Keras中自定义目标函数(损失函数)
在做kaggle项目时,看到一个人设计了一个unet,使用自定义的iou作为损失函数,才想起来原来可以自己设计损失函数…
为了实现自己的目标函数,自然想到先看下Keras中的目标函数是定义的,查下源码发现在/usr/local/lib/python3.5/dist-packages/keras中(我的系统是ubuntu16.04,使用系统自带的python3.5),Keras已经定义了一系列的目标函数。
当然是要先查看下Keras源码了…
"""Built-in loss functions.
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import six
from . import backend as K
from .utils.generic_utils import deserialize_keras_object
from .utils.generic_utils import serialize_keras_object
def mean_squared_error(y_true, y_pred):
return K.mean(K.square(y_pred - y_true), axis=-1)
def mean_absolute_error(y_true, y_pred):
return K.mean(K.abs(y_pred - y_true), axis=-1)
def mean_absolute_percentage_error(y_true, y_pred):
diff = K.abs((y_true - y_pred) / K.clip(K.abs(y_true),
K.epsilon(),
None))
return 100. * K.mean(diff, axis=-1)
def mean_squared_logarithmic_error(y_true, y_pred):
first_log = K.log(K.clip(y_pred, K.epsilon(), None) + 1.)
second_log = K.log(K.clip(y_true, K.epsilon(), None) + 1.)
return K.mean(K.square(first_log - second_log), axis=-1)
def squared_hinge(y_true, y_pred):
return K.mean(K.square(K.maximum(1. - y_true * y_pred, 0.)), axis=-1)
def hinge(y_true, y_pred):
return K.mean(K.maximum(1. - y_true * y_pred, 0.), axis=-1)
def categorical_hinge(y_true, y_pred):
pos = K.sum(y_true * y_pred, axis=-1)
neg = K.max((1. - y_true) * y_pred, axis=-1)
return K.maximum(0., neg - pos + 1.)
def logcosh(y_true, y_pred):
"""Logarithm of the hyperbolic cosine of the prediction error.
`log(cosh(x))` is approximately equal to `(x ** 2) / 2` for small `x` and
to `abs(x) - log(2)` for large `x`. This means that 'logcosh' works mostly
like the mean squared error, but will not be so strongly affected by the
occasional wildly incorrect prediction.
# Arguments
y_true: tensor of true targets.
y_pred: tensor of predicted targets.
# Returns
Tensor with one scalar loss entry per sample.
"""
def _logcosh(x):
return x + K.softplus(-2. * x) - K.log(2.)
return K.mean(_logcosh(y_pred - y_true), axis=-1)
def categorical_crossentropy(y_true, y_pred):
return K.categorical_crossentropy(y_true, y_pred)
def sparse_categorical_crossentropy(y_true, y_pred):
return K.sparse_categorical_crossentropy(y_true, y_pred)
def binary_crossentropy(y_true, y_pred):
return K.mean(K.binary_crossentropy(y_true, y_pred), axis=-1)
def kullback_leibler_divergence(y_true, y_pred):
y_true = K.clip(y_true, K.epsilon(), 1)
y_pred = K.clip(y_pred, K.epsilon(), 1)
return K.sum(y_true * K.log(y_true / y_pred), axis=-1)
def poisson(y_true, y_pred):
return K.mean(y_pred - y_true * K.log(y_pred + K.epsilon()), axis=-1)
def cosine_proximity(y_true, y_pred):
y_true = K.l2_normalize(y_true, axis=-1)
y_pred = K.l2_normalize(y_pred, axis=-1)
return -K.sum(y_true * y_pred, axis=-1)
# Aliases.
mse = MSE = mean_squared_error
mae = MAE = mean_absolute_error
mape = MAPE = mean_absolute_percentage_error
msle = MSLE = mean_squared_logarithmic_error
kld = KLD = kullback_leibler_divergence
cosine = cosine_proximity
def serialize(loss):
return serialize_keras_object(loss)
def deserialize(name, custom_objects=None):
return deserialize_keras_object(name,
module_objects=globals(),
custom_objects=custom_objects,
printable_module_name='loss function')
def get(identifier):
"""Get the `identifier` loss function.
# Arguments
identifier: None or str, name of the function.
# Returns
The loss function or None if `identifier` is None.
# Raises
ValueError if unknown identifier.
"""
if identifier is None:
return None
if isinstance(identifier, six.string_types):
identifier = str(identifier)
return deserialize(identifier)
if isinstance(identifier, dict):
return deserialize(identifier)
elif callable(identifier):
return identifier
else:
raise ValueError('Could not interpret '
'loss function identifier:', identifier)
看到源码后,事情就简单多了
首先定义损失函数IoU
## intersection over union
def IoU(y_true, y_pred, eps=1e-6):
if np.max(y_true) == 0.0:
return IoU(1-y_true, 1-y_pred) ## empty image; calc IoU of zeros
intersection = K.sum(y_true * y_pred, axis=[1,2,3])
union = K.sum(y_true, axis=[1,2,3]) + K.sum(y_pred, axis=[1,2,3]) - intersection
return -K.mean( (intersection + eps) / (union + eps), axis=0)
然后执行编译
seg_model.compile(optimizer=Adam(1e-3, decay=1e-6), loss=IoU, metrics=['binary_accuracy'])就是这么简单…