tf.ConditionalAccumulator

Class ConditionalAccumulator

Inherits From: ConditionalAccumulatorBase

Defined in tensorflow/python/ops/data_flow_ops.py.

See the guide: Inputs and Readers > Conditional Accumulators

A conditional accumulator for aggregating gradients.

Up-to-date gradients (i.e., time step at which gradient was computed is equal to the accumulator's time step) are added to the accumulator.

Extraction of the average gradient is blocked until the required number of gradients has been accumulated.

Properties

accumulator_ref

The underlying accumulator reference.

dtype

The datatype of the gradients accumulated by this accumulator.

name

The name of the underlying accumulator.

Methods

__init__

__init__(
    dtype,
    shape=None,
    shared_name=None,
    name='conditional_accumulator'
)

Creates a new ConditionalAccumulator.

Args:

  • dtype: Datatype of the accumulated gradients.
  • shape: Shape of the accumulated gradients.
  • shared_name: Optional. If non-empty, this accumulator will be shared under the given name across multiple sessions.
  • name: Optional name for the accumulator.

apply_grad

apply_grad(
    grad,
    local_step=0,
    name=None
)

Attempts to apply a gradient to the accumulator.

The attempt is silently dropped if the gradient is stale, i.e., local_step is less than the accumulator's global time step.

Args:

  • grad: The gradient tensor to be applied.
  • local_step: Time step at which the gradient was computed.
  • name: Optional name for the operation.

Returns:

The operation that (conditionally) applies a gradient to the accumulator.

Raises:

  • ValueError: If grad is of the wrong shape

num_accumulated

num_accumulated(name=None)

Number of gradients that have currently been aggregated in accumulator.

Args:

  • name: Optional name for the operation.

Returns:

Number of accumulated gradients currently in accumulator.

set_global_step

set_global_step(
    new_global_step,
    name=None
)

Sets the global time step of the accumulator.

The operation logs a warning if we attempt to set to a time step that is lower than the accumulator's own time step.

Args:

  • new_global_step: Value of new time step. Can be a variable or a constant
  • name: Optional name for the operation.

Returns:

Operation that sets the accumulator's time step.

take_grad

take_grad(
    num_required,
    name=None
)

Attempts to extract the average gradient from the accumulator.

The operation blocks until sufficient number of gradients have been successfully applied to the accumulator.

Once successful, the following actions are also triggered:

  • Counter of accumulated gradients is reset to 0.
  • Aggregated gradient is reset to 0 tensor.
  • Accumulator's internal time step is incremented by 1.

Args:

  • num_required: Number of gradients that needs to have been aggregated
  • name: Optional name for the operation

Returns:

A tensor holding the value of the average gradient.

Raises:

  • InvalidArgumentError: If num_required < 1