AICE.DNN

딥러닝Permalink
라이브러리 importPermalink
import tensorflow as tf
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Dropout

tf.random.set_seed(100)
Sequential 모델 생성Permalink
# Sequential() 모델 정의 하고 model로 저장
# input layer는 input_shape=() 옵션을 사용한다.
# 39개 input layer
# unit 4개 hidden layer
# unit 3개 hidden layer 
# 1개 output layser : 이진분류

model = Sequential()
model.add(Dense(4, activation='relu', input_shape=(39,)))
model.add(Dense(3, activation='relu'))
model.add(Dense(1, activation='sigmoid'))
과적합 방지 모델 생성Permalink
model = Sequential()
model.add(Dense(4, activation='relu', input_shape=(39,)))
model.add(Dropout(0.3))
model.add(Dense(3, activation='relu'))
model.add(Dropout(0.3))
model.add(Dense(1, activation='sigmoid'))
모델 컴파일Permalink
model.compile(optimizer='adam', 
              loss='binary_crossentropy', 
              metrics=['accuracy'])
모델 학습Permalink
# 앞쪽에서 정의된 모델 이름 : model
# Sequential 모델의 fit() 함수 사용
# @인자
### X, y : X_train, y_train
### validation_data=(X_test, y_test)
### epochs 10번
### batch_size 10번

model.fit(X_train, y_train, 
          validation_data=(X_test, y_test),
          epochs=10, 
          batch_size=10)
다중 분류 DNNPermalink
# 39개 input layer
# unit 5개 hidden layer
# dropout
# unit 4개 hidden layer 
# dropout
# 2개 output layser : 다중분류

model = Sequential()
model.add(Dense(5, activation='relu', input_shape=(39,)))
model.add(Dropout(0.3))
model.add(Dense(4, activation='relu'))
model.add(Dropout(0.3))
model.add(Dense(2, activation='softmax'))
모델 컴파일Permalink
model.compile(optimizer='adam', 
              loss='sparse_categorical_crossentropy', 
              metrics=['accuracy'])
모델 학습Permalink
history = model.fit(X_train, y_train, 
          validation_data=(X_test, y_test),
          epochs=20, 
          batch_size=16)
조기 종료 모델Permalink
from tensorflow.keras.callbacks import EarlyStopping, ModelCheckpoint

# val_loss 모니터링해서 성능이 5번 지나도록 좋아지지 않으면 조기 종료
early_stop = EarlyStopping(monitor='val_loss', mode='min', 
                           verbose=1, patience=5)

# val_loss 가장 낮은 값을 가질때마다 모델저장
check_point = ModelCheckpoint('best_model.h5', verbose=1, 
                              monitor='val_loss', mode='min', save_best_only=True)

history = model.fit(x=X_train, y=y_train, 
          epochs=50 , batch_size=20,
          validation_data=(X_test, y_test), verbose=1,
          callbacks=[early_stop, check_point])
OversamplingPermalink
from imblearn.over_sampling import SMOTE

# SMOTE 함수 정의 및 Oversampling 수행

smote = SMOTE(random_state=0)
X_train_over, y_train_over = smote.fit_resample(X_train, y_train)

# 정규화
from sklearn.preprocessing import MinMaxScaler

scaler = MinMaxScaler()
scaler.fit(X_train)
X_train_over = scaler.transform(X_train_over)
X_test = scaler.transform(X_test)

# 모델 개발
model = Sequential()
model.add(Dense(64, activation='relu', input_shape=(39,)))
model.add(Dropout(0.3))
model.add(Dense(32, activation='relu'))
model.add(Dropout(0.3))
model.add(Dense(16, activation='relu'))
model.add(Dropout(0.3))
model.add(Dense(2, activation='softmax'))

model.compile(optimizer='adam', 
              loss='sparse_categorical_crossentropy', 
              metrics=['accuracy'])

# 여기서는 val_accuracy 모니터링해서 성능이 좋아지지 않으면 조기 종료 하게 함.
early_stop = EarlyStopping(monitor='val_accuracy', mode='max', 
                           verbose=1, patience=5)

check_point = ModelCheckpoint('best_model.h5', verbose=1,
                              monitor='val_loss', mode='min', 
                              save_best_only=True)

history = model.fit(x=X_train_over, y=y_train_over, 
          epochs=50 , batch_size=32,
          validation_data=(X_test, y_test), verbose=1,
          callbacks=[early_stop, check_point])