def parse(sentences, weights):
results = []
edge_feats_list, featdict = c.extract(sentences)
size = len(sentences)
count = 0
correct = 0
total = 0
for sentence, edge_feats in zip(sentences, edge_feats_list):
count = count + 1
Log.i("Parsing sentence: %d of %d" % (count, size))
result = []
heads = eisner.parse(sentence, weights, edge_feats)
for index, word in enumerate(sentence):
if index == 0:
continue
token = [
str(word.id),
word.form,
word.lemma,
word.cpostag,
word.postag,
word.feats,
str(heads[index]),
word.deprel
]
line = "\t".join(token)
if heads[index] == word.head:
correct = correct + 1
total = total + 1
result.append(line)
results.append(result)
return results, float(correct / total)
def main(args):
if len(args) < 2:
Log.w("Arguments Error")
return
path = args[0]
iteration = int(args[1])
Log.i("Training file: %s" % path)
Log.i("----------------")
sentences = c.readconllfile(path)
featdict, weights = train(sentences, iteration)
c.save(featdict, weights)
def main(args):
if len(args) < 2:
Log.w("Arguments Error")
return
path = args[0]
mfile = args[1]
Log.i("Training file: %s" % path)
Log.i("----------------")
sentences = c.readconllfile(path)
Log.i("Model file: %s" % mfile)
Log.i("----------------")
featdict, weights = c.load(mfile)
output, score = parse(sentences, weights)
for each in output:
for token in each:
print(token)
print()
Log.i("[DONE] accuracy: {:.2%}".format(score))
def train(sentences, iteration=10):
edge_feats_list, featdict = c.extract(sentences)
weights = np.zeros(len(featdict))
size = len(sentences)
for i in range(1, iteration + 1):
Log.i("Training iteration: %d of %d" % (i, iteration))
count = 0
for sentence, edge_feats in zip(sentences, edge_feats_list):
count = count + 1
Log.i("\tTraining sentence: %d of %d" % (count, size))
predicts = eisner.parse(sentence, weights, edge_feats)
weights = update(weights, predicts, sentence, edge_feats)
"""
print(predicts)
for feat, index in featdict.items():
print(index, feat, weights[index])
"""
return featdict, weights
