CIntToPy */

Implements a Feed-forward layer with multiple inputs, pad-removing, etc. Args: inputs_list: list of input tensors hparams: hyper-parame

Preprocess the input at the beginning of each step. Args: x: input tensor step: step hparams: model hyper-parameters Returns:

RaiseArgTupleInvalid */

GetBuiltinName */

CIntFromPy */

Adds deault hparams for all of the variants of the Universal Transformer. Args: hparams: default hparams (usually one of the standard hparam

ParseKeywords */

CIntToPy */

CIntToPy */

Preprocessing of adjacency matrix for simple GCN model and conversion to tuple representation.

Row-normalize feature matrix and convert to tuple representation

__GNUC__ */

SetupReduce */

CodeObjectCache */

Passes the input through num_layers of vanilla transformer layers. Args: x: input num_layers: number of layers Returns:

Wrapper for tf.matmul (sparse vs dense).

Fills the memory slot at a particular index with the given value. Args: memory: a 4-d tensor [memory_size, batch, length, channel] containin

Glorot & Bengio (AISTATS 2010) init.

dataset: name of dataset test_proportion: ratio of test train split seed: random seed for random splitting of dataset

dataset: name of dataset test_proportion: ratio of test train split seed: random seed for random splitting of dataset

Symmetrically normalize adjacency matrix.

Core function applying the universal transformer layer. Args: x: input hparams: model hyper-parameters ffn_unit: feed-forward unit

InitStrings */

RaiseDoubleKeywords */

RaiseNeedMoreValuesToUnpack */

RaiseNoneIterError */

RaiseTooManyValuesToUnpack */

ExtTypeTest */

CheckBinaryVersion */