[Kaggle DAY17]Real or Not? NLP with Disaster Tweets!

Kaggle 대회 17회차!

오늘은 기존의 데이터 전처리 방식에서

tokenizer를 word_tokenize에서 TreebankWordTokenizer로 변경하고

토큰화 후 stemmer를 사용하지 않고 lemmatizer를 사용하여 보았습니다.

기존의 방식보다 조금 더 원형의 형태를 보존해서 토큰화가 된 것을 볼 수 있었습니다.

 

모델은 16회차의 모델에 그대로 적용해보았습니다.

 

첫번째 제출

from keras import optimizers
adam = optimizers.Adam(lr=0.05, decay=0.1)

model_1 = Sequential()
model_1.add(Embedding(vocab_size, 100))
model_1.add(GRU(32))
model_1.add(Dropout(0.5))
model_1.add(Dense(2, activation='sigmoid'))
model_1.compile(loss='binary_crossentropy', optimizer=adam, metrics=['acc'])
history = model_1.fit(x_train, y_train, batch_size=32, epochs=1, validation_split=0.1)

predict = model_1.predict(x_test)
predict_labels = np.argmax(predict, axis=1)
for i in range(len(predict_labels)):
  predict_labels[i] = predict_labels[i]

ids = list(test['id'])

submission_dic = {"id":ids, "target":predict_labels}
submission_df = pd.DataFrame(submission_dic)
submission_df.to_csv("kaggle_day17.csv", index=False)

결과

 

두번째 제출

adam1 = optimizers.Adam(lr=0.03, decay=0.1)
model_2 = Sequential()
model_2.add(Embedding(vocab_size, 100))
model_2.add(GRU(32))
model_2.add(Dropout(0.5))
model_2.add(Dense(2, activation='sigmoid'))
model_2.compile(loss='binary_crossentropy', optimizer=adam1, metrics=['acc'])
history = model_2.fit(x_train, y_train, batch_size=20, epochs=1, validation_split=0.1)

결과

predict = model_2.predict(x_test)
predict_labels = np.argmax(predict, axis=1)
for i in range(len(predict_labels)):
  predict_labels[i] = predict_labels[i]

ids = list(test['id'])

submission_dic = {"id":ids, "target":predict_labels}
submission_df = pd.DataFrame(submission_dic)
submission_df.to_csv("kaggle_day17_2.csv", index=False)

 

세번째 제출

adam2 = optimizers.Adam(lr=0.05, decay=0.1)
model_3 = Sequential()
model_3.add(Embedding(vocab_size, 100))
model_3.add(GRU(32))
model_3.add(Dropout(0.5))
model_3.add(Dense(2, activation='sigmoid'))
model_3.compile(loss='binary_crossentropy', optimizer=adam2, metrics=['acc'])
history = model_3.fit(x_train, y_train, batch_size=16, epochs=1, validation_split=0.1)

predict = model_3.predict(x_test)
predict_labels = np.argmax(predict, axis=1)
for i in range(len(predict_labels)):
  predict_labels[i] = predict_labels[i]

ids = list(test['id'])

submission_dic = {"id":ids, "target":predict_labels}
submission_df = pd.DataFrame(submission_dic)
submission_df.to_csv("kaggle_day17_3.csv", index=False)

결과

 

네번째 제출

adam3 = optimizers.Adam(lr=0.05, decay=0.1)
model_4 = Sequential()
model_4.add(Embedding(vocab_size, 100))
model_4.add(GRU(32))
model_4.add(Dropout(0.5))
model_4.add(Dense(2, activation='sigmoid'))
model_4.compile(loss='binary_crossentropy', optimizer=adam3, metrics=['acc'])
history = model_4.fit(x_train, y_train, batch_size=20, epochs=1, validation_split=0.1)

predict = model_4.predict(x_test)
predict_labels = np.argmax(predict, axis=1)
for i in range(len(predict_labels)):
  predict_labels[i] = predict_labels[i]

ids = list(test['id'])

submission_dic = {"id":ids, "target":predict_labels}
submission_df = pd.DataFrame(submission_dic)
submission_df.to_csv("kaggle_day17_4.csv", index=False)

결과

 

다섯번째 제출

adam4 = optimizers.Adam(lr=0.05, decay=0.1)
model_5 = Sequential()
model_5.add(Embedding(vocab_size, 100))
model_5.add(GRU(32))
model_5.add(Dropout(0.5))
model_5.add(Dense(2, activation='sigmoid'))
model_5.compile(loss='binary_crossentropy', optimizer=adam4, metrics=['acc'])
history = model_5.fit(x_train, y_train, batch_size=28, epochs=1, validation_split=0.1)

predict = model_5.predict(x_test)
predict_labels = np.argmax(predict, axis=1)
for i in range(len(predict_labels)):
  predict_labels[i] = predict_labels[i]

ids = list(test['id'])

submission_dic = {"id":ids, "target":predict_labels}
submission_df = pd.DataFrame(submission_dic)
submission_df.to_csv("kaggle_day17_5.csv", index=False)

결과

 

다 제출하고 포스팅하는 과정에서 제출할 결과를 만드는 test데이터에

lemmatizer를 사용하지않고 한 것을 발견하여 내일은 lemmatizer를 활용하고 제출을 해보려고합니다.

 

읽어주셔서 감사합니다.