def __init__(self, writer, hparams, trial_id=None):
"""Create a callback for logging hyperparameters to TensorBoard.
As with the standard `tf.keras.callbacks.TensorBoard` class, each
callback object is valid for only one call to `model.fit`.
Args:
writer: The `SummaryWriter` object to which hparams should be
written, or a logdir (as a `str`) to be passed to
`tf.summary.create_file_writer` to create such a writer.
hparams: A `dict` mapping hyperparameters to the values used in
this session. Keys should be the names of `HParam` objects used
in an experiment, or the `HParam` objects themselves. Values
should be Python `bool`, `int`, `float`, or `string` values,
depending on the type of the hyperparameter.
trial_id: An optional `str` ID for the set of hyperparameter
values used in this trial. Defaults to a hash of the
hyperparameters.
Raises:
ValueError: If two entries in `hparams` share the same
hyperparameter name.
"""
# Defer creating the actual summary until we write it, so that the
# timestamp is correct. But create a "dry-run" first to fail fast in
# case the `hparams` are invalid.
self._hparams = dict(hparams)
self._trial_id = trial_id
summary_v2.hparams_pb(self._hparams, trial_id=self._trial_id)
if writer is None:
raise TypeError("writer must be a `SummaryWriter` or `str`, not None")
elif isinstance(writer, str):
self._writer = tf.compat.v2.summary.create_file_writer(writer)
else:
self._writer = writer
def test_duplicate_hparam_names_from_two_objects(self):
hparams = {
hp.HParam("foo"): 1,
hp.HParam("foo"): 1,
}
with six.assertRaisesRegex(
self, ValueError, "multiple values specified for hparam 'foo'"):
hp.hparams_pb(hparams)
def test_duplicate_hparam_names_across_object_and_string(self):
hparams = {
"foo": 1,
hp.HParam("foo"): 1,
}
with six.assertRaisesRegex(
self, ValueError, "multiple values specified for hparam 'foo'"):
hp.hparams_pb(hparams)
def test_serialize_tf_linspace_numpy(self):
# Should be subsumed by `test_serialize_numpy_scalars`; separate
# test because it's a common use case.
hparams = {
"f_default": tf.linspace(1.0, 2.0, 5).numpy()[0],
"f32": tf.cast(tf.linspace(1.0, 2.0, 5), tf.float32).numpy()[0],
"f64": tf.cast(tf.linspace(1.0, 2.0, 5), tf.float64).numpy()[0],
}
hp.hparams_pb(hparams)
def test_serialize_numpy_scalars(self):
hparams = {
"i32": np.array([1, 2], dtype=np.int32)[0],
"i64": np.array([1, 2], dtype=np.int64)[0],
"f_default": np.linspace(1.0, 2.0, 5)[0],
"f32": np.linspace(1.0, 2.0, 5, dtype=np.float32)[0],
"f64": np.linspace(1.0, 2.0, 5, dtype=np.float64)[0],
"bool": np.array([False, True])[0],
}
hp.hparams_pb(hparams)
def test_consistency_across_string_key_and_object_key(self):
hparams_1 = {
hp.HParam("optimizer", hp.Discrete(["adam", "sgd"])): "adam",
"learning_rate": 0.02,
}
hparams_2 = {
"optimizer": "adam",
hp.HParam("learning_rate", hp.RealInterval(1e-2, 1e-1)): 0.02,
}
self.assert_hparams_summaries_equal(
hp.hparams_pb(hparams_1, start_time_secs=self.start_time_secs),
hp.hparams_pb(hparams_2, start_time_secs=self.start_time_secs),
)
def test_invariant_under_permutation(self):
# In particular, the group name should be the same.
hparams_1 = {
"optimizer": "adam",
"learning_rate": 0.02,
}
hparams_2 = {
"learning_rate": 0.02,
"optimizer": "adam",
}
self.assert_hparams_summaries_equal(
hp.hparams_pb(hparams_1, start_time_secs=self.start_time_secs),
hp.hparams_pb(hparams_2, start_time_secs=self.start_time_secs),
)
def test_pb_explicit_trial_id(self):
result = hp.hparams_pb(
self.hparams,
trial_id=self.trial_id,
start_time_secs=self.start_time_secs,
)
self._check_summary(result, check_group_name=True)
def test_pb_is_tensorboard_copy_of_proto(self):
result = hp.hparams_pb(
self.hparams, start_time_secs=self.start_time_secs
)
self.assertIsInstance(result, summary_pb2.Summary)
if tf is not None:
self.assertNotIsInstance(result, tf.compat.v1.Summary)
def _HParamSessionStart(name, hparams):
from tensorboard.plugins.hparams import summary_v2 as hp
try:
# pylint: disable=unexpected-keyword-arg
return hp.hparams_pb(hparams, trial_id=name)
except TypeError:
return _legacy_hparams_pb(hparams, name)
def get_group_name(hparams):
summary_pb = hp.hparams_pb(hparams)
values = summary_pb.value
self.assertEqual(len(values), 1, values)
actual_value = values[0]
self.assertEqual(
actual_value.metadata.plugin_data.plugin_name,
metadata.PLUGIN_NAME,
)
plugin_content = actual_value.metadata.plugin_data.content
info = metadata.parse_session_start_info_plugin_data(plugin_content)
return info.group_name
def test_hparams(self):
old_event = event_pb2.Event()
old_event.step = 0
old_event.wall_time = 456.75
hparams_pb = hparams_summary.hparams_pb({"optimizer": "adam"})
# Simulate legacy event with no tensor content
for v in hparams_pb.value:
v.ClearField("tensor")
old_event.summary.CopyFrom(hparams_pb)
new_events = self._migrate_event(old_event)
self.assertLen(new_events, 1)
self.assertLen(new_events[0].summary.value, 1)
value = new_events[0].summary.value[0]
self.assertEqual(value.tensor, hparams_metadata.NULL_TENSOR)
self.assertEqual(value.metadata.data_class,
summary_pb2.DATA_CLASS_TENSOR)
self.assertEqual(
value.metadata.plugin_data,
hparams_pb.value[0].metadata.plugin_data,
)
def test_pb_invalid_trial_id(self):
with six.assertRaisesRegex(
self, TypeError, "`trial_id` should be a `str`, but got: 12"):
hp.hparams_pb(self.hparams, trial_id=12)
def test_pb_contents(self):
result = hp.hparams_pb(self.hparams,
start_time_secs=self.start_time_secs)
self._check_summary(result)
def Session(name, hparams):
try:
return hp.hparams_pb(hparams, trial_id=name)
except TypeError:
return LegacySession(name, hparams)
