NMT-Keras

Neural Machine Translation with Keras.

Features

• Attention RNN and Transformer models.
• Online learning and Interactive neural machine translation (INMT). See the interactive NMT branch.
• Tensorboard integration. Training process, models and word embeddings visualization.
• Attention model over the input sequence of annotations. - Supporting Bahdanau (Add) and Luong (Dot) attention mechanisms. - Also supports double stochastic attention.
• Peeked decoder: The previously generated word is an input of the current timestep.
• Beam search decoding. - Featuring length and source coverage normalization.
• Ensemble decoding.
• Translation scoring.
• N-best list generation (as byproduct of the beam search process).
• Support for GRU/LSTM networks: - Regular GRU/LSTM units. - Conditional GRU/LSTM units in the decoder. - Multilayered residual GRU/LSTM networks.
• Unknown words replacement.
• Use of pretrained (Glove or Word2Vec) word embedding vectors.
• MLPs for initializing the RNN hidden and memory state.
• Spearmint wrapper for hyperparameter optimization.
• Client-server architecture for web demos.

Guide

• Installation
  • Requirements
• Usage
  • Training
  • Decoding
  • Scoring
• Configuration options
  • Naming and experiment setup
  • Input/output
  • Evaluation
  • Decoding
  • Search normalization
  • Sampling
  • Unknown words treatment
  • Word representation
  • Text representation
  • Input text
  • Output text
  • Optimization
  • Learning rate schedule
  • Training options
  • Early stop
  • Model main hyperparameters
  • AttentionRNNEncoderDecoder model
  • Transformer model
  • Regularizers
  • Tensorboard
  • Storage and plotting
• Resources
  • Theoretical NMT
  • NMT-Keras Step-by-step
  • NMT-Keras Output
  • Tensorboard integration
• Tutorials
• Modules
  • nmt_keras package
  • data_engine package
  • utils package
• Contact
  • Acknowledgement

Indices and tables

• Index
• Module Index
• Search Page