def extract_feature(file,func):
for line in file: # format: i j k l nodes splitted by '\t'
# parse line
contents = line.strip().split('\t')
i,j,k,l = map(int,contents[0:4])
if k<=j or i>j or k > l: continue # unvalid sentence
nodes = ['0/null/null/null/0']
nodes.extend(contents[4:])
tree = Generator.build_tree(nodes)
cat = []
for w in range(len(nodes)):
c = 'N'
if w>0:
if w >= i and w <=j: c = 'C'
if w >= k and w <= l: c = 'E'
cat.append(c)
for w in range(1,len(nodes)):
# for every w (word index), we extract context/features for it
context = get_context(w,tree,cat) # extract contexts from the dependency subtree where w belongs to
print 'context of `',str(tree.sentence[w]),'`',context
# feature incremental for feat_dict
func(context,cat[w])
def predict_file(filein, fileout):
for line in filein:
contents = line.strip().split('\t')
nodes = ['0/null/null/null/0']
nodes.extend(contents[4:])
tree = Generator.build_tree(nodes)
if len(tree.sentence) == 0: continue
results = predict_sentence(tree, 3)
assert (len(tree.sentence) == len(results))
for i in range(1,len(tree.sentence)):
w = tree.sentence[i].word
fileout.write('%s/%s ' % (w, results[i]))
print >> fileout
