편집
1,408
번
바뀜
둘러보기로 가기
검색하러 가기
141번째 줄:
141번째 줄: − + + + 148번째 줄:
150번째 줄: + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
→드롭아웃
|}
|}
= 드롭아웃 =
= 오버피팅 막기 =
== 드롭아웃 ==
<syntaxhighlight lang="python3">
<syntaxhighlight lang="python3">
from keras import model, layers
from keras import model, layers
models.add(....)
models.add(....)
models.add(layers.Dropout(0.5)) # 얼마의 비율로 드롭아웃을 실행할 것인가.
models.add(layers.Dropout(0.5)) # 얼마의 비율로 드롭아웃을 실행할 것인가.
</syntaxhighlight>층 사이사이에 넣을 수 있다.
== 가중치 규제 ==
=== L2 규제 만들기 ===
<syntaxhighlight lang="python3">
from keras import model, layers, regularizers
l2_model = models.Sequential()
l2_model.add(layers.Dense(16, kernel_regularizer=regularizers.l2(0.001), # 0.001은 가중치에 곱할 값.
activation='relu', input_shape=(10000,)))
l2_model.add(layers.Dense(16, kernel_regularizer=regularizers.l2(0.001),
activation='relu'))
l2_model.add(layers.Dense(1, activation='sigmoid'))
</syntaxhighlight>
=== L1 규제 만들기 ===
<syntaxhighlight lang="python3">
from keras import model, layers, regularizers
l1_model = models.Sequential()
l1_model.add(layers.Dense(16, kernel_regularizer=regularizers.l1(0.0001),
activation='relu', input_shape=(10000,)))
l1_model.add(layers.Dense(16, kernel_regularizer=regularizers.l1(0.0001),
activation='relu'))
l1_model.add(layers.Dense(1, activation='sigmoid'))
</syntaxhighlight>
=== L1, L2 동시 사용 ===
<syntaxhighlight lang="python3">
l1l2_model = models.Sequential()
l1l2_model.add(layers.Dense(16, kernel_regularizer=regularizers.l1_l2(l1=0.0001, l2=0.0001),
activation='relu', input_shape=(10000,)))
l1l2_model.add(layers.Dense(16, kernel_regularizer=regularizers.l1_l2(l1=0.0001, l2=0.0001),
activation='relu'))
l1l2_model.add(layers.Dense(1, activation='sigmoid'))
</syntaxhighlight>
</syntaxhighlight>