tf.contrib.estimator.clip_gradients_by_norm

tf.contrib.estimator.clip_gradients_by_norm(
    optimizer,
    clip_norm
)

Returns an optimizer which clips gradients before applying them.

Example:

optimizer = tf.train.ProximalAdagradOptimizer(
    learning_rate=0.1,
    l1_regularization_strength=0.001)
optimizer = tf.contrib.estimator.clip_gradients_by_norm(
    optimizer, clip_norm)
estimator = tf.estimator.DNNClassifier(
    feature_columns=[...],
    hidden_units=[1024, 512, 256],
    optimizer=optimizer)

Args:

  • optimizer: An tf.Optimizer object to apply gradients.
  • clip_norm: A 0-D (scalar) Tensor > 0. The clipping ratio.

Returns:

A tf.Optimizer.