"""
The decorator API is the core way to interact with metrics in torchbearer. All of the classes and functionality handled
here can be reproduced by manually interacting with the classes if necessary. Broadly speaking, the decorator API is
used to construct a :class:`.MetricFactory` which will build a :class:`.MetricTree` that handles data flow between
instances of :class:`.Mean`, :class:`.RunningMean`, :class:`.Std` etc.
"""
import inspect
from torchbearer import metrics
from torchbearer.metrics import EpochLambda, BatchLambda, ToDict, Mean, MetricTree, Std, RunningMean
[docs]def default_for_key(key, *args, **kwargs):
"""The :func:`default_for_key` decorator will register the given metric in the global metric dict
(`metrics.DEFAULT_METRICS`) so that it can be referenced by name in instances of :class:`.MetricList` such as in the
list given to the :class:`.torchbearer.Model`.
Example: ::
@default_for_key('acc')
class CategoricalAccuracy(metrics.BatchLambda):
...
:param key: The key to use when referencing the metric
:type key: str
:param args: Any args to pass to the underlying metric when constructed
:param kwargs: Any keyword args to pass to the underlying metric when constructed
"""
def decorator(arg):
metrics.add_default(key, arg, *args, **kwargs)
return arg
return decorator
[docs]def lambda_metric(name, on_epoch=False):
"""The :func:`lambda_metric` decorator is used to convert a lambda function `y_pred, y_true` into a :class:`.Metric`
instance. This can be used as in the following example: ::
@metrics.lambda_metric('my_metric')
def my_metric(y_pred, y_true):
... # Calculate some metric
model = Model(metrics=[my_metric])
:param name: The name of the metric (e.g. 'loss')
:param on_epoch: If True the metric will be an instance of :class:`.EpochLambda` instead of :class:`.BatchLambda`
:return: A decorator which replaces a function with a :class:`.Metric`
"""
def decorator(metric_function):
if on_epoch:
return EpochLambda(name, metric_function)
else:
return BatchLambda(name, metric_function)
return decorator
[docs]def to_dict(clazz):
"""The :func:`to_dict` decorator is used to wrap either a :class:`.Metric` class or a :class:`.Metric` instance with
a :class:`.ToDict` instance. The result is that future output will be wrapped in a `dict[name, value]`.
Example: ::
>>> from torchbearer import metrics
>>> @metrics.lambda_metric('my_metric')
... def my_metric(y_pred, y_true):
... return y_pred + y_true
...
>>> my_metric.process({'y_pred':4, 'y_true':5})
9
>>> @metrics.to_dict
... @metrics.lambda_metric('my_metric')
... def my_metric(y_pred, y_true):
... return y_pred + y_true
...
>>> my_metric.process({'y_pred':4, 'y_true':5})
{'my_metric': 9}
:param clazz: The class to *decorate*
:return: A :class:`.ToDict` instance or a :class:`.ToDict` wrapper of the given class
"""
if inspect.isclass(clazz):
class Wrapper(ToDict):
def __init__(self, *args, **kwargs):
super(Wrapper, self).__init__(clazz(*args, **kwargs))
Wrapper.__name__ = clazz.__name__
Wrapper.__doc__ = clazz.__doc__
return Wrapper
else:
return ToDict(clazz)
def _wrap_and_add_to_tree(clazz, child_func):
if inspect.isclass(clazz):
class Wrapper(MetricTree):
def __init__(self, *args, **kwargs):
inner = clazz(*args, **kwargs)
if isinstance(inner, MetricTree):
super(Wrapper, self).__init__(inner.root)
self.children = inner.children
else:
super(Wrapper, self).__init__(inner)
self.add_child(child_func(self.root))
Wrapper.__name__ = clazz.__name__
Wrapper.__doc__ = clazz.__doc__
return Wrapper
else:
inner = clazz
if not isinstance(inner, MetricTree):
inner = MetricTree(inner)
inner.add_child(child_func(inner))
return inner
[docs]def mean(clazz):
"""The :func:`mean` decorator is used to add a :class:`.Mean` to the :class:`.MetricTree` which will will output a
mean value at the end of each epoch. At build time, if the inner class is not a :class:`.MetricTree`, one will be
created. The :class:`.Mean` will also be wrapped in a :class:`.ToDict` for simplicity.
Example: ::
>>> import torch
>>> from torchbearer import metrics
>>> @metrics.mean
... @metrics.lambda_metric('my_metric')
... def metric(y_pred, y_true):
... return y_pred + y_true
...
>>> metric.reset({})
>>> metric.process({'y_pred':torch.Tensor([2]), 'y_true':torch.Tensor([2])}) # 4
{}
>>> metric.process({'y_pred':torch.Tensor([3]), 'y_true':torch.Tensor([3])}) # 6
{}
>>> metric.process({'y_pred':torch.Tensor([4]), 'y_true':torch.Tensor([4])}) # 8
{}
>>> metric.process_final()
{'my_metric': 6.0}
:param clazz: The class to *decorate*
:return: A :class:`.MetricTree` with a :class:`.Mean` appended or a wrapper class that extends :class:`.MetricTree`
"""
return _wrap_and_add_to_tree(clazz, lambda metric: ToDict(Mean(metric.name)))
[docs]def std(clazz):
"""The :func:`std` decorator is used to add a :class:`.Std` to the :class:`.MetricTree` which will will output a
population standard deviation value at the end of each epoch. At build time, if the inner class is not a
:class:`.MetricTree`, one will be created. The :class:`.Std` will also be wrapped in a :class:`.ToDict` (with '_std'
appended) for simplicity.
Example: ::
>>> import torch
>>> from torchbearer import metrics
>>> @metrics.std
... @metrics.lambda_metric('my_metric')
... def metric(y_pred, y_true):
... return y_pred + y_true
...
>>> metric.reset({})
>>> metric.process({'y_pred':torch.Tensor([2]), 'y_true':torch.Tensor([2])}) # 4
{}
>>> metric.process({'y_pred':torch.Tensor([3]), 'y_true':torch.Tensor([3])}) # 6
{}
>>> metric.process({'y_pred':torch.Tensor([4]), 'y_true':torch.Tensor([4])}) # 8
{}
>>> '%.4f' % metric.process_final()['my_metric_std']
'1.6330'
:param clazz: The class to *decorate*
:return: A :class:`.MetricTree` with a :class:`.Std` appended or a wrapper class that extends :class:`.MetricTree`
"""
return _wrap_and_add_to_tree(clazz, lambda metric: ToDict(Std(metric.name + '_std')))
[docs]def running_mean(clazz=None, batch_size=50, step_size=10):
"""The :func:`running_mean` decorator is used to add a :class:`.RunningMean` to the :class:`.MetricTree`. If the
inner class is not a :class:`.MetricTree` then one will be created. The :class:`.RunningMean` will be wrapped in a
:class:`.ToDict` (with 'running\_' prepended to the name) for simplicity.
.. note::
The decorator function does not need to be called if not desired, both: `@running_mean` and `@running_mean()`
are acceptable.
Example: ::
>>> import torch
>>> from torchbearer import metrics
>>> @metrics.running_mean(step_size=2) # Update every 2 steps
... @metrics.lambda_metric('my_metric')
... def metric(y_pred, y_true):
... return y_pred + y_true
...
>>> metric.reset({})
>>> metric.process({'y_pred':torch.Tensor([2]), 'y_true':torch.Tensor([2])}) # 4
{'running_my_metric': 4.0}
>>> metric.process({'y_pred':torch.Tensor([3]), 'y_true':torch.Tensor([3])}) # 6
{'running_my_metric': 4.0}
>>> metric.process({'y_pred':torch.Tensor([4]), 'y_true':torch.Tensor([4])}) # 8, triggers update
{'running_my_metric': 6.0}
:param clazz: The class to *decorate*
:param batch_size: See :class:`.RunningMean`
:param step_size: See :class:`.RunningMean`
:return: decorator or :class:`.MetricTree` instance or wrapper
"""
if clazz is None:
def decorator(clazz):
return running_mean(clazz, batch_size=batch_size, step_size=step_size)
return decorator
return _wrap_and_add_to_tree(clazz, lambda metric: ToDict(RunningMean('running_' + metric.name, batch_size=batch_size, step_size=step_size)))