tfrs.experimental.optimizers.ClippyAdagrad

An Adagrad variant with adaptive clipping.

tfrs.experimental.optimizers.ClippyAdagrad(
    learning_rate: Union[float, tf.keras.optimizers.schedules.LearningRateSchedule] = 0.001,
    initial_accumulator_value: float = 0.1,
    variable_relative_threshold: float = 0.1,
    accumulator_relative_threshold: float = 0.0,
    absolute_threshold: float = 1e-07,
    epsilon: float = 1e-07,
    export_clipping_factors: bool = False,
    clip_accumulator_update: bool = False,
    use_standard_accumulator_update: bool = False,
    name: str = 'ClippyAdagrad',
    **kwargs
) -> None

The adaptive clipping mechanism multiplies the learning rate for each model parameter w by a factor in (0, 1] that ensures that at each iteration w is never changed by more than: |w| * variable_relative_threshold

  + accumulator_relative_threshold / sqrt(accum) + absolute_threshold,

where accum is the respective Adagrad accumulator.

learning_rate Initial value for the learning rate: either a floating point value, or a tf.keras.optimizers.schedules.LearningRateSchedule instance. Defaults to 0.001. Note that Adagrad tends to benefit from higher initial learning rate values compared to other optimizers. To match the exact form in the original paper, use 1.0.
initial_accumulator_value A non-negative floating point value. Starting value for the Adagrad accumulators.
variable_relative_threshold A non-negative floating point value. The relative threshold factor for the adaptive clipping, relatively to the updated variable.
accumulator_relative_threshold A non-negative floating point value. The clipping threshold factor relatively to the inverse square root of the Adagrad accumulators. Default to 0.0 but a non-negative value (e.g., 1e-3) allows loosening the clipping threshold in later training.
absolute_threshold A non-negative floating point value. The absolute clipping threshold constant.
epsilon Small floating point value used to maintain numerical stability.
export_clipping_factors When set to True, will add an attribute to the optimizer, called clipping_factors, a list containing the scaling factors used to clip each variable in the model. Otherwise, the clipping_factors attribute is an empty list.
clip_accumulator_update When set to True, will also apply clipping on the Adagrad accumulator update. This may help improve convergence speed in cases where the gradient contains outliers. This cannot be set to True when use_standard_accumulator_update is set to True.
use_standard_accumulator_update When set to True, will update the accumulator before calculating the Adagrad step, as in the classical Adagrad method. This cannot be set to True when clip_accumulator_update is set to True.
name String. The name to use for momentum accumulator weights created by the optimizer.
**kwargs Other arguments. See tf.keras.optimizers.Optimizer docs.

ValueError If both clip_accumulator_update and use_standard_accumulator_update are set to True.

iterations The number of training steps this optimizer has run.
learning_rate The learning rate constant or schedule.
clipping_factors When the argument export_clipping_factors is set to True will contain a list of the scaling factors used to clip each variable in the model. Otherwise, contains an empty list.
variables Returns variables of this optimizer.

Methods

add_variable

add_variable(
    shape, dtype=None, initializer='zeros', name=None
)

Create an optimizer variable.

Args
shape A list of integers, a tuple of integers, or a 1-D Tensor of type int32. Defaults to scalar if unspecified.
dtype The DType of the optimizer variable to be created. Defaults to tf.keras.backend.floatx if unspecified.
initializer string or callable. Initializer instance.
name The name of the optimizer variable to be created.

Returns
An optimizer variable, in the format of tf.Variable.

add_variable_from_reference

add_variable_from_reference(
    model_variable, variable_name, shape=None, initial_value=None
)

Create an optimizer variable from model variable.

Create an optimizer variable based on the information of model variable. For example, in SGD optimizer momemtum, for each model variable, a corresponding momemtum variable is created of the same shape and dtype.

Args
model_variable tf.Variable. The corresponding model variable to the optimizer variable to be created.
variable_name String. The name prefix of the optimizer variable to be created. The create variables name will follow the pattern {variable_name}/{model_variable.name}, e.g., momemtum/dense_1.
shape List or Tuple, defaults to None. The shape of the optimizer variable to be created. If None, the created variable will have the same shape as model_variable.
initial_value A Tensor, or Python object convertible to a Tensor, defaults to None. The initial value of the optimizer variable, if None, the initial value will be default to 0.

Returns
An optimizer variable.

aggregate_gradients

aggregate_gradients(
    grads_and_vars
)

Aggregate gradients on all devices.

By default, we will perform reduce_sum of gradients across devices. Users can implement their own aggregation logic by overriding this method.

Args
grads_and_vars List of (gradient, variable) pairs.

Returns
List of (gradient, variable) pairs.

apply_gradients

apply_gradients(
    grads_and_vars, name=None, skip_gradients_aggregation=False, **kwargs
)

Apply gradients to variables.

Args
grads_and_vars List of (gradient, variable) pairs.
name string, defaults to None. The name of the namescope to use when creating variables. If None, self.name will be used.
skip_gradients_aggregation If true, gradients aggregation will not be performed inside optimizer. Usually this arg is set to True when you write custom code aggregating gradients outside the optimizer.
**kwargs keyword arguments only used for backward compatibility.

Returns
A tf.Variable, representing the current iteration.

Raises
TypeError If grads_and_vars is malformed.
RuntimeError If called in a cross-replica context.

build

View source

build(
    var_list: Sequence[tf.Variable]
) -> None

Initialize the optimizer's variables, such as momemtum variables.

This function has to be implemented by subclass optimizers, and subclass optimizers need to call super().build(var_list).

Args
var_list List of model variables to build optimizers on. For example, SGD optimizer with momentum will store one momentum variable corresponding to each model variable.

compute_gradients

compute_gradients(
    loss, var_list, tape=None
)

Compute gradients of loss on trainable variables.

Args
loss Tensor or callable. If a callable, loss should take no arguments and return the value to minimize.
var_list list or tuple of Variable objects to update to minimize loss, or a callable returning the list or tuple of Variable objects. Use callable when the variable list would otherwise be incomplete before minimize since the variables are created at the first time loss is called.
tape (Optional) tf.GradientTape. If loss is provided as a Tensor, the tape that computed the loss must be provided.

Returns
A list of (gradient, variable) pairs. Variable is always present, but gradient can be None.

exclude_from_weight_decay

exclude_from_weight_decay(
    var_list=None, var_names=None
)

Exclude variables from weight decay.

This method must be called before the optimizer's build method is called. You can set specific variables to exclude out, or set a list of strings as the anchor words, if any of which appear in a variable's name, then the variable is excluded.

Args
var_list A list of tf.Variables to exclude from weight decay.
var_names A list of strings. If any string in var_names appear in the model variable's name, then this model variable is excluded from weight decay. For example, var_names=['bias'] excludes all bias variables from weight decay.

finalize_variable_values

finalize_variable_values(
    var_list
)

Set the final value of model's trainable variables.

Sometimes there are some extra steps before ending the variable updates, such as overriding the model variables with its average value.

Args
var_list list of model variables.

from_config

@classmethod
from_config(
    config, custom_objects=None
)

Creates an optimizer from its config.

This method is the reverse of get_config, capable of instantiating the same optimizer from the config dictionary.

Args
config A Python dictionary, typically the output of get_config.
custom_objects A Python dictionary mapping names to additional user-defined Python objects needed to recreate this optimizer.

Returns
An optimizer instance.

get_config

View source

get_config() -> Dict[str, Any]

Returns the config of the optimizer.

An optimizer config is a Python dictionary (serializable) containing the configuration of an optimizer. The same optimizer can be reinstantiated later (without any saved state) from this configuration.

Subclass optimizer should override this method to include other hyperparameters.

Returns
Python dictionary.

load_own_variables

load_own_variables(
    store
)

Set the state of this optimizer object.

minimize

minimize(
    loss, var_list, tape=None
)

Minimize loss by updating var_list.

This method simply computes gradient using tf.GradientTape and calls apply_gradients(). If you want to process the gradient before applying then call tf.GradientTape and apply_gradients() explicitly instead of using this function.

Args
loss Tensor or callable. If a callable, loss should take no arguments and return the value to minimize.
var_list list or tuple of Variable objects to update to minimize loss, or a callable returning the list or tuple of Variable objects. Use callable when the variable list would otherwise be incomplete before minimize since the variables are created at the first time loss is called.
tape (Optional) tf.GradientTape.

Returns
None

save_own_variables

save_own_variables(
    store
)

Get the state of this optimizer object.

set_weights

set_weights(
    weights
)

Set the weights of the optimizer.

Args
weights a list of tf.Variables or numpy arrays, the target values of optimizer variables. It should have the same order as self._variables.

update_step

View source

update_step(
    grad: Union[tf.Tensor, tf.IndexedSlices], variable: tf.Variable
) -> None

Update step given gradient and the associated model variable.