Returns an op that deletes this tensor array from its resource container.
Returns an op that deletes this tensor array from its resource container.
This enables the user to close and release the resource in the middle of a step/run.
Name for the created op.
Created op.
Boolean value indicating whether to place the tensor array on the same device as
the tensor used on its first write call (write operations include write
,
unstack
, and split
).
Boolean value indicating whether to place the tensor array on the same device as
the tensor used on its first write call (write operations include write
,
unstack
, and split
). If false
, the tensor array will be placed on the device
determined by the op creation context available during its initialization.
Creates an op that concatenates the elements of the tensor array.
Creates an op that concatenates the elements of the tensor array.
The op takes T
elements with shapes [n0, d0, d1, ...]
, [n1, d0, d1, ...]
, ..., [n(T-1), d0, d1, ...]
and
concatenates them into a tensor with shape [n0 + n1 + ... + n(T-1), d0, d1, ...]
.
All elements must have been written and must have the same shape, except for their first dimension.
Name for the created op.
Tensor with all of the elements in the tensor array, concatenated along the first axis.
Data type of the tensor array elements.
Float scalar tensor for the tensor array, used to control gradient flow.
Creates an op that gathers specific elements from this tensor array.
Creates an op that gathers specific elements from this tensor array.
Note that all elements selected by indices
must have been written and must have the same shape.
One-dimensional tensor containing the positions in the tensor array from which to read tensor elements.
Name for the created op.
Tensor containing the gathered elements, concatenated along a new axis (the new dimension 0
).
Tensor handle to the tensor array.
Returns a tensor array with the same content and properties as this one.
Returns a tensor array with the same content and properties as this one.
New TensorArray object with a flow that ensures the control dependencies from the contexts will become control dependencies for writes, reads, etc. Use this object for all subsequent operations.
Boolean value indicating whether shape inference is enabled.
Boolean value indicating whether shape inference is enabled. If true
, all elements
must have the same shape.
Creates an op that reads an element from this tensor array.
Creates an op that reads an element from this tensor array.
Position to read from, inside the tensor array.
Name for the created op.
Tensor in the specified position of the tensor array.
Creates an op that scatters the provided elements along indices of this tensor array.
Creates an op that scatters the provided elements along indices of this tensor array.
Note that indices
must be a vector and its length must match the first dimension of value
.
One-dimensional tensor containing the positions in the tensor array at which to write the tensor elements.
Concatenated tensor to write to the tensor array.
Name for the created op.
Output flow of the tensor array, used to enforce proper chaining of operations.
Returns an op that gets the current size of the tensor array.
Returns an op that gets the current size of the tensor array.
Name for the created op.
Created op output, containing the current size of the tensor array.
Splits the values of a tensor into a tensor array.
Splits the values of a tensor into a tensor array.
(N+1)-dimensional tensor to split. Must have the same data type as this tensor array.
1-D integer tensor with the lengths to use when splitting input
along its first dimension.
Name for the created op.
Tensor array with flow that ensures the split occurs. Use this object for all subsequent operations.
Creates an op that returns the elements in this tensor array as a stacked tensor.
Creates an op that returns the elements in this tensor array as a stacked tensor.
Note that all elements of this tensor array must have been written and must have the same shape.
If the elements have rank R
, then the returned tensor shape will be equal to R + 1
.
Name for the created op.
Stacked tensor.
Converts this tensor array to an output (i.e., dense symbolic tensor), by stacking it.
Converts this tensor array to an output (i.e., dense symbolic tensor), by stacking it.
Creates an op that unstacks the values of a tensor in this tensor array.
Creates an op that unstacks the values of a tensor in this tensor array.
If the input value shapes have rank R
, then the output tensor array will contain elements whose shapes have
rank R - 1
.
Tensor to unstack.
Name for the created op.
New tensor array object with flow that ensures the unstack occurs. Use this object for all subsequent operations.
Creates an op that writes an element to this tensor array.
Creates an op that writes an element to this tensor array.
Position to write to, inside the tensor array.
Tensor to write to the tensor array.
Name for the created op.
Output flow of the tensor array, used to enforce proper chaining of operations.
Class wrapping dynamic-sized, per-time-step, write-once tensor arrays.
This class is meant to be used with dynamic iteration primitives such as
whileLoop
andmapFunction
. It supports gradient back-propagation via special "flow" control flow dependencies.Note that the name of the
TensorArray
(even if passed in) is uniquified automatically. Each time a newTensorArray
is created at runtime it is assigned its own name for the duration of the run. This avoids name collisions if aTensorArray
is created within awhileLoop
.