Python tensorflow.Event() Examples
The following are 20
code examples of tensorflow.Event().
You can vote up the ones you like or vote down the ones you don't like,
and go to the original project or source file by following the links above each example.
You may also want to check out all available functions/classes of the module
tensorflow
, or try the search function
.
![](https://www.programcreek.com/common/static/images/search.png)
Example #1
Source File: prof.py From tensorpack with Apache License 2.0 | 6 votes |
def _setup_graph(self): # special heuristics for Horovod from ..train import HorovodTrainer if isinstance(self.trainer, HorovodTrainer): if self.trainer.mpi_enabled(): logger.warn("GPUUtilizationTracker is disabled under MPI.") self._enabled = False return else: self._devices = [self.trainer.hvd.local_rank()] if self._devices is None: self._devices = self._guess_devices() assert len(self._devices), "[GPUUtilizationTracker] No GPU device given!" self._evt = mp.Event() self._stop_evt = mp.Event() self._queue = mp.Queue() self._proc = mp.Process(target=self.worker, args=( self._evt, self._queue, self._stop_evt, self._devices)) ensure_proc_terminate(self._proc) start_proc_mask_signal(self._proc)
Example #2
Source File: debugger_server_lib.py From Serverless-Deep-Learning-with-TensorFlow-and-AWS-Lambda with MIT License | 6 votes |
def on_graph_def(self, graph_def, device_name, wall_time): """Implementation of the GraphDef-carrying Event proto callback. Args: graph_def: A GraphDef proto. N.B.: The GraphDef is from the core runtime of a debugged Session::Run() call, after graph partition. Therefore it may differ from the GraphDef available to the general TensorBoard. For example, the GraphDef in general TensorBoard may get partitioned for multiple devices (CPUs and GPUs), each of which will generate a GraphDef event proto sent to this method. device_name: Name of the device on which the graph was created. wall_time: An epoch timestamp (in microseconds) for the graph. """ # For now, we do nothing with the graph def. However, we must define this # method to satisfy the handler's interface. Furthermore, we may use the # graph in the future (for instance to provide a graph if there is no graph # provided otherwise). del device_name del wall_time del graph_def
Example #3
Source File: prof.py From ADL with MIT License | 6 votes |
def _setup_graph(self): # special heuristics for Horovod from ..train import HorovodTrainer if isinstance(self.trainer, HorovodTrainer): if self.trainer.mpi_enabled(): logger.warn("GPUUtilizationTracker is disabled under MPI.") self._enabled = False return else: self._devices = [self.trainer.hvd.local_rank()] if self._devices is None: self._devices = self._guess_devices() assert len(self._devices), "[GPUUtilizationTracker] No GPU device given!" self._evt = mp.Event() self._stop_evt = mp.Event() self._queue = mp.Queue() self._proc = mp.Process(target=self.worker, args=( self._evt, self._queue, self._stop_evt, self._devices)) ensure_proc_terminate(self._proc) start_proc_mask_signal(self._proc)
Example #4
Source File: event_file_loader.py From Serverless-Deep-Learning-with-TensorFlow-and-AWS-Lambda with MIT License | 6 votes |
def Load(self): """Loads all new values from disk. Calling Load multiple times in a row will not 'drop' events as long as the return value is not iterated over. Yields: All values that were written to disk that have not been yielded yet. """ tf.logging.debug('Loading events from %s', self._file_path) while True: try: with tf.errors.raise_exception_on_not_ok_status() as status: self._reader.GetNext(status) except (tf.errors.DataLossError, tf.errors.OutOfRangeError): # We ignore partial read exceptions, because a record may be truncated. # PyRecordReader holds the offset prior to the failed read, so retrying # will succeed. break event = tf.Event() event.ParseFromString(self._reader.record()) yield event tf.logging.debug('No more events in %s', self._file_path)
Example #5
Source File: server.py From crayon with MIT License | 6 votes |
def tb_add_histogram(experiment, name, wall_time, step, histo): # Tensorflow does not support key being unicode histo_string = {} for k,v in histo.items(): histo_string[str(k)] = v histo = histo_string writer = tb_get_xp_writer(experiment) summary = tf.Summary(value=[ tf.Summary.Value(tag=name, histo=histo), ]) event = tf.Event(wall_time=wall_time, step=step, summary=summary) writer.add_event(event) writer.flush() tb_modified_xp(experiment, modified_type="histograms", wall_time=wall_time) # Perform requests to tensorboard http api
Example #6
Source File: interactive_debugger_server_lib.py From tensorboard with Apache License 2.0 | 6 votes |
def on_graph_def(self, graph_def, device_name, wall_time): """Implementation of the GraphDef-carrying Event proto callback. Args: graph_def: A GraphDef proto. N.B.: The GraphDef is from the core runtime of a debugged Session::Run() call, after graph partition. Therefore it may differ from the GraphDef available to the general TensorBoard. For example, the GraphDef in general TensorBoard may get partitioned for multiple devices (CPUs and GPUs), each of which will generate a GraphDef event proto sent to this method. device_name: Name of the device on which the graph was created. wall_time: An epoch timestamp (in microseconds) for the graph. """ # For now, we do nothing with the graph def. However, we must define this # method to satisfy the handler's interface. Furthermore, we may use the # graph in the future (for instance to provide a graph if there is no graph # provided otherwise). del wall_time self._graph_defs[device_name] = graph_def if not self._graph_defs_arrive_first: self._add_graph_def(device_name, graph_def) self._incoming_channel.get()
Example #7
Source File: prof.py From tensorpack with Apache License 2.0 | 5 votes |
def _write_event(self, metadata): evt = tf.Event() evt.tagged_run_metadata.tag = 'trace-{}'.format(self.global_step) evt.tagged_run_metadata.run_metadata = metadata.SerializeToString() self.trainer.monitors.put_event(evt)
Example #8
Source File: generate_testdata.py From deep_image_model with Apache License 2.0 | 5 votes |
def WriteScalarSeries(writer, tag, f, n=5): """Write a series of scalar events to writer, using f to create values.""" step = 0 wall_time = _start_time for i in xrange(n): v = f(i) value = tf.Summary.Value(tag=tag, simple_value=v) summary = tf.Summary(value=[value]) event = tf.Event(wall_time=wall_time, step=step, summary=summary) writer.add_event(event) step += 1 wall_time += 10
Example #9
Source File: debugger_server_lib.py From Serverless-Deep-Learning-with-TensorFlow-and-AWS-Lambda with MIT License | 5 votes |
def on_core_metadata_event(self, event): """Implementation of the core metadata-carrying Event proto callback. Args: event: An Event proto that contains core metadata about the debugged Session::Run() in its log_message.message field, as a JSON string. See the doc string of debug_data.DebugDumpDir.core_metadata for details. """ self._session_run_index = self._parse_session_run_index(event)
Example #10
Source File: server.py From crayon with MIT License | 5 votes |
def tb_add_scalar(experiment, name, wall_time, step, value): writer = tb_get_xp_writer(experiment) summary = tf.Summary(value=[ tf.Summary.Value(tag=name, simple_value=value), ]) event = tf.Event(wall_time=wall_time, step=step, summary=summary) writer.add_event(event) writer.flush() tb_modified_xp(experiment, modified_type="scalars", wall_time=wall_time)
Example #11
Source File: debugger_plugin_testlib.py From tensorboard with Apache License 2.0 | 5 votes |
def _CreateEventWithDebugNumericSummary( self, device_name, op_name, output_slot, wall_time, step, list_of_values ): """Creates event with a health pill summary. Note the debugger plugin only works with TensorFlow and, thus, uses TF protos and TF EventsWriter. Args: device_name: The name of the op's device. op_name: The name of the op to which a DebugNumericSummary was attached. output_slot: The numeric output slot for the tensor. wall_time: The numeric wall time of the event. step: The step of the event. list_of_values: A python list of values within the tensor. Returns: A `tf.Event` with a health pill summary. """ event = tf.compat.v1.Event(step=step, wall_time=wall_time) tensor = tf.compat.v1.make_tensor_proto( list_of_values, dtype=tf.float64, shape=[len(list_of_values)] ) value = event.summary.value.add( tag=op_name, node_name="%s:%d:DebugNumericSummary" % (op_name, output_slot), tensor=tensor, ) content_proto = debugger_event_metadata_pb2.DebuggerEventMetadata( device=device_name, output_slot=output_slot ) value.metadata.plugin_data.plugin_name = constants.DEBUGGER_PLUGIN_NAME value.metadata.plugin_data.content = tf.compat.as_bytes( json_format.MessageToJson( content_proto, including_default_value_fields=True ) ) return event
Example #12
Source File: interactive_debugger_server_lib.py From tensorboard with Apache License 2.0 | 5 votes |
def on_core_metadata_event(self, event): """Implementation of the core metadata-carrying Event proto callback. Args: event: An Event proto that contains core metadata about the debugged Session::Run() in its log_message.message field, as a JSON string. See the doc string of debug_data.DebugDumpDir.core_metadata for details. """ core_metadata = json.loads(event.log_message.message) input_names = ",".join(core_metadata["input_names"]) output_names = ",".join(core_metadata["output_names"]) target_nodes = ",".join(core_metadata["target_nodes"]) self._run_key = RunKey(input_names, output_names, target_nodes) if not self._graph_defs: self._graph_defs_arrive_first = False else: for device_name in self._graph_defs: self._add_graph_def(device_name, self._graph_defs[device_name]) self._outgoing_channel.put( _comm_metadata(self._run_key, event.wall_time) ) # Wait for acknowledgement from client. Blocks until an item is got. logger.info("on_core_metadata_event() waiting for client ack (meta)...") self._incoming_channel.get() logger.info("on_core_metadata_event() client ack received (meta).") # TODO(cais): If eager mode, this should return something to yield.
Example #13
Source File: interactive_debugger_server_lib.py From tensorboard with Apache License 2.0 | 5 votes |
def _extract_device_name_from_event(event): """Extract device name from a tf.Event proto carrying tensor value.""" plugin_data_content = json.loads( tf.compat.as_str(event.summary.value[0].metadata.plugin_data.content) ) return plugin_data_content["device"]
Example #14
Source File: generate_testdata.py From tensorboard with Apache License 2.0 | 5 votes |
def WriteHistogramSeries(writer, tag, mu_sigma_tuples, n=20): """Write a sequence of normally distributed histograms to writer.""" step = 0 wall_time = _start_time for [mean, stddev] in mu_sigma_tuples: data = [random.normalvariate(mean, stddev) for _ in xrange(n)] histo = _MakeHistogram(data) summary = tf.Summary(value=[tf.Summary.Value(tag=tag, histo=histo)]) event = tf.Event(wall_time=wall_time, step=step, summary=summary) writer.add_event(event) step += 10 wall_time += 100
Example #15
Source File: generate_testdata.py From tensorboard with Apache License 2.0 | 5 votes |
def WriteScalarSeries(writer, tag, f, n=5): """Write a series of scalar events to writer, using f to create values.""" step = 0 wall_time = _start_time for i in xrange(n): v = f(i) value = tf.Summary.Value(tag=tag, simple_value=v) summary = tf.Summary(value=[value]) event = tf.Event(wall_time=wall_time, step=step, summary=summary) writer.add_event(event) step += 1 wall_time += 10
Example #16
Source File: prof.py From ADL with MIT License | 5 votes |
def _write_event(self, metadata): evt = tf.Event() evt.tagged_run_metadata.tag = 'trace-{}'.format(self.global_step) evt.tagged_run_metadata.run_metadata = metadata.SerializeToString() self.trainer.monitors.put_event(evt)
Example #17
Source File: prof.py From petridishnn with MIT License | 5 votes |
def _write_event(self, metadata): evt = tf.Event() evt.tagged_run_metadata.tag = 'trace-{}'.format(self.global_step) evt.tagged_run_metadata.run_metadata = metadata.SerializeToString() self.trainer.monitors.put_event(evt)
Example #18
Source File: prof.py From petridishnn with MIT License | 5 votes |
def _before_train(self): assert gpu_available_in_session(), "[GPUUtilizationTracker] needs GPU!" self._evt = mp.Event() self._stop_evt = mp.Event() self._queue = mp.Queue() self._proc = mp.Process(target=self.worker, args=( self._evt, self._queue, self._stop_evt)) ensure_proc_terminate(self._proc) start_proc_mask_signal(self._proc)
Example #19
Source File: generate_testdata.py From deep_image_model with Apache License 2.0 | 5 votes |
def WriteHistogramSeries(writer, tag, mu_sigma_tuples, n=20): """Write a sequence of normally distributed histograms to writer.""" step = 0 wall_time = _start_time for [mean, stddev] in mu_sigma_tuples: data = [random.normalvariate(mean, stddev) for _ in xrange(n)] histo = _MakeHistogram(data) summary = tf.Summary(value=[tf.Summary.Value(tag=tag, histo=histo)]) event = tf.Event(wall_time=wall_time, step=step, summary=summary) writer.add_event(event) step += 10 wall_time += 100
Example #20
Source File: debugger_server_lib.py From Serverless-Deep-Learning-with-TensorFlow-and-AWS-Lambda with MIT License | 4 votes |
def on_value_event(self, event): """Records the summary values based on an updated message from the debugger. Logs an error message if writing the event to disk fails. Args: event: The Event proto to be processed. """ if not event.summary.value: tf.logging.warn("The summary of the event lacks a value.") return # The node name property is actually a watch key, which is a concatenation # of several pieces of data. watch_key = event.summary.value[0].node_name if not watch_key.endswith(constants.DEBUG_NUMERIC_SUMMARY_SUFFIX): # Ignore events that lack a DebugNumericSummary. # NOTE(@chihuahua): We may later handle other types of debug ops. return # We remove the constants.DEBUG_NUMERIC_SUMMARY_SUFFIX from the end of the # watch name because it is not distinguishing: every health pill entry ends # with it. node_name_and_output_slot = watch_key[ :-len(constants.DEBUG_NUMERIC_SUMMARY_SUFFIX)] shape = tf.make_ndarray(event.summary.value[0].tensor).shape if (len(shape) != 1 or shape[0] < constants.MIN_DEBUG_NUMERIC_SUMMARY_TENSOR_LENGTH): tf.logging.warning("Health-pill tensor either lacks a dimension or is " "shaped incorrectly: %s" % shape) return match = re.match(r"^(.*):(\d+)$", node_name_and_output_slot) if not match: tf.logging.warning( ("A event with a health pill has an invalid node name and output " "slot combination, (i.e., an unexpected debug op): %r"), node_name_and_output_slot) return if self._session_run_index >= 0: event.step = self._session_run_index else: # Data from parameter servers (or any graphs without a master) do not # contain core metadata. So the session run count is missing. Set its # value to a microsecond epoch timestamp. event.step = int(time.time() * 1e6) # Write this event to the events file designated for data from the # debugger. self._events_writer_manager.write_event(event) alert = numerics_alert.extract_numerics_alert(event) if self._numerics_alert_callback and alert: self._numerics_alert_callback(alert)