Base abstract class that allows the user to customize sampling.

Augmented decoder output that also includes the attention scores.

Structure for the output of a beam search decoder. This class is used to define the output at each step as well as the final output of the decoder.

Output of an RNN decoder. Note that we output both the logits and predictions because during dynamic decoding the predictions may not correspond

Calculates BLEU score using the Moses multi-bleu.perl script.

Augmented decoder output that also includes the attention scores.

Augmented decoder output that also includes the attention scores and copy vals.

Augmented decoder output that also includes the attention scores.

An attention layer that calculates attention scores using a dot product.

State for a single step of beam search. Args: log_probs: The current log probabilities of all beams finished: A boolean vector that specifi

A simple evaluator

A simple evaluator

A simple evaluator

Simple schema which maps table&column to a unique identifier

An RNN Decoder that uses attention over an input sequence. Args: cell: An instance of ` tf.contrib.rnn.RNNCell` helper: An instance of `tf.

An attention layer that calculates attention scores using a parameterized multiplication.

Sequence2Sequence model with attention mechanism. Args: source_vocab_info: An instance of `VocabInfo` for the source vocabulary targe

Simple RNN decoder that performed a softmax operations on the cell output.

Basic Sequence2Sequence model with a unidirectional encoder and decoder. The last encoder state is used to initialize the decoder and thus both mu

The BeamSearchDecoder wraps another decoder to perform beam search instead of greedy selection. This decoder must be used with batch size of 1, whic

Outputs for a single step of beam search. Args: scores: Score for each beam, a float32 vector predicted_ids: predictions for this step step

Calculates BLEU score using the Moses multi-bleu.perl script.

Conditional LSTM with Attention

A deep convolutional encoder, as described in https://arxiv.org/abs/1611.02344. The encoder supports optional positions embeddings. Params:

Turn positive integers (indexes) into denses vectors of fixed size. eg. [[4], [20]] -> [[0.25, 0.1], [0.6, -0.2]] @input_dim

Final outputs returned by the beam search after all decoding is finished. Args: predicted_ids: The final prediction. A tensor of shape `[

A bridge that creates an initial decoder state based on the output of the encoder. This state is created by passing the encoder outputs through an

extract Main Entity from the query

Passes the encoder state through to the decoder as-is. This bridge can only be used if encoder and decoder have the exact same state size, i.e. us

An encoder that pools over embeddings, as described in https://arxiv.org/abs/1611.02344. The encoder supports optional positions embeddings and a

Simple schema which maps table&column to a unique identifier

Simple schema which maps table&column to a unique identifier

Simple schema which maps table&column to a unique identifier

Simple schema which maps table&column to a unique identifier

A collection of options that are passed to the training script and can be saved to perform inference later. Args: task: Name of the training

A bridge that does not pass any information between encoder and decoder and sets the initial decoder state to 0. The input function is not modified.

Reference: http://arxiv.org/abs/1212.5701

Reference: http://arxiv.org/abs/1212.5701

Reference: http://arxiv.org/abs/1412.6980v8 Default parameters follow those provided in the original paper.

aggr type

Tests the `AttentionDecoder` class.

Attention layer according to https://arxiv.org/abs/1409.0473. Params: num_units: Number of units used in the attention layer

Tests the AttentionLayerBahdanau class

Tests the AttentionLayerDot class

Tests the AttentionLayer module.

Base class for Sequence2Sequence model with attention-based copying. Args: source_vocab_info: An instance of `VocabInfo` for the sou

Tests the `BasicDecoder` class.

Configuration object for beam search. Args: beam_width: Number of beams to use, an integer vocab_size: Output vocabulary size eos_token

A bidirectional RNN encoder. Uses the same cell for both the forward and backward RNN. Stacking should be performed as part of the cell. Arg

Tests the BidirectionalRNNEncoder class.

An abstract bridge class. A bridge defines how state is passed between encoder and decoder. All logic is contained in the `_create` method, which

Abstract class for bridge tests

Conditional LSTM with Attention

Interface for all classes that are configurable via a parameters dictionary. Args: params: A dictionary of parameters. mode: A value in t

Tests the ConvEncoder class.

a convolutional layer with max pooling

Tests Vocabulary lookup table operations.

Test for `sqe2seq.losses.sequence_mask`.