nsl.estimator.add_adversarial_regularization
Stay organized with collections
Save and categorize content based on your preferences.
Adds adversarial regularization to a tf.estimator.Estimator.
nsl.estimator.add_adversarial_regularization(
estimator, optimizer_fn=None, adv_config=None
)
The returned estimator will include the adversarial loss as a regularization
term in its training objective, and will be trained using the optimizer
provided by optimizer_fn. optimizer_fn (along with the hyperparameters)
should be set to the same one used in the base estimator.
If optimizer_fn is not set, a default optimizer tf.train.AdagradOptimizer
with learning_rate=0.05 will be used.
Args |
|---|
estimator
|
A tf.estimator.Estimator object, the base model.
|
optimizer_fn
|
A function that accepts no arguments and returns an instance
of tf.train.Optimizer. This optimizer (instead of the one used in
estimator) will be used to train the model. If not specified, default to
tf.train.AdagradOptimizer with learning_rate=0.05.
|
adv_config
|
An instance of nsl.configs.AdvRegConfig that specifies various
hyperparameters for adversarial regularization.
|
Returns |
|---|
A modified tf.estimator.Estimator object with adversarial regularization
incorporated into its loss.
|
Except as otherwise noted, the content of this page is licensed under the Creative Commons Attribution 4.0 License, and code samples are licensed under the Apache 2.0 License. For details, see the Google Developers Site Policies. Java is a registered trademark of Oracle and/or its affiliates.
Last updated 2024-01-26 UTC.
[null,null,["Last updated 2024-01-26 UTC."],[],[],null,["# nsl.estimator.add_adversarial_regularization\n\n\u003cbr /\u003e\n\n|---------------------------------------------------------------------------------------------------------------------------------------------------------------------------|\n| [View source on GitHub](https://github.com/tensorflow/neural-structured-learning/blob/v1.4.0/neural_structured_learning/estimator/adversarial_regularization.py#L30-L153) |\n\nAdds adversarial regularization to a `tf.estimator.Estimator`. \n\n nsl.estimator.add_adversarial_regularization(\n estimator, optimizer_fn=None, adv_config=None\n )\n\nThe returned estimator will include the adversarial loss as a regularization\nterm in its training objective, and will be trained using the optimizer\nprovided by `optimizer_fn`. `optimizer_fn` (along with the hyperparameters)\nshould be set to the same one used in the base `estimator`.\n\nIf `optimizer_fn` is not set, a default optimizer `tf.train.AdagradOptimizer`\nwith `learning_rate=0.05` will be used.\n\n\u003cbr /\u003e\n\n\u003cbr /\u003e\n\n\u003cbr /\u003e\n\n| Args ---- ||\n|----------------|--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|\n| `estimator` | A `tf.estimator.Estimator` object, the base model. |\n| `optimizer_fn` | A function that accepts no arguments and returns an instance of `tf.train.Optimizer`. This optimizer (instead of the one used in `estimator`) will be used to train the model. If not specified, default to `tf.train.AdagradOptimizer` with `learning_rate=0.05`. |\n| `adv_config` | An instance of [`nsl.configs.AdvRegConfig`](../../nsl/configs/AdvRegConfig) that specifies various hyperparameters for adversarial regularization. |\n\n\u003cbr /\u003e\n\n\u003cbr /\u003e\n\n\u003cbr /\u003e\n\n\u003cbr /\u003e\n\n| Returns ------- ||\n|---|---|\n| A modified `tf.estimator.Estimator` object with adversarial regularization incorporated into its loss. ||\n\n\u003cbr /\u003e"]]
View source on GitHub