def __init__(self, session, corpus_dir, knbase_dir, result_dir, result_file):
"""
Args:
session: The TensorFlow session.
corpus_dir: Name of the folder storing corpus files and vocab information.
knbase_dir: Name of the folder storing data files for the knowledge base.
result_dir: The folder containing the trained result files.
result_file: The file name of the trained model.
"""
self.session = session
# Prepare data and hyper parameters
print("# Prepare dataset placeholder and hyper parameters ...")
tokenized_data = TokenizedData(corpus_dir=corpus_dir, training=False)
self.knowledge_base = KnowledgeBase()
self.knowledge_base.load_knbase(knbase_dir)
self.session_data = SessionData()
self.hparams = tokenized_data.hparams
self.src_placeholder = tf.placeholder(shape=[None], dtype=tf.string)
src_dataset = tf.data.Dataset.from_tensor_slices(self.src_placeholder)
self.infer_batch = tokenized_data.get_inference_batch(src_dataset)
# Create model
print("# Creating inference model ...")
self.model = ModelCreator(training=False, tokenized_data=tokenized_data,
batch_input=self.infer_batch)
# Restore model weights
print("# Restoring model weights ...")
self.model.saver.restore(session, os.path.join(result_dir, result_file))
self.session.run(tf.tables_initializer())
def __init__(self,
session,
corpus_dir,
knbase_dir,
result_dir,
hparams_dir=None):
self.session = session
hparams = HParams(hparams_dir).hparams if hparams_dir else None
# Prepare data and hyper parameters
print("# Prepare dataset placeholder and hyper parameters ...")
self.tokenized_data = TokenizedData(corpus_dir=corpus_dir,
hparams=hparams,
knbase_dir=knbase_dir,
training=False)
self.hparams = self.tokenized_data.hparams
self.src_placeholder = tf.placeholder(shape=[None], dtype=tf.string)
src_dataset = tf.contrib.data.Dataset.from_tensor_slices(
self.src_placeholder)
self.infer_batch = self.tokenized_data.get_inference_batch(src_dataset)
# Create model
print("# Creating inference model ...")
self.model = ModelCreator(training=False,
tokenized_data=self.tokenized_data,
batch_input=self.infer_batch)
latest_ckpt = tf.train.latest_checkpoint(result_dir)
print("# Restoring model weights ...")
self.model.saver.restore(session, latest_ckpt)
self.session.run(tf.tables_initializer())
def __init__(self, session, corpus_dir, knbase_dir, result_dir,
result_file):
self.session = session
# Prepare data and hyper parameters
print("# Prepare dataset placeholder and hyper parameters ...")
self.tokenized_data = TokenizedData(corpus_dir=corpus_dir,
knbase_dir=knbase_dir,
training=False)
self.hparams = self.tokenized_data.hparams
self.src_placeholder = tf.placeholder(shape=[None], dtype=tf.string)
src_dataset = tf.data.Dataset.from_tensor_slices(self.src_placeholder)
self.infer_batch = self.tokenized_data.get_inference_batch(src_dataset)
# Create model
print("# Creating inference model ...")
self.model = ModelCreator(training=False,
tokenized_data=self.tokenized_data,
batch_input=self.infer_batch)
# Restore model weights
print("# Restoring model weights ...")
self.model.saver.restore(session, os.path.join(result_dir,
result_file))
self.session.run(tf.tables_initializer())
def __init__(self, corpus_dir):
self.graph = tf.Graph()
with self.graph.as_default():
tokenized_data = TokenizedData(corpus_dir=corpus_dir)
self.hparams = tokenized_data.hparams
self.train_batch = tokenized_data.get_training_batch() # Return BatchInput namedtuple from .chatbot/tokenizedata.py
self.model = ModelCreator(training=True, tokenized_data=tokenized_data,
batch_input=self.train_batch)
def __init__(self, corpus_dir):
self.graph = tf.Graph()
with self.graph.as_default():
tokenized_data = TokenizedData(corpus_dir=corpus_dir)
self.hparams = tokenized_data.hparams
self.train_batch = tokenized_data.get_training_batch()
self.model = ModelCreator(training=True, tokenized_data=tokenized_data,
batch_input=self.train_batch)
def __init__(self, session, corpus_dir, knbase_dir, result_dir, aiml_dir,
result_file):
"""
Args:
session: The TensorFlow session.
corpus_dir: Name of the folder storing corpus files and vocab information.
knbase_dir: Name of the folder storing data files for the knowledge base.
result_dir: The folder containing the trained result files.
result_file: The file name of the trained model.
"""
self.session = session
# Prepare data and hyper parameters
print("# Prepare dataset placeholder and hyper parameters ...")
tokenized_data = TokenizedData(corpus_dir=corpus_dir, training=False)
self.knowledge_base = KnowledgeBase()
self.knowledge_base.load_knbase(knbase_dir)
self.session_data = SessionData()
self.hparams = tokenized_data.hparams
self.src_placeholder = tf.placeholder(shape=[None], dtype=tf.string)
src_dataset = tf.data.Dataset.from_tensor_slices(self.src_placeholder)
self.infer_batch = tokenized_data.get_inference_batch(src_dataset)
# Create Retrival model
self.kmodel = aiml.Kernel()
brain_file_name = os.path.join(aiml_dir, BRAIN_FILE)
print(aiml_dir)
# Restore model rules
if os.path.exists(brain_file_name):
print("# Loading from brain file ... ")
self.kmodel.loadBrain(brain_file_name)
else:
print("# Parsing aiml files ...")
aimls_file_name = os.path.join(aiml_dir, AIMLS_FILE)
self.kmodel.bootstrap(learnFiles=os.path.abspath(aimls_file_name),
commands="load aiml b")
print("# Saving brain file: " + BRAIN_FILE)
self.kmodel.saveBrain(brain_file_name)
# Create Generative model
print("# Creating inference model ...")
self.model = ModelCreator(training=False,
tokenized_data=tokenized_data,
batch_input=self.infer_batch)
# Restore model weights
print("# Restoring model weights ...")
self.model.saver.restore(session, os.path.join(result_dir,
result_file))
self.session.run(tf.tables_initializer())
def __init__(self, corpus_dir):
"""
Constructor of the BotTrainer.
Args:
corpus_dir: The folder to save all the training related data.
"""
self.graph = tf.Graph()
with self.graph.as_default():
tokenized_data = TokenizedData(corpus_dir=corpus_dir)
self.hparams = tokenized_data.hparams
self.train_batch = tokenized_data.get_training_batch()
self.model = ModelCreator(training=True, tokenized_data=tokenized_data,
batch_input=self.train_batch)
class BotPredictor(object):
def __init__(self, session, corpus_dir, knbase_dir, result_dir,
result_file):
self.session = session
# Prepare data and hyper parameters
print("# Prepare dataset placeholder and hyper parameters ...")
self.tokenized_data = TokenizedData(corpus_dir=corpus_dir,
knbase_dir=knbase_dir,
training=False)
self.hparams = self.tokenized_data.hparams
self.src_placeholder = tf.placeholder(shape=[None], dtype=tf.string)
src_dataset = tf.data.Dataset.from_tensor_slices(self.src_placeholder)
self.infer_batch = self.tokenized_data.get_inference_batch(src_dataset)
# Create model
print("# Creating inference model ...")
self.model = ModelCreator(training=False,
tokenized_data=self.tokenized_data,
batch_input=self.infer_batch)
# Restore model weights
print("# Restoring model weights ...")
self.model.saver.restore(session, os.path.join(result_dir,
result_file))
self.session.run(tf.tables_initializer())
def predict(self, sentence, html_format=False):
if sentence.strip() == '':
return "Don't you want to say something to me?"
pat_matched, new_sentence, num_list = \
FunctionData.check_arithmetic_pattern_and_replace(sentence)
for pre_time in range(2):
tokens = nltk.word_tokenize(new_sentence.lower())
tmp_sentence = [' '.join(tokens[:]).strip()]
self.session.run(self.infer_batch.initializer,
feed_dict={self.src_placeholder: tmp_sentence})
outputs, _ = self.model.infer(self.session)
if self.hparams.beam_width > 0:
outputs = outputs[0]
eos_token = self.hparams.eos_token.encode("utf-8")
outputs = outputs.tolist()[0]
if eos_token in outputs:
outputs = outputs[:outputs.index(eos_token)]
if pat_matched and pre_time == 0:
out_sentence, if_func_val = self._get_final_output(
outputs, para_list=num_list, html_format=html_format)
if if_func_val:
return out_sentence
else:
new_sentence = sentence
else:
out_sentence, _ = self._get_final_output(
outputs, html_format=html_format)
return out_sentence
def _get_final_output(self, sentence, para_list=None, html_format=False):
sentence = b' '.join(sentence).decode('utf-8')
if sentence == '':
return "I don't know what to say.", False
if_func_val = False
last_word = None
word_list = []
for word in sentence.split(' '):
word = word.strip()
if not word:
continue
if word.startswith('_func_val_'):
if_func_val = True
word = call_function(word[10:],
tokenized_data=self.tokenized_data,
para_list=para_list,
html_format=html_format)
else:
if word in self.tokenized_data.upper_words:
word = self.tokenized_data.upper_words[word]
if (last_word is None or last_word
in ['.', '!', '?']) and not word[0].isupper():
word = word.capitalize()
if not word.startswith('\'') and word != 'n\'t' \
and (word not in string.punctuation or word in ['(', '[', '{', '``', '$']) \
and last_word not in ['(', '[', '{', '``', '$']:
word = ' ' + word
word_list.append(word)
last_word = word
return ''.join(word_list).strip(), if_func_val