File size: 8,458 Bytes
cf2a15a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
# Copyright 2019 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.
# ==============================================================================
"""Summary creation methods for the HParams plugin.

Typical usage for exporting summaries in a hyperparameters-tuning experiment:
1. Create the experiment (once) by calling experiment_pb() and exporting
   the resulting summary into a top-level (empty) run.
2. In each training session in the experiment, call session_start_pb() before
   the session starts, exporting the resulting summary into a uniquely named
   run for the session, say <session_name>.
3. Train the model in the session, exporting each metric as a scalar summary
   in runs of the form <session_name>/<sub_dir>, where <sub_dir> can be empty a
   (in which case the run is just the <session_name>) and depends on the
   metric. The name of such a metric is a (group, tag) pair given by
   (<sub_dir>, tag) where tag is the tag of the scalar summary.
   When calling experiment_pb in step 1, you'll need to pass all the metric
   names used in the experiemnt.
4. When the session completes, call session_end_pb() and export the resulting
   summary into the same session run <session_name>.
"""


import time


import tensorflow as tf

from tensorboard.plugins.hparams import api_pb2
from tensorboard.plugins.hparams import metadata
from tensorboard.plugins.hparams import plugin_data_pb2


def experiment_pb(
    hparam_infos, metric_infos, user="", description="", time_created_secs=None
):
    """Creates a summary that defines a hyperparameter-tuning experiment.

    Args:
      hparam_infos: Array of api_pb2.HParamInfo messages. Describes the
          hyperparameters used in the experiment.
      metric_infos: Array of api_pb2.MetricInfo messages. Describes the metrics
          used in the experiment. See the documentation at the top of this file
          for how to populate this.
      user: String. An id for the user running the experiment
      description: String. A description for the experiment. May contain markdown.
      time_created_secs: float. The time the experiment is created in seconds
      since the UNIX epoch. If None uses the current time.

    Returns:
      A summary protobuffer containing the experiment definition.
    """
    if time_created_secs is None:
        time_created_secs = time.time()
    experiment = api_pb2.Experiment(
        description=description,
        user=user,
        time_created_secs=time_created_secs,
        hparam_infos=hparam_infos,
        metric_infos=metric_infos,
    )
    return _summary(
        metadata.EXPERIMENT_TAG,
        plugin_data_pb2.HParamsPluginData(experiment=experiment),
    )


def session_start_pb(
    hparams, model_uri="", monitor_url="", group_name="", start_time_secs=None
):
    """Constructs a SessionStartInfo protobuffer.

    Creates a summary that contains a training session metadata information.
    One such summary per training session should be created. Each should have
    a different run.

    Args:
      hparams: A dictionary with string keys. Describes the hyperparameter values
               used in the session, mapping each hyperparameter name to its value.
               Supported value types are  `bool`, `int`, `float`, `str`, `list`,
               `tuple`.
               The type of value must correspond to the type of hyperparameter
               (defined in the corresponding api_pb2.HParamInfo member of the
               Experiment protobuf) as follows:

                +-----------------+---------------------------------+
                |Hyperparameter   | Allowed (Python) value types    |
                |type             |                                 |
                +-----------------+---------------------------------+
                |DATA_TYPE_BOOL   | bool                            |
                |DATA_TYPE_FLOAT64| int, float                      |
                |DATA_TYPE_STRING | str, tuple, list   |
                +-----------------+---------------------------------+

               Tuple and list instances will be converted to their string
               representation.
      model_uri: See the comment for the field with the same name of
                 plugin_data_pb2.SessionStartInfo.
      monitor_url: See the comment for the field with the same name of
                   plugin_data_pb2.SessionStartInfo.
      group_name:  See the comment for the field with the same name of
                   plugin_data_pb2.SessionStartInfo.
      start_time_secs: float. The time to use as the session start time.
                       Represented as seconds since the UNIX epoch. If None uses
                       the current time.
    Returns:
      The summary protobuffer mentioned above.
    """
    if start_time_secs is None:
        start_time_secs = time.time()
    session_start_info = plugin_data_pb2.SessionStartInfo(
        model_uri=model_uri,
        monitor_url=monitor_url,
        group_name=group_name,
        start_time_secs=start_time_secs,
    )
    for hp_name, hp_val in hparams.items():
        # Boolean typed values need to be checked before integers since in Python
        # isinstance(True/False, int) returns True.
        if isinstance(hp_val, bool):
            session_start_info.hparams[hp_name].bool_value = hp_val
        elif isinstance(hp_val, (float, int)):
            session_start_info.hparams[hp_name].number_value = hp_val
        elif isinstance(hp_val, str):
            session_start_info.hparams[hp_name].string_value = hp_val
        elif isinstance(hp_val, (list, tuple)):
            session_start_info.hparams[hp_name].string_value = str(hp_val)
        else:
            raise TypeError(
                "hparams[%s]=%s has type: %s which is not supported"
                % (hp_name, hp_val, type(hp_val))
            )
    return _summary(
        metadata.SESSION_START_INFO_TAG,
        plugin_data_pb2.HParamsPluginData(
            session_start_info=session_start_info
        ),
    )


def session_end_pb(status, end_time_secs=None):
    """Constructs a SessionEndInfo protobuffer.

    Creates a summary that contains status information for a completed
    training session. Should be exported after the training session is completed.
    One such summary per training session should be created. Each should have
    a different run.

    Args:
      status: A tensorboard.hparams.Status enumeration value denoting the
          status of the session.
      end_time_secs: float. The time to use as the session end time. Represented
          as seconds since the unix epoch. If None uses the current time.

    Returns:
      The summary protobuffer mentioned above.
    """
    if end_time_secs is None:
        end_time_secs = time.time()

    session_end_info = plugin_data_pb2.SessionEndInfo(
        status=status, end_time_secs=end_time_secs
    )
    return _summary(
        metadata.SESSION_END_INFO_TAG,
        plugin_data_pb2.HParamsPluginData(session_end_info=session_end_info),
    )


def _summary(tag, hparams_plugin_data):
    """Returns a summary holding the given HParamsPluginData message.

    Helper function.

    Args:
      tag: string. The tag to use.
      hparams_plugin_data: The HParamsPluginData message to use.
    """
    summary = tf.compat.v1.Summary()
    tb_metadata = metadata.create_summary_metadata(hparams_plugin_data)
    raw_metadata = tb_metadata.SerializeToString()
    tf_metadata = tf.compat.v1.SummaryMetadata.FromString(raw_metadata)
    summary.value.add(
        tag=tag,
        metadata=tf_metadata,
        tensor=_TF_NULL_TENSOR,
    )
    return summary


# Like `metadata.NULL_TENSOR`, but with the TensorFlow version of the
# proto. Slight kludge needed to expose the `TensorProto` type.
_TF_NULL_TENSOR = type(tf.make_tensor_proto(0)).FromString(
    metadata.NULL_TENSOR.SerializeToString()
)