1.MSE
def mean_squared_error(y_true, y_pred):
return K.mean(math_ops.square(y_pred - y_true), axis=-1)
以语义分割为例：
y_true :shape (224,224,2)
y_pred:shape (224,224,2)
代码中axis=-1，代表取出最后一维度取平均值，公式中的N=2
2.绝对值
def mean_absolute_error(y_true, y_pred):
return K.mean(math_ops.abs(y_pred - y_true), axis=-1)
3. 预测值百分率
|y_pred - y_true| / | y_true |
def mean_absolute_percentage_error(y_true, y_pred):
diff = math_ops.abs(
(y_true - y_pred) / K.clip(math_ops.abs(y_true), K.epsilon(), None))
return 100. * K.mean(diff, axis=-1)
4. MSE + log
def mean_squared_logarithmic_error(y_true, y_pred):
first_log = math_ops.log(K.clip(y_pred, K.epsilon(), None) + 1.)
second_log = math_ops.log(K.clip(y_true, K.epsilon(), None) + 1.)
return K.mean(math_ops.square(first_log - second_log), axis=-1)
5.Hinge损失
def hinge(y_true, y_pred):
return K.mean(math_ops.maximum(1. - y_true * y_pred, 0.), axis=-1)
6.categorical_hinge 损失
def categorical_hinge(y_true, y_pred):
pos = math_ops.reduce_sum(y_true * y_pred, axis=-1)
neg = math_ops.reduce_max((1. - y_true) * y_pred, axis=-1)
return math_ops.maximum(0., neg - pos + 1.)
7.交叉熵
def categorical_crossentropy(y_true, y_pred):
return K.categorical_crossentropy(y_true, y_pred)
8. one-hot 交叉熵
def sparse_categorical_crossentropy(y_true, y_pred):
return K.sparse_categorical_crossentropy(y_true, y_pred)
9. 二值交叉熵
def binary_crossentropy(y_true, y_pred):
return K.mean(K.binary_crossentropy(y_true, y_pred), axis=-1)
