mastues / daikon

Simple encoder-decoder neural machine translation written in tensorflow

daikon

A simple encoder-decoder model based on recurrent neural networks (RNNs) for machine translation. Supports model training and translation with trained models.

daikon is derived from romanesco written by Samuel Läubli.

Installation

Make sure you have an NVIDIA GPU at your disposal, with all drivers and CUDA installed. Make sure you also have python >= 3.5, pip and git installed, and run

git clone https://github.com/ZurichNLP/daikon
cd daikon
pip install --user -e .

If you have sudo privileges and prefer to install daikon for all users on your system, omit the --user flag. The -e flag installs the app in “editable mode”, meaning you can change source files (such as daikon/constants.py) at any time.

Model training

Models are trained from plaintext files with one sentence per line. Symbols – e.g., words or characters – are delimited by blanks. You need to specify two parallel files: one in the source language, and one in the target language.

Example file (word-level):

I love the people of Iowa .
So that 's the way it is .
Very simple .

Example file (character-level):

I <blank> l o v e <blank> t h e <blank> p e o p l e <blank> o f <blank> I o w a .
S o <blank> t h a t &apos; s <blank> t h e <blank> w a y <blank> i t <blank> i s .
V e r y <blank> s i m p l e .

daikon doesn't preprocess training data. If you want to train a model on lowercased input, for example, you'll need to lowercase the training data yourself.

To train a model from source.txt and target.txt using GPU 0, run

CUDA_VISIBLE_DEVICES=0 daikon train --source source.txt --target target.txt

By default, the trained model and vocabulary will be stored in a directory called model, and logs (for monitoring with Tensorboard) in logs. You can use custom destinations through the -m and -l command line arguments, respectively. Folders will be created if they don't exist.

Some hyperparameters can be adjusted from the command line; run daikon train -h for details. Other hyperparameters are currently hardcoded in daikon/constants.py.

Translation

A trained model can be used to translate new text. To translate a string on GPU 0 run

CUDA_VISIBLE_DEVICES=0 echo "Here is a sample input text" | daikon translate

This assumes there is a folder called model in your current working directory, containing a model trained with daikon (see above). If your model is stored somewhere else, use the -m command line argument.

For further options, run daikon translate -h.

Scoring

Finally, daikon can score existing translations (pairs of source and target sentences):

CUDA_VISIBLE_DEVICES=0 daikon score --source source.txt --target target.txt 

Assuming, again, that there is a folder called model in your current working directory that contains a trained model.

For further options, run daikon score -h.