1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
|
# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Iteration over tf.data.Datasets when eager execution is enabled."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import threading
from tensorflow.contrib.data.python.ops import prefetching_ops
from tensorflow.python.data.ops import iterator_ops
from tensorflow.python.data.util import nest
from tensorflow.python.eager import context
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import errors
from tensorflow.python.framework import function
from tensorflow.python.framework import ops
from tensorflow.python.ops import gen_dataset_ops
from tensorflow.python.ops import resource_variable_ops
_uid_counter = 0
_uid_lock = threading.Lock()
def _generate_shared_name(prefix):
with _uid_lock:
global _uid_counter
uid = _uid_counter
_uid_counter += 1
return "{}_{}".format(prefix, uid)
class Iterator(object):
"""An iterator producing tf.Tensor objects from a tf.data.Dataset."""
def __init__(self, dataset):
"""Creates a new iterator over the given dataset.
For example:
```python
dataset = tf.data.Dataset.range(4)
for x in Iterator(dataset):
print(x)
```
Tensors produced will be placed on the device on which this iterator object
was created.
Args:
dataset: A `tf.data.Dataset` object.
Raises:
RuntimeError: When invoked without eager execution enabled.
"""
if not context.in_eager_mode():
raise RuntimeError(
"{} objects can only be used when eager execution is enabled, use "
"tf.data.Dataset.make_iterator or "
"tf.data.Dataset.make_one_shot_iterator for graph construction".
format(type(self)))
with ops.device("/device:CPU:0"):
ds_variant = dataset._as_variant_tensor() # pylint: disable=protected-access
self._output_types = dataset.output_types
self._output_shapes = dataset.output_shapes
self._flat_output_types = nest.flatten(dataset.output_types)
self._flat_output_shapes = nest.flatten(dataset.output_shapes)
self._resource = gen_dataset_ops.iterator(
container="",
shared_name=_generate_shared_name("eager_iterator"),
output_types=self._flat_output_types,
output_shapes=self._flat_output_shapes)
gen_dataset_ops.make_iterator(ds_variant, self._resource)
# Delete the resource when this object is deleted
self._resource_deleter = resource_variable_ops.EagerResourceDeleter(
handle=self._resource, handle_device="/device:CPU:0")
self._device = context.context().device_name
self._buffer_resource_handle = None
if not context.context().device_spec.device_type:
is_remote_device = False
else:
is_remote_device = context.context().device_spec.device_type != "CPU"
if is_remote_device:
with ops.device("/device:CPU:0"):
iter_string_handle = gen_dataset_ops.iterator_to_string_handle(
self._resource)
@function.Defun(dtypes.string)
def remote_fn(h):
remote_iterator = iterator_ops.Iterator.from_string_handle(
h, self._output_types, self._output_shapes)
return remote_iterator.get_next()
remote_fn.add_to_graph(None)
target = constant_op.constant("/device:CPU:0")
with ops.device(self._device):
self._buffer_resource_handle = prefetching_ops.function_buffering_resource(
string_arg=iter_string_handle,
f=remote_fn,
target_device=target,
buffer_size=10,
thread_pool_size=1,
container="",
shared_name=_generate_shared_name("function_buffer_resource"))
self._buffer_resource_deleter = resource_variable_ops.EagerResourceDeleter(
handle=self._buffer_resource_handle, handle_device=self._device)
def __iter__(self):
return self
def __next__(self): # For Python 3 compatibility
return self.next()
def next(self):
"""Return the next tf.Tensor from the dataset."""
with ops.device(self._device):
try:
if self._buffer_resource_handle is not None:
ret = prefetching_ops.function_buffering_resource_get_next(
function_buffer_resource=self._buffer_resource_handle,
output_types=self._flat_output_types)
else:
# TODO(ashankar): Consider removing this ops.device() contextmanager
# and instead mimic ops placement in graphs: Operations on resource
# handles execute on the same device as where the resource is placed.
ret = gen_dataset_ops.iterator_get_next(
self._resource,
output_types=self._flat_output_types,
output_shapes=self._flat_output_shapes)
except errors.OutOfRangeError:
raise StopIteration
return nest.pack_sequence_as(self._output_types, ret)
|
