from chainer.training import extension
[docs]def observe_value(observation_key, target_func):
"""Returns a trainer extension to continuously record a value.
Args:
observation_key (str): Key of observation to record.
target_func (function): Function that returns the value to record.
It must take one argument: :class:~chainer.training.Trainer object.
Returns:
The extension function.
"""
@extension.make_extension(
trigger=(1, 'epoch'), priority=extension.PRIORITY_WRITER)
def _observe_value(trainer):
trainer.observation[observation_key] = target_func(trainer)
return _observe_value
[docs]def observe_lr(optimizer_name='main', observation_key='lr'):
"""Returns a trainer extension to record the learning rate.
Args:
optimizer_name (str): Name of optimizer whose learning rate is
recorded.
observation_key (str): Key of observation to record.
Returns:
The extension function.
"""
return observe_value(
observation_key,
lambda trainer: trainer.updater.get_optimizer(optimizer_name).lr)