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 x n, Tensors. output_size: int,

Initialize the parameters for a NAS cell. Args: num_units: int, The number of units in the NAS cell num_proj: (optional) int, The out

Get a sharded variable concatenated into one tensor.

Run the actual computation of one step LSTM. Args: freq_inputs: list of Tensors, 2D, [batch, feature_size]. block: int, current frequ

Make the frequency features. Args: input_feat: input Tensor, 2D, [batch, num_units]. slice_offset: (optional) Python int, default 0,

Convert the given sentence into word indices and masks. WARNING: Unknown words (not in vocabulary) are revmoed. Args:

Create a cell with an added input embedding. Args: cell: an RNNCell, an embedding will be put before its inputs. embedding_classes: i

Produce `out`, s.t. out(i, j) = data(indices(i), i, j).

Slices inputs into groups to prepare for processing by cell's groups Args: inputs: cell input or it's previous state, a Tenso

Time reverse the provided tensor or list of tensors. Assumes the top dimension is the time dimension. Args: t: 3D tensor or list of 2D

borrowed this implementation from @karpathy's neuraltalk.

Tokenization/string cleaning for strings. Taken from https://github.com/yoonkim/CNN_sentence/blob/master/process_data.py

Split the given sentence (str) and enumerate the words as strs. Each word is normalized, i.e. lower-cased, non-alphabet characters

Create a RNN cell composed sequentially of a number of RNNCells. Args: cells: list of RNNCells that will be composed in this order. s

Run this LSTM on inputs, starting from the given state. This method must be implemented by subclasses and does the actual work of calling the

convolution: Args: args: a Tensor or a list of Tensors of dimension 3D, 4D or 5D, batch x n, Tensors. filter_size: int tuple of filter h

Compute the cycle ratio in the dtype of the time.

Get a list of sharded variables with the given dtype.

r"""Computes the LSTM cell forward propagation for 1 time step. This implementation uses 1 weight matrix and 1 bias vector, and there's an option

Make the frequency features. Args: input_feat: input Tensor, 2D, batch x num_units. Returns: A list of frequency features, with

Modulo function that propagates x gradients.

Returns an exponential distribution initializer. Args: minval: float or a scalar float Tensor. With value > 0. Lower bound of the range

Generates a batch iterator for a dataset.

Remove unexpected character in root.

Convert sentence consisting of string to indexed sentence.

Return question string for given key.

Iterate caption strings associated in the vid/gifs.

Return question string for given key.