def complete(user_query):
#from word_vectors import TfIdfEmbedding
#print(dir())
nd = decider([user_query])
if nd == 1:
classification = classify([user_query])
print(classification)
call_model(user_query, classification[0])
else:
print("OUTLIER")
def classify_and_ner(statement):
from sklearn.externals import joblib
import numpy as np
from ner import call_model
model_file = open("../data/word2vec_final/lr_model.pkl", "rb")
selected_model = joblib.load(model_file)
statement_vector = np.asarray(statement)
x = selected_model.predict(statement_vector)
print(x)
#return x
call_model(statement[0], x[0])
def transform(self,X):
from nltk import word_tokenize
sentence_vector = np.empty([1, self.dimsize])
i = 0
for sentences in X:
if i == 0:
sentence_vector = np.array([np.mean(
[self.word2vec[w] for w in
word_tokenize(sentences) if w in self.word2vec ] or [np.zeros(self.dimsize)],axis=0
)
])
i += 1
else:
try :
sentence_vector = np.append(sentence_vector, np.array([np.mean(
[self.word2vec[w] for w in word_tokenize(sentences) if w in self.word2vec ] or [np.zeros(self.dimsize)
], axis=0
]), axis=0)
i+=1
except :
print("Sentence vector is giving the error : ",str(i+1))
i+=1
return sentence_vector
def create_model(csv_path):
w2v,train_data = create_gensim_word2vec("../data/word2vec_final/"+csv_path)
from sklearn.pipeline import Pipeline
from sklearn.externals import joblib
from sklearn.ensemble import ExtraTreesClassifier
from sklearn.svm import SVC
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import GridSearchCV
from sklearn.metrics import accuracy_score
from sklearn.model_selection import train_test_split,cross_val_score
#extree_w2v = Pipeline([("word2vec",Word2VecEmbedding(w2v)),("extree",ExtraTreesClassifier(n_estimators=400))])
extree_w2v_tfidf = Pipeline([("word2vec_tfidf",TfIdfEmbedding(w2v)),("extree",ExtraTreesClassifier(n_estimators = 400))])
param_grid = {'C': [1e3, 5e3, 2e3,3e3,4e3,1e4, 5e4, 1e5,5e5,1e2,10,1],
'gamma': [0.0001, 0.0005, 0.001, 0.005, 0.01, 0.1],
'kernel' : ['rbf','poly','linear','sigmoid']}
grid_search = GridSearchCV(SVC(class_weight='balanced', probability=True),
param_grid, n_jobs=10, verbose=1)
svm_w2v_tfidf = Pipeline([("word2vec_tfidf",TfIdfEmbedding(w2v)),("svm",grid_search)])
param_grid_logistic = { 'C' : [1e3, 5e3,2e3,3e3,4e3,1e4, 5e4, 1e5,5e5,1e2,10,1],
'solver':['newton-cg','liblinear','lbfgs']}
log_grid = GridSearchCV(LogisticRegression(penalty="l2"),param_grid_logistic,n_jobs=10,verbose=1)
lr_w2v_tfidf = Pipeline([("word2vec_tfidf",TfIdfEmbedding(w2v)),("log_reg",log_grid)])
X_train, X_test, y_train, y_test = train_test_split(train_data.values[:,1],train_data.values[:,2], test_size=0.15)
#extree_w2v.fit(X,y)
print("TRAINING THE EXTREE MODEL : \n")
extree_w2v_tfidf.fit(X_train,y_train)
print("\n")
print("TRAINING THE SVM MODEL : \n")
svm_w2v_tfidf.fit(X_train,y_train)
print("\n")
print("TRAINING THE LOG REG MODEL : \n")
lr_w2v_tfidf.fit(X_train,y_train)
print("\n")
with open("../data/word2vec_final/lr_model.pkl","wb") as f:
joblib.dump(lr_w2v_tfidf,f)
#### Evaluating the models
svm_validation = cross_val_score(svm_w2v_tfidf,X_train,y_train,scoring='accuracy',cv=5)
extree_validation = cross_val_score(extree_w2v_tfidf,X_train,y_train,scoring='accuracy',cv=5)
lr_validation = cross_val_score(lr_w2v_tfidf,X_train,y_train,scoring="accuracy",cv =5)
print("SVM CROSS VALIDATION : ")
print(svm_validation)
print("MEAN CV SCORE : ")
print(np.mean(svm_validation))
print("SVM BEST FIT : ")
print(svm_w2v_tfidf.named_steps['svm'].best_params_)
svm_w2v_tfidf_pred = svm_w2v_tfidf.predict(X_test)
print("ACCURACY FOR SVM ON TEST DATA: ", accuracy_score(y_test, svm_w2v_tfidf_pred))
#evaluate_models(svm_w2v_tfidf, "SVM")
print("EXTREE CROSS VALIDATION : ")
print(extree_validation)
print("MEAN CV SCORE : ")
print(np.mean(extree_validation))
extree_w2v_tfidf_pred = extree_w2v_tfidf.predict(X_test)
print("ACCURACY FOR EXTREE W2V-TFIDIF ON TEST DATA: ", accuracy_score(y_test, extree_w2v_tfidf_pred))
#evaluate_models(extree_w2v_tfidf, "EXTREE")
print("LR CROSS VALIDATION : ")
print(lr_validation)
print("MEAN CV SCORE : ")
print(np.mean(lr_validation))
lr_w2v_tfidf_pred = lr_w2v_tfidf.predict(X_test)
print("ACCURACY FOR LR ON TEST DATA: ", accuracy_score(y_test, lr_w2v_tfidf_pred))
print("LR BEST FIT : ")
print(lr_w2v_tfidf.named_steps['log_reg'].best_params_)
def evaluate_models(model_file,model_name):
ques = ["Can you book a flight for me please?",
"Just book a meeting with my flight manager please",
"Send a text message not a mail to book a meeting on a flight",
"Book a meeting",
"I would like you to book a flight please",
"Schedule a flight before a meeting",
"I would like you to schedule a meeting before booking the flight",
"Book a flight from delhi to mumbai",
"Send a text to david",
# "Send a formal mail to mary telling her to come early in the morning",
"Let sara know i am in delhi via text message",
# "Mail the students that the class has been canceled today",
"Schedule a meeting in a flight",
# "Send an email not a message to book a meeting in the flight",
# "Remind everyone about the meeting",
"Can you schedule a meeting with my friend please",
"book a meeting with david if done book a flight to mumbai",
# "I don't want you to book a flight",
"Would you mind turning off the fan and lights",
"Can you please send an email to Mr. Mogambo",
"What is the level of precipitation in the Bahamas",
"I am really bored",
"Tell me a joke about Mr Donald Trump"
# "I want you to shut up"
]
gold_standard = ["FLIGHT", "MEETING", "CONNECT", "MEETING", "FLIGHT", "FLIGHT", "MEETING", "FLIGHT", "CONNECT",
"CONNECT", "MEETING", "MEETING", "MEETING",
"IOT", "CONNECT", "CLIMATE", "JOKE", "JOKE"]
def calculate_accuracies(answers,model,questions = ques):
score = 0
for i in range(len(answers)):
if answers[i] == gold_standard[i].lower():
score = score + 1
else:
questions[i] = questions[i] + " * "
print("ACCURACY FOR " + str(model) + " : " + str(score) + " out of " + str(len(answers)))
print("QUESTIONS WITH * ARE CLASSIFIED WRONG")
for q in questions:
print(q + " :: " + str(answers[questions.index(q)]) + " CORRECT ANSWER : " + gold_standard[questions.index(q)])
calculate_accuracies(model_file.predict(ques),model_name)
def classify_and_ner(statement):
from sklearn.externals import joblib
import numpy as np
from ner import call_model
model_file = open("../data/word2vec_final/lr_model.pkl","rb")
selected_model = joblib.load(model_file)
statement_vector = np.asarray(statement)
x = selected_model.predict(statement_vector)
print(x)
#return x
call_model(statement[0],x[0])
def classify(statement):
from sklearn.externals import joblib
import numpy as np
model_file = open("lr_model.pkl", "rb")
selected_model = joblib.load(model_file)
statement_vector = np.asarray(statement)
x = selected_model.predict(statement_vector)
return x
# def classify_and_ner(statement):
#
# #from sklearn.feature_extraction.text import TfidfVectorizer
# os.chdir("..")
#
# from ner import call_model
# from sklearn.externals import joblib
#
# import numpy as np
#
#
# model_file = open("lr_model.pkl","rb")
# selected_model = joblib.load(model_file)
#
# #tf_file = open("tfidf.pkl","rb")
# #tfidf_trans = pkl.load(tf_file)
#
#
# statement_vector = np.asarray(statement)
# print(statement_vector)
# #x_test_counts = count_vect.fit_transform(statement_vector)
# #x_test_tfidf = tfidf_transformer.fit_transform(statement_vector)
# #print(x_test_tfidf)
# x = selected_model.predict(statement_vector)
# print(x)
#
# #call_model(statement[0],x[0])
#create_model("commands.csv")
#classify_and_ner(["Just schedule a meeting for me please","Book a meeting"])
# if len(sys.argv) == 2:
# if os.path.exists(sys.argv[1]):
# print("File has been found ! " + sys.argv[1])
# create_model(sys.argv[1])
# else:
# print("Invalid file path")
# for elements in sys.argv[1:]:
# if re.findall(".csv",elements):
# print("Found the file")
# create_gensim_word2vec(elements)
# else:
# create_model()
#
# else:
# create_model()
#classify_and_ner(["Book a meeting with Piyush for 4pm"])