This repository contains an implementation of a Seq2seq neural network model for machine translation. More details on sequence to sequence machine translation and hyperparameter tuning may be found in Massive Exploration of Neural Machine Translation Architectures.
To train our model, clone the repo and run main.py:
usage: main.py [-h] [--lr N] [--hs N] [--emb N] [--nlayers N] [--dp N]
[--unidir] [--attn STR] [--reverse_input] [-v N] [-b N]
[--epochs N] [--model DIR] [-e] [--visualize] [--predict DIR]
[--predict_outfile DIR] [--predict_from_input STR]
Machine Translation with Attention
optional arguments:
-h, --help show this help message and exit
--lr N learning rate, default: 2e-3
--hs N size of hidden state, default: 300
--emb N embedding size, default: 300
--nlayers N number of layers in rnn, default: 2
--dp N dropout probability, default: 0.30
--unidir use unidirectional encoder, default: bidirectional
--attn STR attention: dot-product, additive or none, default:
dot-product
--reverse_input reverse input to encoder, default: False
-v N vocab size, use 0 for maximum size, default: 0
-b N batch size, default: 64
--epochs N number of epochs, default: 50
--model DIR path to model, default: None
-e, --evaluate only evaluate model, default: False
--visualize visualize model attention distribution
--predict DIR directory with final input data for predictions,
default: None
For example, to train with the default parameters, run:
python main.py
Note: To generate smaller or larger versions of the training dataset for experimenting with the model, you may use the following commands.
cp .data/iwslt/de-en/train.de-en-full.en .data/iwslt/de-en/train.de-en.en && cp .data/iwslt/de-en/train.de-en-full.de .data/iwslt/de-en/train.de-en.decp .data/iwslt/de-en/train.de-en-small.en .data/iwslt/de-en/train.de-en.en && cp .data/iwslt/de-en/train.de-en-small.de .data/iwslt/de-en/train.de-en.de
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent