adjust file paths and license info

preprocess/normalise-romanian.py

@@ -1,9 +1,10 @@
 #!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+# Author: Barry Haddow
+# Distributed under MIT license
 
 #
 # Normalise Romanian s-comma and t-comma
-#
-# author: Barry Haddow
 
 import io
 import sys

preprocess/remove-diacritics.py

@@ -1,9 +1,10 @@
 #!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+# Author: Barry Haddow
+# Distributed under MIT license
 
 #
 # Remove Romanian diacritics. Assumes s-comma and t-comma are normalised
-#
-# author: Barry Haddow
 
 import io
 import sys

r2l/README.md

@@ -12,11 +12,11 @@ use the same vocabulary, and apply reverse.py *after* truecasing/BPE, to simplif
 
 3. at test time, produce an n-best list with the l2r model(s):
 
-  time THEANO_FLAGS=mode=FAST_RUN,floatX=float32,device=gpu,on_unused_input=warn python /path/to/nematus/nmt/translate.py -m model.npz -i test.bpe.de -o test.output.50best -k 50 -n -p 1 --n-best
+  time THEANO_FLAGS=mode=FAST_RUN,floatX=float32,device=gpu,on_unused_input=warn python /path/to/nematus/nematus/translate.py -m model.npz -i test.bpe.de -o test.output.50best -k 50 -n -p 1 --n-best
 
 4. reverse the outputs in the n-best list, and re-score with the r2l model(s).
 
   python reverse_nbest.py < test.output.50best > test.output.50best.reversed
 
-  time THEANO_FLAGS=mode=FAST_RUN,floatX=float32,device=gpu,on_unused_input=warn python /path/to/nematus/nmt/rescore.py -m /path/to/r2l_model/model.npz -s test.bpe.de -i test.output.50best.reversed -o test.output.50best.rescored -b 80 -n
+  time THEANO_FLAGS=mode=FAST_RUN,floatX=float32,device=gpu,on_unused_input=warn python /path/to/nematus/nematus/rescore.py -m /path/to/r2l_model/model.npz -s test.bpe.de -i test.output.50best.reversed -o test.output.50best.rescored -b 80 -n
   python rerank.py < test.output.50best.rescored | python reverse.py > test.output.reranked

r2l/rerank.py

@@ -1,6 +1,7 @@
 #!/usr/bin/python
 # -*- coding: utf-8 -*-
 # Author: Rico Sennrich
+# Distributed under MIT license
 
 import sys
 from collections import defaultdict

r2l/reverse.py

@@ -1,5 +1,7 @@
 #!/usr/bin/python
 # -*- coding: utf-8 -*-
+# Author: Rico Sennrich
+# Distributed under MIT license
 
 import sys
 

r2l/reverse_nbest.py

@@ -1,5 +1,7 @@
 #!/usr/bin/python
 # -*- coding: utf-8 -*-
+# Author: Rico Sennrich
+# Distributed under MIT license
 
 import sys
 

sample/config.py

@@ -2,9 +2,7 @@
 import os
 import sys
 
-sys.path.append('/path/to/nematus/nmt')
-
-from nmt import train
+sys.path.append('/path/to/nematus/nematus')
 
 VOCAB_SIZE = 90000
 SRC = "ro"

sample/preprocess.sh

@@ -1,4 +1,4 @@
-#/bin/sh
+#!/bin/sh
 
 # this sample script preprocesses a sample corpus, including tokenization,
 # truecasing, and subword segmentation. 

sample/translate.sh

@@ -1,12 +1,12 @@
-#/bin/sh
+#!/bin/sh
 
 # theano device
 device=gpu
 
 # path to nematus ( https://www.github.com/rsennrich/nematus )
 nematus=/path/to/nematus
 
-THEANO_FLAGS=mode=FAST_RUN,floatX=float32,device=$device,on_unused_input=warn python $nematus/nmt/translate.py \
+THEANO_FLAGS=mode=FAST_RUN,floatX=float32,device=$device,on_unused_input=warn python $nematus/nematus/translate.py \
      -m model/model.npz \
      -i data/newsdev2016.bpe.ro \
      -o data/newsdev2016.output \

sample/validate.sh

@@ -16,7 +16,7 @@ dev=data/newsdev2016.bpe.ro
 ref=data/newsdev2016.tok.en
 
 # decode
-THEANO_FLAGS=mode=FAST_RUN,floatX=float32,device=$device,on_unused_input=warn python $nematus/nmt/translate.py \
+THEANO_FLAGS=mode=FAST_RUN,floatX=float32,device=$device,on_unused_input=warn python $nematus/nematus/translate.py \
      -m $prefix.dev.npz \
      -i $dev \
      -o $dev.output.dev \