216 lines
6.7 KiB
Python
216 lines
6.7 KiB
Python
import random
|
|
import re
|
|
|
|
try: input = raw_input
|
|
except NameError: pass
|
|
|
|
|
|
class Key:
|
|
def __init__(self, word, weight, decomps):
|
|
self.word = word
|
|
self.weight = weight
|
|
self.decomps = decomps
|
|
|
|
|
|
class Decomp:
|
|
def __init__(self, parts, save, reasmbs):
|
|
self.parts = parts
|
|
self.save = save
|
|
self.reasmbs = reasmbs
|
|
self.next_reasmb_index = 0
|
|
|
|
|
|
class Eliza:
|
|
def __init__(self):
|
|
self.initials = []
|
|
self.finals = []
|
|
self.quits = []
|
|
self.pres = {}
|
|
self.posts = {}
|
|
self.synons = {}
|
|
self.keys = {}
|
|
self.memory = []
|
|
|
|
def load(self, path):
|
|
key = None
|
|
decomp = None
|
|
with open(path) as file:
|
|
for line in file:
|
|
if not line.strip():
|
|
continue
|
|
tag, content = [part.strip() for part in line.split(':')]
|
|
if tag == 'initial':
|
|
self.initials.append(content)
|
|
elif tag == 'final':
|
|
self.finals.append(content)
|
|
elif tag == 'quit':
|
|
self.quits.append(content)
|
|
elif tag == 'pre':
|
|
parts = content.split(' ')
|
|
self.pres[parts[0]] = parts[1:]
|
|
elif tag == 'post':
|
|
parts = content.split(' ')
|
|
self.posts[parts[0]] = parts[1:]
|
|
elif tag == 'synon':
|
|
parts = content.split(' ')
|
|
self.synons[parts[0]] = parts
|
|
elif tag == 'key':
|
|
parts = content.split(' ')
|
|
word = parts[0]
|
|
weight = int(parts[1]) if len(parts) > 1 else 1
|
|
key = Key(word, weight, [])
|
|
self.keys[word] = key
|
|
elif tag == 'decomp':
|
|
parts = content.split(' ')
|
|
save = False
|
|
if parts[0] == '$':
|
|
save = True
|
|
parts = parts[1:]
|
|
decomp = Decomp(parts, save, [])
|
|
key.decomps.append(decomp)
|
|
elif tag == 'reasmb':
|
|
parts = content.split(' ')
|
|
decomp.reasmbs.append(parts)
|
|
|
|
def _match_decomp_r(self, parts, words, results):
|
|
if not parts and not words:
|
|
return True
|
|
if not parts or (not words and parts != ['*']):
|
|
return False
|
|
if parts[0] == '*':
|
|
for index in range(len(words), -1, -1):
|
|
results.append(words[:index])
|
|
if self._match_decomp_r(parts[1:], words[index:], results):
|
|
return True
|
|
results.pop()
|
|
return False
|
|
elif parts[0].startswith('@'):
|
|
root = parts[0][1:]
|
|
if not root in self.synons:
|
|
raise ValueError("Unknown synonym root {}".format(root))
|
|
if not words[0].lower() in self.synons[root]:
|
|
return False
|
|
results.append([words[0]])
|
|
return self._match_decomp_r(parts[1:], words[1:], results)
|
|
elif parts[0].lower() != words[0].lower():
|
|
return False
|
|
else:
|
|
return self._match_decomp_r(parts[1:], words[1:], results)
|
|
|
|
def _match_decomp(self, parts, words):
|
|
results = []
|
|
if self._match_decomp_r(parts, words, results):
|
|
return results
|
|
return None
|
|
|
|
def _next_reasmb(self, decomp):
|
|
index = decomp.next_reasmb_index
|
|
result = decomp.reasmbs[index % len(decomp.reasmbs)]
|
|
decomp.next_reasmb_index = index + 1
|
|
return result
|
|
|
|
def _reassemble(self, reasmb, results):
|
|
output = []
|
|
for reword in reasmb:
|
|
if not reword:
|
|
continue
|
|
if reword[0] == '(' and reword[-1] == ')':
|
|
index = int(reword[1:-1])
|
|
if index < 1 or index > len(results):
|
|
raise ValueError("Invalid result index {}".format(index))
|
|
insert = results[index - 1]
|
|
for punct in [',', '.', ';']:
|
|
if punct in insert:
|
|
insert = insert[:insert.index(punct)]
|
|
output.extend(insert)
|
|
else:
|
|
output.append(reword)
|
|
return output
|
|
|
|
def _sub(self, words, sub):
|
|
output = []
|
|
for word in words:
|
|
word_lower = word.lower()
|
|
if word_lower in sub:
|
|
output.extend(sub[word_lower])
|
|
else:
|
|
output.append(word)
|
|
return output
|
|
|
|
def _match_key(self, words, key):
|
|
for decomp in key.decomps:
|
|
results = self._match_decomp(decomp.parts, words)
|
|
if results is None:
|
|
continue
|
|
results = [self._sub(words, self.posts) for words in results]
|
|
reasmb = self._next_reasmb(decomp)
|
|
if reasmb[0] == 'goto':
|
|
goto_key = reasmb[1]
|
|
if not goto_key in self.keys:
|
|
raise ValueError("Invalid goto key {}".format(goto_key))
|
|
return self._match_key(words, self.keys[goto_key])
|
|
output = self._reassemble(reasmb, results)
|
|
if decomp.save:
|
|
self.memory.append(output)
|
|
continue
|
|
return output
|
|
return None
|
|
|
|
def respond(self, text):
|
|
if text.lower() in self.quits:
|
|
return None
|
|
|
|
text = re.sub(r'\s*\.+\s*', ' . ', text)
|
|
text = re.sub(r'\s*,+\s*', ' , ', text)
|
|
text = re.sub(r'\s*;+\s*', ' ; ', text)
|
|
|
|
words = [w for w in text.split(' ') if w]
|
|
|
|
words = self._sub(words, self.pres)
|
|
|
|
keys = [self.keys[w.lower()] for w in words if w.lower() in self.keys]
|
|
keys = sorted(keys, key=lambda k: -k.weight)
|
|
|
|
output = None
|
|
|
|
for key in keys:
|
|
output = self._match_key(words, key)
|
|
if output:
|
|
break
|
|
if not output:
|
|
if self.memory:
|
|
index = random.randrange(len(self.memory))
|
|
output = self.memory.pop(index)
|
|
else:
|
|
output = self._next_reasmb(self.keys['xnone'].decomps[0])
|
|
|
|
return " ".join(output)
|
|
|
|
def initial(self):
|
|
return random.choice(self.initials)
|
|
|
|
def final(self):
|
|
return random.choice(self.finals)
|
|
|
|
def run(self):
|
|
print(self.initial())
|
|
|
|
while True:
|
|
sent = input('> ')
|
|
|
|
output = self.respond(sent)
|
|
if output is None:
|
|
break
|
|
|
|
print(output)
|
|
|
|
print(self.final())
|
|
|
|
|
|
def main():
|
|
eliza = Eliza()
|
|
eliza.load('doctor.txt')
|
|
eliza.run()
|
|
|
|
if __name__ == '__main__':
|
|
main() |