NMT-Koean-To-English

작성중...

• 한영 기계 번역(Korean-English Machine Translation) 모델 개발 스터디
• PyTorch, koNLPY, NLPY, Gensim package 활용

Requirements

• Python 3.6 (may work with other versions, but I used 3.6)
• PyTorch 1.1.0
• Gensim 3.8.0
• konlpy 0.5.1
• nltk 3.4.4

Datasets

• https://github.com/jungyeul/korean-parallel-corpora 데이터셋 사용

git clone https://github.com/dlcjfgmlnasa/NMT-Koean-To-English.git --recursive

pip install -r requirement.txt

---

목차

1. Sequence to Sequence (Seq2Seq)
2. Sequence to Sequence with Attention (Seq2Seq with Attention)
3. Convolution Sequence to Sequence
4. ByteNet
5. SliceNet
6. Transformer

1. Seq2Seq

• Parameter List

import argparse
parser = argparse.ArgumentParser()
parser.add_argument('--rnn_sequence_size', default=30, type=int)
parser.add_argument('--min_count', default=3, type=int)
parser.add_argument('--max_count', default=10000, type=int)
parser.add_argument('--embedding_size', default=200, type=int)
parser.add_argument('--rnn_dim', default=200, type=int)
parser.add_argument('--rnn_layer', default=3, type=int)
parser.add_argument('--batch_size', default=128, type=int)

• loss

• translation

2. Seq2Seq with Attention

• Parameter List

import argparse
parser = argparse.ArgumentParser()
parser.add_argument('--rnn_sequence_size', default=30, type=int)
parser.add_argument('--min_count', default=3, type=int)
parser.add_argument('--max_count', default=100000, type=int)
parser.add_argument('--embedding_size', default=200, type=int)
parser.add_argument('--rnn_dim', default=123, type=int)
parser.add_argument('--rnn_layer', default=3, type=int)
parser.add_argument('--rnn_dropout_rate', default=0.5, type=float)
parser.add_argument('--use_residual', default=True, type=bool)
parser.add_argument('--attention_method', default='general', choices=['dot', 'general', 'concat'], type=str)
parser.add_argument('--batch_size', default=128, type=int)

3. Convolution Seq2Seq

• 구현 완료 (2019.08.05)

4. ByteNet

5. SliceNet

6. Transformer

Reference

• https://pytorch.org/tutorials/intermediate/seq2seq_translation_tutorial.html
• https://github.com/bentrevett/pytorch-seq2seq
• https://wikidocs.net/24996