def main():
try:
inputset = sys.argv[1]
outputset = sys.argv[2]
except:
print("Syntax: inputset outputset",file=sys.stderr)
sys.exit(2)
reader = Reader(inputset)
writer = Writer(outputset)
for sentencepair in reader:
sentencepair.ref = None
sentencepair.source = None
sentencepair.category = None
writer.write(sentencepair)
writer.close()
reader.close()
def main():
try:
outputset = sys.argv[1]
inputsets = sys.argv[2:]
except:
print("Syntax: outputset inputset inputset2...", file=sys.stderr)
sys.exit(2)
id = 0
writer = Writer(outputset)
for inputset in inputsets:
reader = Reader(inputset)
for sentencepair in reader:
id += 1
sentencepair.id = id
writer.write(sentencepair)
reader.close()
writer.close()
def main():
try:
inputset = sys.argv[1]
outputset = sys.argv[2]
except:
print("Syntax: inputset outputset",file=sys.stderr)
sys.exit(2)
reader = Reader(inputset)
sentencepairs = []
for sentencepair in reader:
sentencepairs.append(sentencepair)
reader.close()
writer = Writer(outputset)
random.shuffle(sentencepairs)
for i, sentencepair in enumerate(sentencepairs):
sentencepair.id = i+1
writer.write(sentencepair)
writer.close()
def main():
if len(sys.argv) < 3 or len(sys.argv) > 5:
print("Syntax: inputset outputset offset maxwords",file=sys.stderr)
sys.exit(2)
try:
inputset, outputset, offset,maxwords = sys.argv[1:]
offset = int(offset)
maxwords = int(maxwords)
except:
maxwords = 99
try:
inputset, outputset, offset = sys.argv[1:]
offset = int(offset)
except:
inputset, outputset = sys.argv[1:]
offset = 1
buffer = []
BUFFERSIZE = 10
tmpfile=False
inputs = set()
if os.path.exists(outputset):
writer = Writer(outputset + '.tmp')
reader = Reader(outputset)
for sentencepair in reader:
inputs.add( hash(sentencepair.input) )
writer.write(sentencepair)
tmpfile=True
else:
writer = Writer(outputset)
num = 0
reader = Reader(inputset)
quit = False
for sentencepair in reader:
if len(sentencepair.input) <= maxwords:
num += 1
if not hash(sentencepair.input) in inputs:
if num >= offset:
buffer.append(sentencepair)
if len(buffer) == BUFFERSIZE:
buffer, quit = processbuffer(buffer, reader,writer, inputs,num-BUFFERSIZE)
if quit: break
if buffer and not quit: processbuffer(buffer, reader,writer, inputs,num)
writer.close()
if tmpfile:
os.rename(outputset+'.tmp',outputset)
def main():
global sources, categories
if len(sys.argv) < 4:
print("Syntax: set L1 L2",file=sys.stderr)
sys.exit(2)
setfile= sys.argv[1]
l1= sys.argv[2]
l2= sys.argv[3]
sentencepairs = []
if os.path.exists(setfile):
print("Loading existing file: ", setfile)
reader = Reader(setfile)
for sentencepair in reader:
sentencepairs.append(sentencepair)
if sentencepair.source:
sources[sentencepair.source] += 1
if sentencepair.category:
categories[sentencepair.category] += 1
print(str(len(sentencepairs)) + " sentences loaded")
else:
print("New file: ", setfile,file=sys.stderr)
print("Type h for help")
cursor = None
quit = False
while not quit:
cmd = input("> ")
if cmd.lower() == 'q':
writer = Writer(setfile,l1,l2)
for sentencepair in sentencepairs:
writer.write(sentencepair)
writer.close()
quit = True
elif cmd.lower() == 'h':
print("q\tSave and quit",file=sys.stderr)
print("n\tNew sentence pair",file=sys.stderr)
#print("d\tDelete sentence pair",file=sys.stderr)
print("a\tAdd alternative",file=sys.stderr)
print(">\tNext sentence pair",file=sys.stderr)
print("<\tPrevious sentence pair",file=sys.stderr)
print("12\tGo to sentence pair #12", file=sys.stderr)
print("w\tWrite changes to disk", file=sys.stderr)
elif cmd.lower() == "<":
if cursor is None:
cursor = len(sentencepairs) - 1
else:
cursor = cursor - 1
if cursor < 0:
cursor = len(sentencepairs) - 1
showsentencepair(sentencepairs, cursor)
elif cmd.lower() == ">":
if cursor is None:
cursor = 0
else:
cursor = cursor + 1
if cursor >= len(sentencepairs):
cursor = 0
showsentencepair(sentencepairs, cursor)
elif cmd.lower().isdigit():
cursor = int(cmd.lower()) - 1
if cursor < 0:
cursor = 0
if cursor >= len(sentencepairs):
cursor = len(sentencepairs) - 1
elif cmd.lower() == 'n':
cursor = newsentencepair(sentencepairs)
elif cmd.lower() == 'w':
writer = Writer(setfile,l1,l2)
for sentencepair in sentencepairs:
writer.write(sentencepair)
writer.close()
elif cmd.lower() == 'p':
if cursor is None:
cursor = 0
showsentencepair(sentencepairs, cursor)
elif cmd.lower() == 'a':
if cursor is None:
cursor = 0
addalternative(sentencepairs[cursor])
else:
print("No such command, type h for help", file=sys.stderr)
def main():
try:
inputset = sys.argv[1]
outputset = sys.argv[2]
l1 = sys.argv[3]
l2 = sys.argv[4]
except:
print("Syntax: inputset outputset l1 l2", file=sys.stderr)
sys.exit(2)
writer = Writer(outputset)
reader = Reader(inputset)
for sentencepair in reader:
if sentencepair.ref:
for left, fragment, right in sentencepair.fragments(sentencepair.ref):
print("Tokenising reference: L=", left, file=sys.stderr)
print(" F=", fragment.value, file=sys.stderr)
print(" R=", right, file=sys.stderr)
if left.strip():
left = tok(left, l2)
else:
left = ""
alts = fragment.alternatives
fragment = Fragment(tok(fragment.value, l2), id=fragment.id)
for alt in alts:
fragment.alternatives.append(Alternative(tok(alt.value, l2)))
if right.strip():
right = tok(right, l2)
else:
right = ""
if left and right:
ref = left + (fragment,) + right
elif left:
ref = left + (fragment,)
elif right:
ref = (fragment,) + right
sentencepair.ref = ref
if sentencepair.output:
for left, fragment, right in sentencepair.fragments(sentencepair.output):
print("Tokenising output: L=", left, file=sys.stderr)
print(" F=", fragment.value, file=sys.stderr)
print(" R=", right, file=sys.stderr)
if left.strip():
left = tok(left, l2)
else:
left = ""
alts = fragment.alternatives
fragment = Fragment(tok(fragment.value, l2))
for alt in alts:
fragment.alternatives.append(Alternative(tok(alt)))
if right.strip():
right = tok(right, l2)
else:
right = ""
if left and right:
out = left + (fragment,) + right
elif left:
out = left + (fragment,)
elif right:
out = (fragment,) + right
sentencepair.output = out
if sentencepair.input:
for left, fragment, right in sentencepair.fragments(sentencepair.input):
print("Tokenising input: L=", left, file=sys.stderr)
print(" F=", fragment.value, file=sys.stderr)
print(" R=", right, file=sys.stderr)
if left.strip():
left = tok(left, l2)
else:
left = ""
alts = fragment.alternatives
fragment = Fragment(tok(fragment.value, l1), id=fragment.id)
if right.strip():
right = tok(right, l2)
else:
right = ""
if left and right:
inp = left + (fragment,) + right
elif left:
inp = left + (fragment,)
elif right:
inp = (fragment,) + right
sentencepair.input = inp
writer.write(sentencepair)
reader.close()
writer.close()
def generate(testoutput, ttablefile, gizamodelfile_s2t, gizamodelfile_t2s, patternmodelfile_source, patternmodelfile_target, classfile_source, classfile_target, size =0, joinedprobabilitythreshold = 0.01, divergencefrombestthreshold=0.8,DEBUG = False):
if size > 0:
print("Extracting instances, writing to " + testoutput + '.tmp',file=sys.stderr)
writer = Writer(testoutput+'.tmp')
else:
print("Extracting instances, writing to " + testoutput,file=sys.stderr)
writer = Writer(testoutput)
prevsentence = -1
id = 0
for sourcepattern, targetpattern, sourceoffset, targetoffset, sourcesentence, targetsentence, sentence in extractpairs(ttablefile, gizamodelfile_s2t, gizamodelfile_t2s, patternmodelfile_source, patternmodelfile_target, classfile_source, classfile_target, joinedprobabilitythreshold, divergencefrombestthreshold, DEBUG):
id += 1
if sentence != prevsentence:
print(datetime.datetime.now().strftime('%H:%M:%S'), "Input sentence #" + str(sentence) + " , Output sentence #" + str(id), file=sys.stderr)
prevsentence = sentence
valid, sentencepair = makesentencepair(id, sourcepattern, targetpattern, sourceoffset, targetoffset, sourcesentence, targetsentence)
if valid:
writer.write(sentencepair)
writer.close()
if size > 0:
print("Sampling " + str(size),file=sys.stderr)
selected_ids = set(random.sample( range(1,id+1), size ))
writer = Writer(testoutput)
reader = Reader(testoutput+'.tmp')
newid = 0
for sentencepair in reader:
if int(sentencepair.id) in selected_ids:
newid += 1
sentencepair.id = newid
writer.write(sentencepair)
reader.close()
writer.close()
def main():
parser = argparse.ArgumentParser(description="Colibrita - Translation Assistance using Moses", formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument('-f','--dataset', type=str,help="Dataset file", action='store',default="",required=True)
parser.add_argument('--debug','-d', help="Debug", action='store_true', default=False)
parser.add_argument('-o','--output',type=str,help="Output prefix", required = True)
parser.add_argument('-T','--ttable', type=str,help="Phrase translation table (file) to use when testing with --lm and without classifier training", action='store',required=True)
parser.add_argument('--lm',type=str, help="Language model (file in ARPA format, as produced by for instance SRILM)", action='store',required=True)
parser.add_argument('--lmweight',type=float, help="Language model weight for Moses ", action='store',default=0.5)
parser.add_argument('--lmorder',type=float, help="Language model order", action='store',default=3)
parser.add_argument('--dweight',type=float, help="Distortion weight for Moses", action='store',default=0.6)
parser.add_argument('--tmweights',type=str, help="Translation model weights for Moses (comma separated)", action='store',default="0.20,0.20,0.20,0.20,0.20")
parser.add_argument('--lmweightrr',type=float, help="Language model weight in reranking", action='store',default=1)
parser.add_argument('--tweightrr',type=float, help="Translation model weight in reranking", action='store',default=1)
parser.add_argument('-n','--n',type=int,help="Number of output hypotheses per sentence", default=25)
parser.add_argument('-a','--a',type=int,help="Add alternative translations, up to the specified numer", default=0)
args = parser.parse_args()
#if os.path.exists(args.output):
# print("Output already " + args.output + " already exists, doing nothing..",file=sys.stderr)
# sys.exit(2)
#else:
# os.mkdir(args.output)
if not os.path.exists(args.ttable):
print("Translation table " + args.ttable + " does not exist", file=sys.stderr)
sys.exit(2)
if not os.path.exists(args.lm):
print("Language model " + args.lm + " does not exist", file=sys.stderr)
sys.exit(2)
data = Reader(args.dataset)
f = open(args.output + '.moses.ini','w',encoding='utf-8')
f.write("[input-factors]\n0\n\n")
f.write("[mapping]\n0 T 0\n\n")
f.write("[ttable-file]\n0 0 0 5 " + args.ttable + "\n\n")
f.write("[lmodel-file]\n0 0 " + str(args.lmorder) + " " + args.lm + "\n\n")
f.write("[ttable-limit]\n20\n\n")
f.write("[weight-d]\n" + str(args.dweight) + "\n\n")
f.write("[weight-l]\n" + str(args.lmweight) + "\n\n")
f.write("[weight-t]\n" + "\n".join(args.tmweights.split(',')) + "\n\n")
f.write("[weight-w]\n-1\n")
f.write("[distortion-limit]\n6\n")
f.close()
if not os.path.exists(args.output + ".nbestlist"):
cmd = 'moses -f ' + args.output + '.moses.ini -n-best-list ' + args.output + '.nbestlist ' + str(args.n)
print("Calling moses: " + cmd,file=sys.stderr)
p = subprocess.Popen(cmd,shell=True,stdout=subprocess.PIPE,stdin=subprocess.PIPE,stderr=subprocess.PIPE)
for sentencepair in data:
for left, sourcefragment, right in sentencepair.inputfragments():
p.stdin.write( (str(sourcefragment) + "\n").encode('utf-8'))
p.communicate()
p.stdin.close()
data.reset()
else:
print("Moses output already exists, not overwriting. Delete " + args.output + ".nbestlist if you want a fresh run.",file=sys.stderr)
print("Loading Language model", file=sys.stderr)
lm = ARPALanguageModel(args.lm)
print("Processing moses output...",file=sys.stderr)
previndex = -1
sentenceoutput = []
hypotheses = []
with open(args.output+'.nbestlist','r',encoding='utf-8') as f:
for line in f:
fields = [ x.strip() for x in line.strip().split("|||") ]
print(fields,file=sys.stderr)
index = int(fields[0])
if index != previndex:
if hypotheses:
sentenceoutput.append( hypotheses )
hypotheses = []
previndex = index
solution = fields[1]
rawscores = fields[2].split(' ')
print(rawscores,file=sys.stderr)
tscore = float(rawscores[9])
hypotheses.append( (solution, tscore) )
sentenceoutput.append( hypotheses ) #don't forget last one
writer = Writer(args.output + '.output.xml')
for i, sentencepair in enumerate(data):
sentencepair.output = copy(sentencepair.input)
hypotheses = sentenceoutput[i]
for left, inputfragment, right in sentencepair.inputfragments():
candidatesentences = []
bestlmscore = -999999999
besttscore = -999999999
for hypothesis, tscore in hypotheses:
#compute new lm score
outputfragment = Fragment(tuple(hypothesis.split(' ')), inputfragment.id)
candidatesentence = sentencepair.replacefragment(inputfragment, outputfragment, sentencepair.output)
lminput = " ".join(sentencepair._str(candidatesentence)).split(" ") #joining and splitting deliberately to ensure each word is one item
lmscore = lm.score(lminput)
assert lmscore <= 0
if lmscore > bestlmscore:
bestlmscore = lmscore
if tscore > besttscore:
besttscore = tscore
candidatesentences.append( ( candidatesentence, hypothesis, tscore, lmscore ) )
#compute scores
solutions = []
for candidatesentence, targetpattern, tscore, lmscore in candidatesentences:
tscore = args.tweightrr * (tscore-besttscore)
lmscore = args.lmweightrr * (lmscore-bestlmscore)
score = tscore + lmscore
print(targetpattern + " --- tscore=" + str(tscore) + ", lmscore=" + str(lmscore),file=sys.stderr)
solutions.append( (score, targetpattern) )
solutions = sorted(solutions, key=lambda x: -1 * x[0])
translation = tuple(solutions[0][1].split())
outputfragment = Fragment(translation, inputfragment.id)
print("\t" + str(inputfragment) + " -> " + str(outputfragment), file=sys.stderr)
if args.a:
for score, solution in solutions[1:1+args.a]:
outputfragment.alternatives.append( Alternative( tuple(solution.split()), confidence=score) )
sentencepair.output = sentencepair.replacefragment(inputfragment, outputfragment, sentencepair.output)
writer.write(sentencepair)
break #only support one iteration for now, one fragment per sentence
writer.close()
print("All done.", file=sys.stderr)