def conv2d_linear_last_bn()

in benchmarks/horovod-resnet/train_imagenet_resnet_hvd.py [0:0]

• 21 lines of code
• 2 McCabe index (conditional complexity)

    def conv2d_linear_last_bn(self, inputs, *args, **kwargs):
        x = tf.layers.conv2d(
            inputs,
            data_format=self.data_format,
            use_bias=False,
            kernel_initializer=self.conv_initializer,
            activation=None,
            *args,
            **kwargs
        )
        param_initializers = {
            "moving_mean": tf.zeros_initializer(),
            "moving_variance": tf.ones_initializer(),
            "beta": tf.zeros_initializer(),
        }
        if self.adv_bn_init:
            param_initializers["gamma"] = tf.zeros_initializer()
        else:
            param_initializers["gamma"] = tf.ones_initializer()
        x = self.batch_norm(x, param_initializers=param_initializers)
        return x