1 Source code

First, the source code ：https://github.com/keon/seq2seq
We usually use this machine translation model to learn the source code , But there will be many small technical problems .

2 python -m spacy download en Prompt that the server connection has failed all the time and the installation cannot be installed

The official description of the following command line is not available , There will be an error that the connection is not connected ：

python -m spacy download de
python -m spacy download en

So , You need to use these two packages de_core_news_sm-2.3.0.tar.gz and en_core_web_sm-2.2.5.tar.gz, If you don't have a dial, you can click in and download it by yourself . Close test available .

3 Modify the code

The code needs to be modified in two places ：
utils.py First few lines in ：

import re
import spacy
from torchtext.legacy.data import Field, BucketIterator #  add legacy！！！
from torchtext.legacy.datasets import Multi30k #  add legacy！！！

Then the next few lines change to ：

def load_dataset(batch_size):
    spacy_de = spacy.load('de_core_news_sm') #  The parameter here used to be de
    spacy_en = spacy.load('en_core_web_md') #  The parameter here used to be en
    url = re.compile('(<url>.*</url>)')
    ……

4 pycharm/anaconda+torch+cuda+gpu Environmental Science

I can't match the reference of the environment This blog .

stay win10+anaconda After installation in the environment , stay pycharm Use in conda The virtual environment kernel can be used .

5 effect