def add_summary_by_name(summary_name: str,
summary_value: tf.Tensor,
max_outputs_tb: int = 1):
"""
Add the summary defining the type of it by name and subtracting
the prefix from name
Parameters
----------
summary_name
name of the summary
summary_value :
value of summary
max_outputs_tb
number of maximum outputs in tensorboard e.g. for images
"""
name_splitted = summary_name.split('/')
if len(name_splitted) > 1:
family = name_splitted[0]
else:
family = None
if 'scalar_' in summary_name:
tf.summary.scalar(summary_name.replace('scalar_', ''),
summary_value,
family=family)
elif 'image_' in summary_name:
tf.summary.image(summary_name.replace('image_', ''),
summary_value,
max_outputs=max_outputs_tb,
family=family)
elif 'histogram_' in summary_name:
main_name = summary_name.replace('histogram_', '')
if isinstance(summary_value, dict):
for each_summary_name, each_summary_value in summary_value.items():
histogram_name = '_'.join([main_name, each_summary_name])
add_histogram_summary(histogram_name, each_summary_value)
else:
add_histogram_summary(main_name, summary_value)
elif 'text_' in summary_name:
tf.summary.text(summary_name.replace('text_', ''), summary_value)
elif 'audio_' in summary_name:
# TODO([email protected]) Find a way to set the sample_rate
tf.summary.audio(summary_name.replace('audio_', ''),
summary_value,
max_outputs=max_outputs_tb,
family=family,
sample_rate=16000)
else:
msg = ('Warning: summary with name {} will not be added '
'to tensorboard!'.format(summary_name))
warnings.warn(msg, RuntimeWarning, stacklevel=2)
def __call__(self, values: Tensor, raveled: bool = True) -> Tensor:
if isinstance(values, dict):
return {k: self(v, raveled) for k, v in values.items()}
return self._call_with_tensor_values(values=values, raveled=raveled)
