def build(tree_head_dict):
#def build():
pb_instances = propbank_ptb.instances()
all_sub_trees = []
for inst in pb_instances[:]:
if str(inst).find('*') == -1 and str(inst).find(',') == -1:
arguments = []
tree = inst.tree
#tree.draw()
pred_tree = inst.predicate.select(tree)
(pred,r) = parseExpr(str(pred_tree),0,0)
(parsed,r) = parseExpr(str(tree),0,0)
remove_functional_tags(parsed)
#all_sub_trees = all_sub_trees + print_all_subtrees(parsed,[]) # use this function to print all subtrees and then use java program to find heads
for (argloc,argid) in inst.arguments:
if str(argloc.select(tree)).split(')')[0].find('*') == -1:
#print('%s' % (argloc.select(tree).pprint(10000)[:]))
wordNum = int(str(argloc).split(':')[0])
h = int(str(argloc).split(':')[1])
#print wordNum
#print h
arg = traverse_tree_depth(parsed,wordNum,h)
arguments.append((arg,argid))
#for a in arguments:
#print a.data
#pruned = pruning(parsed,pred,inst.predicate.wordnum,[])
#print
t_word = pred.word
t_w_pos = pred.data
#t_word = inst.roleset.split('.')[0]
#t_w_pos = inst.predicate.select(inst.tree).node
pred_parrent = find_pred_parrent(parsed,inst.predicate.wordnum,None)
subcat = find_subcat(pred_parrent)
#print t_word
#print t_w_pos
for (arg,label) in arguments:
path_list = get_path(arg,parsed,inst.predicate.wordnum)
(h,h_pos) = extract_head(arg,tree_head_dict)
path = ''.join(path_list)
distance = len(path_list)
pt = arg.data.rstrip()
t_word_pls_pt = str(t_word)+str(pt)
t_word_pls_h_word = str(t_word)+str(h)
distance_pls_t_word = str(distance)+str(t_word)
subcatStar = find_subcat(arg.parent)
subcatAt = find_subcat(arg)
print 'h='+str(h)+' h_pos='+str(h_pos)+' h_word='+str(h)+' h_word_pos='+str(h_pos)+' path='+str(path)+' t_word_pls_pt='+t_word_pls_pt+' t_word_pls_h_word='+t_word_pls_h_word+' subcat='+str(subcat)+ ' subcatAt='+str(subcatAt)+ ' subcatStar='+str(subcatStar)+ ' '+label
from argext import *
from nltk.corpus import propbank_ptb
from math import floor
from sys import stdout
if __name__ == '__main__' :
print('\n- acquiring experiment data -\n')
# vars
exp_name = 'SemanticArgumentClassification' # the experiment's name
files = [exp_name + '_data_train.arff', exp_name + '_data_dev.arff', exp_name + '_data_test.arff'] # the output filenames
ratios = [0.6, 0.2, 0.2] # their corresponding ratios (60%, 20%, 20%)
pbi_ratio = 1. # ratio of the total PropBank corpus to acquire data from
# init
pbi = propbank_ptb.instances()
featurelist = ['predicate', 'path', 'phraseType', 'position', 'voice', 'class'] # initialize ARGInstanceBuilder with featurelist
arg_the_builder = ARGInstanceBuilder(dict.fromkeys(featurelist))
arglist = [] # arglist for the extracted ARGInstances
# extract ARGInstances
pbi_ratio_index = floor(len(pbi)*pbi_ratio)
for i in range(pbi_ratio_index) :
if (i%20) == 0 :
stdout.write("\rextracting ARGInstances...%.2f%%" % (i*100/pbi_ratio_index))
stdout.flush()
try :
arglist += arg_the_builder.get_arginstances(pbi[i]) # add extracted ARGInstances from current Propbank Instance to arglist
except :
print("Error at PropBankInstance with index : " + str(i))
stdout.write("\rextracting ARGInstances...done \n")
def build(tree_head_dict):
#def build():
pb_instances = propbank_ptb.instances()
all_sub_trees = []
##predicted_context_labels = read_without_context_labels() ##
##f = 0 ##
##t = 0 ##
for inst in pb_instances[:]:
if int(str(inst).split('/')[1]) > 01 and int(str(inst).split('/')[1]) < 22: # for getting training data according to CoNLL 2005 task
#if int(str(inst).split('/')[1]) == 23: # for getting testing data according to CoNLL 2005 task
if str(inst).find('*') == -1 and str(inst).find(',') == -1:
arguments = []
tree = inst.tree
#tree.draw()
pred_tree = inst.predicate.select(tree)
(pred,r) = parseExpr(str(pred_tree),0,0)
(parsed,r) = parseExpr(str(tree),0,0)
#remove_functional_tags(parsed)
#all_sub_trees = all_sub_trees + print_all_subtrees(parsed,[]) # use this function to print all subtrees and then use java program to find heads
gold_context_labels_list = []
for (argloc,argid) in inst.arguments:
if str(argloc.select(tree)).split(')')[0].find('*') == -1:
#print('%s' % (argloc.select(tree).pprint(10000)[:]))
wordNum = int(str(argloc).split(':')[0])
h = int(str(argloc).split(':')[1])
#print wordNum
#print h
arg = traverse_tree_depth(parsed,wordNum,h)
arguments.append((arg,wordNum,argid))
gold_context_labels_list.append(argid)
#for a in arguments:
#print a.data
#pruned = pruning(parsed,pred,inst.predicate.wordnum,[])
#print
t_word = pred.word
t_w_pos = pred.data
#t_word = inst.roleset.split('.')[0]
#t_w_pos = inst.predicate.select(inst.tree).node
pred_parrent = find_pred_parrent(parsed,inst.predicate.wordnum,None)
subcat = find_subcat(pred_parrent)
if pred_parrent.word != None:
ParentWord = pred_parrent.word # word of the parrent node of pred
else:
ParentWord = 'none'
ParentWordPos = pred_parrent.data.rstrip() # pos of the parrent node of pred
## gold context labels
context_labels_all = gold_context_labels_list # for training
#predicted
##t = t + len(arguments) ##
##context_labels_all = predicted_context_labels[f:t] ## for testing
##f = t ##
d = 0
#print len(inst.arguments)
for (arg,wordNum,label) in arguments:
path_list = get_path(arg,parsed,inst.predicate.wordnum)
(h,h_pos) = extract_head(arg,tree_head_dict)
path = ''.join(path_list)
distance = len(path_list)
pt = arg.data.rstrip()
t_word_pls_pt = str(t_word)+str(pt)
t_word_pls_h_word = str(t_word)+str(h)
distance_pls_t_word = str(distance)+str(t_word)
subcatStar = find_subcat(arg.parent)
subcatAt = find_subcat(arg)
if wordNum < inst.wordnum:
position = 'before'
else:
position = 'after'
temp = []
for i in range(0,len(context_labels_all)):
if i != d:
temp.append(context_labels_all[i])
context_labels = ':'.join(temp)
#context_labels = ':'.join([a for a in context_labels_all if context_labels_all.index(a) != d]) ##
d = d + 1 ##
#print ft + ' ' + str(context_labels_all) + ' ' + context_labels + ' ' + str(len(inst.arguments))
#without context labels
#print 't_word='+str(t_word)+' t_w_pos='+str(t_w_pos)+' h_word='+str(h)+' h_word_pos='+str(h_pos)+' path='+str(path)+' t_word_pls_pt='+t_word_pls_pt+' t_word_pls_h_word='+t_word_pls_h_word+' subcat='+str(subcat)+ ' subcatAt='+str(subcatAt)+ ' subcatStar='+str(subcatStar)+' pt='+pt+' position='+position+' ParentWord='+ParentWord+' ParentWordPos='+ParentWordPos+ ' '+label
#print 't_word='+str(t_word)+' t_w_pos='+str(t_w_pos)+' h_word='+str(h)+' h_word_pos='+str(h_pos)+' path='+str(path)+' t_word_pls_pt='+t_word_pls_pt+' t_word_pls_h_word='+t_word_pls_h_word+' subcat='+str(subcat)+ ' subcatAt='+str(subcatAt)+ ' subcatStar='+str(subcatStar)+' pt='+pt+' position='+position+' ParentWord='+ParentWord+' ParentWordPos='+ParentWordPos+ ' ?'
#with context labels
print 't_word='+str(t_word)+' t_w_pos='+str(t_w_pos)+' h_word='+str(h)+' h_word_pos='+str(h_pos)+' path='+str(path)+' t_word_pls_pt='+t_word_pls_pt+' t_word_pls_h_word='+t_word_pls_h_word+' subcat='+str(subcat)+ ' subcatAt='+str(subcatAt)+ ' subcatStar='+str(subcatStar)+' pt='+pt+' position='+position+' ParentWord='+ParentWord+' ParentWordPos='+ParentWordPos+' context_labels='+context_labels+ ' '+label
