66 lines
1.5 KiB
Python
66 lines
1.5 KiB
Python
|
import jsgf
|
||
|
from os import listdir
|
||
|
from os.path import isfile, join
|
||
|
|
||
|
|
||
|
gram_dir = './gramatyki/'
|
||
|
grammar_files = [file for file in listdir(gram_dir) if isfile(join(gram_dir, file))]
|
||
|
|
||
|
grammars = []
|
||
|
|
||
|
for grammarFile in grammar_files:
|
||
|
grammar = jsgf.parse_grammar_file(gram_dir + grammarFile)
|
||
|
grammars.append(grammar)
|
||
|
|
||
|
lines = []
|
||
|
|
||
|
data_dir = './data/'
|
||
|
data_files = [file for file in listdir(data_dir) if isfile(join(data_dir, file))]
|
||
|
|
||
|
detected = 0
|
||
|
|
||
|
for file in data_files:
|
||
|
f = open(f'{data_dir}{file}', "r")
|
||
|
for line in f:
|
||
|
sep = line.split('\t')
|
||
|
if sep[0] == 'user':
|
||
|
lines.append([sep[1], sep[2]])
|
||
|
|
||
|
|
||
|
def get_dialog_act(rule):
|
||
|
slots = []
|
||
|
get_slots(rule.expansion, slots)
|
||
|
global detected
|
||
|
detected = detected + 1
|
||
|
return {'act': rule.grammar.name, 'slots': slots}
|
||
|
|
||
|
|
||
|
def get_slots(expansion, slots):
|
||
|
if expansion.tag != '':
|
||
|
slots.append((expansion.tag, expansion.current_match))
|
||
|
return
|
||
|
|
||
|
for child in expansion.children:
|
||
|
get_slots(child, slots)
|
||
|
|
||
|
if not expansion.children and isinstance(expansion, jsgf.NamedRuleRef):
|
||
|
get_slots(expansion.referenced_rule.expansion, slots)
|
||
|
|
||
|
|
||
|
def nlu(utterance):
|
||
|
matched = None
|
||
|
for _grammar in grammars:
|
||
|
matched = _grammar.find_matching_rules(utterance)
|
||
|
if matched:
|
||
|
break
|
||
|
|
||
|
if matched:
|
||
|
return get_dialog_act(matched[0])
|
||
|
else:
|
||
|
return {'act': 'null', 'slots': []}
|
||
|
|
||
|
|
||
|
for line in lines:
|
||
|
nlu(line[0])
|
||
|
|
||
|
print(f'{round((detected/len(lines)*100),2)}%')
|