def construct_weights()

in models.py [0:0]

• 23 lines of code
• 5 McCabe index (conditional complexity)

    def construct_weights(self, scope=''):
        weights = {}
        dtype = tf.float32

        if FLAGS.cclass and FLAGS.dataset == "cifar10":
            classes = 10
        elif FLAGS.cclass and FLAGS.dataset == "imagenet":
            classes = 1000
        else:
            classes = 1

        with tf.variable_scope(scope):
            # First block
            init_conv_weight(weights, 'c1_pre', 3, self.channels, 128)
            init_res_weight(weights, 'res_optim', 3, 128, self.dim_hidden, classes=classes)
            init_res_weight(weights, 'res_1', 3, self.dim_hidden, self.dim_hidden, classes=classes)
            init_res_weight(weights, 'res_2', 3, self.dim_hidden, self.dim_hidden, classes=classes)
            init_res_weight(weights, 'res_3', 3, self.dim_hidden, 2*self.dim_hidden, classes=classes)
            init_res_weight(weights, 'res_4', 3, 2*self.dim_hidden, 2*self.dim_hidden, classes=classes)
            init_res_weight(weights, 'res_5', 3, 2*self.dim_hidden, 2*self.dim_hidden, classes=classes)
            init_res_weight(weights, 'res_6', 3, 2*self.dim_hidden, 4*self.dim_hidden, classes=classes)
            init_res_weight(weights, 'res_7', 3, 4*self.dim_hidden, 4*self.dim_hidden, classes=classes)
            init_res_weight(weights, 'res_8', 3, 4*self.dim_hidden, 4*self.dim_hidden, classes=classes)
            init_fc_weight(weights, 'fc5', 4*self.dim_hidden , 1, spec_norm=False)

            init_attention_weight(weights, 'atten', self.dim_hidden, self.dim_hidden / 2, trainable_gamma=True)

        return weights