def call(self, target, prediction):
size = len(prediction.get_shape().as_list())
reduce_ax = list(range(1, size))
eps = 1e-8
true_positive = K.sum(prediction * target, axis=reduce_ax)
target_positive = K.sum((target), axis=reduce_ax)
predicted_positive = K.sum((prediction), axis=reduce_ax)
fb_numerator = (1 + self.beta**2) * true_positive + eps
fb_denominator = (self.beta**
2) * target_positive + predicted_positive + eps
return 1 - fb_numerator / fb_denominator
def loss_fn(output):
return K.sum(output * (K.variable(np.array(targets)) - output))
def loss_fn(output):
return K.sum(output * (K.constant(np.array(targets)) - output))
def loss_fn(output):
# tp = K.cast(np.array(targets), K.floatx())
return K.sum(output * K.cast(np.array(targets), K.floatx()))
def loss_fn(output):
return K.sum(output * (targets - output))
def loss_fn(output):
return K.sum(output * targets)