def __init__(self, batch_size, every_n_steps=1, metric_logger=None):
self._last_recorded_time = time.time()
self._train_start_time = time.time()
self._batch_size = batch_size
self._every_n_steps = every_n_steps
self._logger = metric_logger or logger.BaseBenchmarkLogger()
self._global_step = 0 # Initialize it in __init__
super(ExamplesPerSecondCallback, self).__init__()
def test_log_metric(self):
log = logger.BaseBenchmarkLogger()
log.log_metric("accuracy",
0.999,
global_step=1e4,
extras={"name": "value"})
expected_log_prefix = "Benchmark metric:"
self.assertRegexpMatches(str(self.logged_message), expected_log_prefix)
def __init__(self,
batch_size,
every_n_steps=None,
every_n_secs=None,
warm_steps=0,
metric_logger=None):
"""Initializer for ExamplesPerSecondHook.
Args:
batch_size: Total batch size across all workers used to calculate
examples/second from global time.
every_n_steps: Log stats every n steps.
every_n_secs: Log stats every n seconds. Exactly one of the
`every_n_steps` or `every_n_secs` should be set.
warm_steps: The number of steps to be skipped before logging and running
average calculation. warm_steps steps refers to global steps across all
workers, not on each worker
metric_logger: instance of `BenchmarkLogger`, the benchmark logger that
hook should use to write the log. If None, BaseBenchmarkLogger will
be used.
Raises:
ValueError: if neither `every_n_steps` or `every_n_secs` is set, or
both are set.
"""
if (every_n_steps is None) == (every_n_secs is None):
raise ValueError("exactly one of every_n_steps"
" and every_n_secs should be provided.")
self._logger = metric_logger or logger.BaseBenchmarkLogger()
self._timer = tf.estimator.SecondOrStepTimer(every_steps=every_n_steps,
every_secs=every_n_secs)
self._step_train_time = 0
self._total_steps = 0
self._batch_size = batch_size
self._warm_steps = warm_steps
# List of examples per second logged every_n_steps.
self.current_examples_per_sec_list = []
def __init__(self, metric_logger=None):
self._logger = metric_logger or logger.BaseBenchmarkLogger()
self._per_batch_metrics = _PER_BATCH_METRICS
self._per_epoch_metrics = _PER_EPOCH_METRICS
self._global_step = 0 # Initialize it in __init__
super(LoggingMetricCallback, self).__init__()
