(tensor_proto, False)
return tensor_value
开发者ID:PacktPublishing,项目名称:Serverless-Deep-Learning-with-TensorFlow-and-AWS-Lambda,代码行数:38,代码来源:debug_data.py
示例5: quantize_weight_eightbit
# 需要导入模块: from tensorflow.python.framework import tensor_util [as 别名]
# 或者: from tensorflow.python.framework.tensor_util import TensorShapeProtoToList [as 别名]
def quantize_weight_eightbit(input_node, quantization_mode):
"""Returns replacement nodes for input_node using the Dequantize op."""
base_name = input_node.name + "_"
quint8_const_name = base_name + "quint8_const"
min_name = base_name + "min"
max_name = base_name + "max"
float_tensor = tensor_util.MakeNdarray(
input_node.attr["value"].tensor)
min_value = np.min(float_tensor.flatten())
max_value = np.max(float_tensor.flatten())
# min_value == max_value is a tricky case. It can occur for general
# tensors, and of course for scalars. The quantized ops cannot deal
# with this case, so we set max_value to something else.
# It's a tricky question what is the numerically best solution to
# deal with this degeneracy.
# TODO(petewarden): Better use a tolerance than a hard comparison?
if min_value == max_value:
if abs(min_value) < 0.000001:
max_value = min_value + 1.0
elif min_value > 0:
max_value = 2 * min_value
else:
max_value = min_value / 2.0
sess = tf.Session()
with sess.as_default():
quantize_op = tf.contrib.quantization.python.quantize_v2(
float_tensor,
min_value,
max_value,
tf.quint8,
mode=quantization_mode)
quint8_tensor = quantize_op[0].eval()
shape = tensor_util.TensorShapeProtoToList(input_node.attr[
"value"].tensor.tensor_shape)
quint8_const_node = create_constant_node(quint8_const_name,
quint8_tensor,
tf.quint8,
shape=shape)
min_node = create_constant_node(min_name, min_value, tf.float32)
max_node = create_constant_node(max_name, max_value, tf.float32)
dequantize_node = create_node("Dequantize", input_node.name,
[quint8_const_name, min_name, max_name])
set_attr_dtype(dequantize_node, "T", tf.quint8)
set_attr_string(dequantize_node, "mode", quantization_mode)
return [quint8_const_node, min_node, max_node, dequantize_node]
示例6: quantize_weight_eightbit
# 需要导入模块: from tensorflow.python.framework import tensor_util [as 别名]
# 或者: from tensorflow.python.framework.tensor_util import TensorShapeProtoToList [as 别名]
def quantize_weight_eightbit(input_node, quantization_mode):
"""Returns replacement nodes for input_node using the Dequantize op."""
base_name = input_node.name + "_"
quint8_const_name = base_name + "quint8_const"
min_name = base_name + "min"
max_name = base_name + "max"
float_tensor = tensor_util.MakeNdarray(input_node.attr["value"].tensor)
min_value = np.min(float_tensor.flatten())
max_value = np.max(float_tensor.flatten())
# Make sure that the range includes zero.
if min_value > 0.0:
min_value = 0.0
# min_value == max_value is a tricky case. It can occur for general
# tensors, and of course for scalars. The quantized ops cannot deal
# with this case, so we set max_value to something else.
# It's a tricky question what is the numerically best solution to
# deal with this degeneracy.
# TODO(petewarden): Better use a tolerance than a hard comparison?
if min_value == max_value:
if abs(min_value) < 0.000001:
max_value = min_value + 1.0
elif min_value > 0:
max_value = 2 * min_value
else:
max_value = min_value / 2.0
sess = session.Session()
with sess.as_default():
quantize_op = array_ops.quantize_v2(
float_tensor,
min_value,
max_value,
dtypes.quint8,
mode=quantization_mode)
quint8_tensor = quantize_op[0].eval()
shape = tensor_util.TensorShapeProtoToList(input_node.attr["value"]
.tensor.tensor_shape)
quint8_const_node = create_constant_node(
quint8_const_name, quint8_tensor, dtypes.quint8, shape=shape)
min_node = create_constant_node(min_name, min_value, dtypes.float32)
max_node = create_constant_node(max_name, max_value, dtypes.float32)
dequantize_node = create_node("Dequantize", input_node.name,
[quint8_const_name, min_name, max_name])
set_attr_dtype(dequantize_node, "T", dtypes.quint8)
set_attr_string(dequantize_node, "mode", quantization_mode)
return [quint8_const_node, min_node, max_node, dequantize_node]