class IntentAgent:
def __init__(self, config):
self.config = copy.deepcopy(config)
self.agent = None
self.classes = None
self.preprocessor = None
def init_agent(self):
self.config['model_from_saved'] = True
self.agent = KerasMulticlassModel(self.config)
self.classes = self.agent.classes
self.preprocessor = NLTKTokenizer()
def _run_score(self, observation):
task = observation[0]
infer_result = self.agent.infer(self.preprocessor.infer(task))
prediction = self.classes[np.argmax(infer_result)]
return prediction
def answer(self, input_task):
try:
if isinstance(input_task, list):
result = self._run_score(input_task)
return result
elif isinstance(input_task, int):
result = 'There is no Intent Classifier testing API provided'
return result
else:
return {"ERROR": "Parameter error - {} belongs to unknown type".format(
str(input_task))}
except Exception as e:
print(e)
return {"ERROR": "{}".format(traceback.extract_stack())}
from intent_model.preprocessing import NLTKTokenizer
from intent_model.multiclass import KerasMulticlassModel
config_file = sys.argv[1]
# Reading parameters
with open(config_file, "r") as f:
opt = json.load(f)
# Infering is possible only for saved intent_model
opt['model_from_saved'] = True
# Initializing intent_model
print("Initializing intent_model")
model = KerasMulticlassModel(opt)
# Initializing classes
classes = model.classes
# Initializing preprocessor
preprocessor = NLTKTokenizer()
phrase = "I want you to add 'I love you, baby' to my playlist"
# Predicting
predictions = model.infer(preprocessor.infer(phrase))
# Result
print(np.max(predictions), classes[np.argmax(predictions)])
config_file = sys.argv[1]
data_file = sys.argv[2]
# Reading full data
comment_name = "request"
train_data = pd.read_csv(Path(data_file), sep=',')
print(train_data.head())
values = {"istask": 0, "request": "element"}
train_data.fillna(values, inplace=True)
# Tokenization that splits words and punctuation by space
preprocessor = NLTKTokenizer()
for k in range(3839):
inds = np.arange(k * 10000, min((k + 1) * 10000, train_data.shape[0]))
train_data.loc[inds, comment_name] = preprocessor.infer(train_data.loc[inds, comment_name].values)
# Reading parameters of intent_model from json
with open(config_file, "r") as f:
opt = json.load(f)
# Initializing classes from dataset
columns = list(train_data.columns)
columns.remove(comment_name)
classes = np.array(columns)
opt["classes"] = " ".join(list(classes))
print(classes)
# Constructing data
data_dict = dict()
train_pairs = []
config_file = sys.argv[1]
data_file = sys.argv[2]
# Reading full data
comment_name = "request"
train_data = pd.read_csv(Path(data_file), sep=',')
print(train_data.head())
values = {"istask": 0, "request": "пропущено"}
train_data.fillna(values, inplace=True)
# Tokenization that splits words and punctuation by space
preprocessor = NLTKTokenizer()
for k in range(3839):
inds = np.arange(k * 10000, min((k + 1) * 10000, train_data.shape[0]))
train_data.loc[inds, comment_name] = preprocessor.infer(train_data.loc[inds, comment_name].values)
# Reading parameters of intent_model from json
with open(config_file, "r") as f:
opt = json.load(f)
# Initializing classes from dataset
columns = list(train_data.columns)
columns.remove(comment_name)
classes = np.array(columns)
opt["classes"] = " ".join(list(classes))
print(classes)
# Constructing data
data_dict = dict()
train_pairs = []
from intent_model.preprocessing import NLTKTokenizer
from intent_model.multiclass import KerasMulticlassModel
config_file = sys.argv[1]
# Reading parameters
with open(config_file, "r") as f:
opt = json.load(f)
# Infering is possible only for saved intent_model
opt['model_from_saved'] = True
# Initializing intent_model
print("Initializing intent_model")
model = KerasMulticlassModel(opt)
# Initializing classes
classes = model.classes
# Initializing preprocessor
preprocessor = NLTKTokenizer()
phrase = "I want you to add 'I love you, baby' to my playlist"
# Predicting
predictions = model.infer(preprocessor.infer(phrase))
# Result
print(np.max(predictions), classes[np.argmax(predictions)])