Fix best practice instruction

rsennrich / subword-nmt

Unsupervised Word Segmentation for Neural Machine Translation and Text Generation

MIT License

2.18k stars 464 forks source link

Fix best practice instruction #60

Closed bastings closed 6 years ago

bastings commented 6 years ago

I think there is a mistake in the best practice instructions.

the first apply-bpe output should be saved or piped through
get-vocab should get a temporary training corpus as input that has BPE applied without vocabulary threshold.
get_vocab now takes -i and -o arguments (or stdin)

I fixed it in this PR. Please check it :-)

rsennrich commented 6 years ago

thanks Joost; this got broken in 1b5456906b6d75e2352c95ee416c7596ea18fceb.

I have a slight preference for using a pipe instead of a temp file (less chance of mistakenly reading/writing to the same file), so I committed that in 36bfdd3a7a63502e786e04e6d2d772fb353ee354.

bastings commented 6 years ago

Perfect!