Structure describing permissible absolute and relative error bounds.

A description of a single command line flag, holding its name, type, usage text, and a pointer to the corresponding variable.

Initializer base class: all initializers inherit from this class.

This is a helper for creating a Node and adding it to a Graph. Internally, it uses a NodeDefBuilder to automatically set attrs that can be inferred fr

Builder class passed to the REGISTER_OP() macro.

Shape rank and dimensions are accessed through InferenceContext.

An OpTestBuilder is a graph builder class that takes as input an operator to test, its inputs and attributes, and builds a graph that executes the ope

@ingroup core Denotes success or failure of a call in Tensorflow.

A typed tuple. A correct type is a tuple with the correct length where each element has the correct type.

Scanner provides simplified string parsing, in which a string is parsed as a series of scanning calls (e.g. One, Any, Many, OneLiteral, Eos), and then

Handles caching of transformations while building the model. `_FeatureColumn` specifies how to digest an input column to the network. Some featur

A DeviceType is just a string, but we wrap it up in a class to give some type checking as we're passing these around

Non-assignable values should work.

Identifier for a tensor within a step. first == operation_name, second == output_index Note: does not own backing storage for name.

MetricSpec connects a model to metric functions. The MetricSpec class contains all information necessary to connect the output of a `model_fn` to

See the @{$variables$Variables How To} for a high level overview. A variable maintains state in the graph across calls to `run()`. You add a vari

GradientStats holds first and second order gradient stats.

Mock subclass of CursesUI that bypasses actual terminal manipulations.

2D convolution layer (e.g. spatial convolution over images). This layer creates a convolution kernel that is convolved with the layer input to pr

Represents the type of the elements in a `Tensor`. The following `DType` objects are defined: * `tf.float16`: 16-bit half-precision floating-poi

@ingroup core Represents an n-dimensional array of values.

A helper class to make AttrSlice from initializer lists

SessionOptions contains configuration information for a session.

Specifies the ndim, dtype and shape of every input to a layer. Every layer should expose (if appropriate) an `input_spec` attribute: a list of in

TensorFlow Debugger (tfdbg) stepper. The stepper provides ability to perform "continue to" actions on a graph, given fetch and feeds. The stepper

A tensorflow Op may need access to different kinds of memory that are not simply a function of the device to which the Op has been assigned. For exam

For the purpose of control flow, every tensor produced by TensorFlow is conceptually tagged by a 'FrameAndIter'. FrameAndIter consists of a 'frame_id'

Compute `Y = g(X; shift, scale) = scale @ X + shift`. Here `scale = c * I + diag(D1) + tril(L) + V @ diag(D2) @ V.T`. In TF parlance, the `scale

Represents the output of 'node' at 'index'.

Batch normalization layer (Ioffe and Szegedy, 2014). Normalize the activations of the previous layer at each batch, i.e. applies a transformation

Subclasses the wrapper class for testing. Overrides its CLI-related methods for headless testing environments. Inserts observer variables for ass

Non-assignable values should work.

Represents a potentially large set of elements. A `Dataset` can be used to represent an input pipeline as a collection of elements (nested struct

A ShapeLayout object encapsulates the layout of a particular shape (including tuples). This differs from the Layout proto which describes the layout o

Applies an activation function to an output. Arguments: activation: name of activation function to use or alternatively, a Theano o

This is exactly the output of TF's ParseExample Op. Documentation is available in: tensorflow/core/ops/parsing_ops.cc

Compute `Y = g(X) = exp(X)`. Example Use: ```python # Create the Y=g(X)=exp(X) transform which works only on Tensors with 1 # batch

Stochastic dual coordinate ascent solver for linear models. This class currently only supports a single machine (multi-threaded) implementati

Max pooling operation for spatial data. Arguments: pool_size: integer or tuple of 2 integers, factors by which to downscale (vertic

Tuple used by LSTM Cells for `state_size`, `zero_state`, and output state. Stores two elements: `(c, h)`, in that order. Where `c` is the hidden st

Manage and manipulate `Distribution` shape. Terminology: Recall that a `Tensor` has: - `shape`: size of `Tensor` dimensions, - `ndi

Fake implementation of `SessionRunHook`.

Linear map: sum_i(args[i] * W[i]), where W[i] is a variable. Args: args: a 2D Tensor or a list of 2D, batch, n, Tensors. output_size: int,

Describes the configuration for the algorithms that will used. Arguments: algorithm: the primary algorithm that should be used. algorithm_no_scratch:

Optimize TF computations by reducing the arithmetic complexity required to run a model.

Options for calling ImportGraphDef().

Prune a model to make it more efficient: Remove unnecessary operations. Optimize gradient computations.

Represents the value of one dimension in a TensorShape.