keras_ver_Onet_v7.py 文件源码

def myloss (random_int,type): # need to make sure input type
    if print_progress: random_int = tf.Print(random_int, ['random in cls',random_int])
    condition1 = random_int[0] > tf.constant(rand_threshold[1])
    condition0 = random_int[0] > tf.constant(rand_threshold[0])
    condition_default =  condition0 & condition1
    if type =='cls':
        def lossfun(y_true, y_pred):
            if print_progress: condition = tf.Print(condition1, ['rand int[0]:', random_int[0],
                                                    ' tf.constant:', tf.constant(rand_threshold[1]),
                                                    ' condition1:', condition1 ])
            val= tf.case({ condition1: lambda: K.mean(K.square(y_pred - y_true), axis=-1),
                           condition0: lambda: 0 * K.mean(K.square(y_true), axis=-1)
                           },
                         default=lambda:0 * K.mean(K.square(y_true), axis=-1),
                         exclusive=False )

            if print_progress: val = tf.Print(val, ['cls loss out:',val,
                                                    ' rand int received:',random_int,
                                                    'condition',condition1])
            val.set_shape(K.mean(K.square(y_true), axis=-1).shape)
            return val
    elif type =='roi':
        def lossfun(y_true, y_pred):
            if print_progress: condition = tf.Print(condition1, ['rand int[0]:', random_int[0],
                                                                ' tf.constant:', tf.constant(rand_threshold),
                                                                ' condition:', condition1])
            val= tf.case({ condition1: lambda: 0 * K.mean(K.square(y_true), axis=-1),
                           condition0: lambda: K.mean(K.square(y_pred - y_true), axis=-1)
                         },
                         default=lambda: 0 * K.mean(K.square(y_true), axis=-1),exclusive=False)
            if print_progress: val = tf.Print(val, ['roi loss out :', val,
                                                    ' rand int received:', random_int,
                                                    'condition', condition1])
            val.set_shape(K.mean(K.square(y_true), axis=-1).shape)
            return val
    else :
        def lossfun(y_true, y_pred):
            if print_progress: condition = tf.Print(condition1, ['rand int[0]:', random_int[0],
                                                                 ' tf.constant:', tf.constant(rand_threshold),
                                                                 ' condition:', condition1])
            val = tf.case({condition1: lambda: 0 * K.mean(K.square(y_true), axis=-1),
                           condition0: lambda: 0 * K.mean(K.square(y_true), axis=-1)
                           },
                          default=lambda: K.mean(K.square(y_pred - y_true), axis=-1),exclusive=False)
            val.set_shape(K.mean(K.square(y_true), axis=-1).shape)
            if print_progress: val = tf.Print(val, ['pts loss out :', val,
                                                    ' rand int received:', random_int,
                                                    'condition', condition1])
            return val
    return lossfun