如何在TensorFlow的Adam优化器中添加正则化（L1 / L2）？

Question

我目前正在研究Google's Machine Learning Crash Course，并且正在尝试使用DNNClassifier估计器来解决二进制分类问题。我正在尝试将正则化（L1 / L2）添加到Adam优化器，因为它尚未在函数中定义为自变量。有什么想法如何实施吗？下面是我的代码：

steps = 1000
periods = 10
steps_per_period = steps / periods

my_optimiser = tf.train.AdamOptimizer(learning_rate = learning_rate)
my_optimiser = tf.contrib.estimator.clip_gradients_by_norm(my_optimiser, 5.0)
dnn_classifier = tf.estimator.DNNClassifier(
        feature_columns = construct_feature_columns(training_features),
        n_classes = 2,
        hidden_units = hidden_units,
        optimizer = my_optimiser)

training_input_fn = lambda: my_input_fn(
  training_features, 
  training_targets, 
  batch_size = batch_size)
predict_training_input_fn = lambda: my_input_fn(
  training_features, 
  training_targets, 
  num_epochs = 1, 
  shuffle = False)
predict_validation_input_fn = lambda: my_input_fn(
  validation_features, 
  validation_targets, 
  num_epochs = 1, 
  shuffle = False)

training_log_losses = []
validation_log_losses = []

for period in range (0, periods):

    dnn_classifier.train(
            input_fn = training_input_fn,
            steps = steps_per_period
            )