keras 기본 사용법

keras는 아래 6가지 순서대로 생성 할 수 있다.

1. input node 설정
2. layer 연결 또는 output 설정
3. Model 생성 및 실행
4. Compile (for deep learning)
5. Training
6. Evaluation

referece by tensorflow officaial tutorial : Keras > Functional API

위 과정을 따라 Keras 을 사용하는데 있어 크게 4가지로 분류 된다.

1. single input, single output
2. single input, Multi output
3. Multi input, single output
4. Multi input, Multi output

하지만 여기서는 대표적으로 2가지 과정만 살펴본다.

1. single input, single output
2. Multi input, Multi output

1. input node 설정(single input, multi input)

from tensorflow.keras.layers import Input

1.1 Single Input

x1 = Input(shape=[], name='x1')

1.2 Multi Input

x1 = Input(shape=[], name='x1')
x2 = Input(shape=[], name='x2')

1.3 Input Datatype

Input Datartype 으로는 tuple, list 가 가능하다. ㅁ

x1 = Input(shape=(256,), name='x1')
x1 = Input(shape=[256], name='x1')

2 . 여러 layer 연결 또는 연산

2.1 연산

y = x1 + x2

2.2 여러 Layer 연결

input_ = Input(shape=[784])
layer1 = Desne(units=256, activation='relu')(input_)
layer2 = Desne(units=256, activation='relu')(layer1)
layer3 = Desne(units=256, activation='relu')(layer2)

3 . Output

3.1 single output

input_ = Input(shape=[784])
layer1 = Desne(units=256, activation='relu')
layer2 = Desne(units=256, activation='relu')(layer1)
layer3 = Desne(units=256, activation='relu')(layer2)

output = layer3

3.2 multi output

input_ = Input(shape=[784])
layer1 = Desne(units=256, activation='relu')
layer2 = Desne(units=256, activation='relu')(layer1)
layer3 = Desne(units=256, activation='relu')(layer2)

output1 = layer2
output2 = layer3

Model 생성

1. single input, single output

   from tensorflow.keras.models import Model
   from tensorflow.keras.layers import Input, Dense
   import tensorflow as tf
   
   input_ = Input(shape=[784], name='input_')
   layer1 = Dense(units=256, activation='relu')(input_)
   layer2 = Dense(units=256, activation='relu')(layer1)
   layer3 = Dense(units=10, activation='relu')(layer2)
   output = layer3
   
   # model 생성 
   model = Model(input_, output)
   
   # model 실행
   x = tf.zeros(shape=[1, 784])
   print(model(x))
   print(model.predict(x))
   
   

2. multi input, multi output

   from tensorflow.keras.models import Model
   from tensorflow.keras.layers import Input, Dense
   import tensorflow as tf
   
   input_1 = Input(shape=[784], name='input_1')
   layer1 = Dense(units=10, activation='relu')(input_1)
   
   input_2 = Input(shape=[784], name='input_2')
   layer2 = Dense(units=10, activation='relu')(input_2)
   
   
   output = layer1 + layer2
   inputs = {"input_1": input_1, "input_2": input_2}
   model = Model(inputs, output)
   
   # model 실행
   x = tf.zeros(shape=[1, 784])
   print(model(x))
   print(model.predict(x))
   
   

Model 실행

• 모델 실행 방법

  1. __call__()
  2. Model.predict()

• multi input , single output

from tensorflow.keras.models import Model
from tensorflow.keras.layers import Input, Dense
import tensorflow as tf

# multi input 
input_1 = Input(shape=[784])
layer1 = Dense(units=10, activation='relu')(input_1)

input_2 = Input(shape=[784])
layer2 = Dense(units=10, activation='relu')(input_2)

# single output
output = layer1 + layer2
inputs = {"input_1": input_1, "input_2": input_2}
model = Model(inputs, output)

x = tf.zeros(shape=[1, 784])
input_values = {"input_1": x, "input_2": x}

# model 실행 : __call__() => numpy 
print(model.predict(input_values))

# model 실행 : Model.predict() => tensor 
print(model(input_values))

• Multi input, multi output

  from tensorflow.keras.models import Model
  from tensorflow.keras.layers import Input, Dense
  import tensorflow as tf
  
  # multi input 
  input_1 = Input(shape=[784])
  output_1 = Dense(units=10, activation='relu', name='output_1')(input_1)
  
  input_2 = Input(shape=[784])
  output_2 = Dense(units=10, activation='relu', name='output_2')(input_2)
  
  # single output
  inputs = {"input_1": input_1, "input_2": input_2}
  outputs = {"output_1": output_1, "output_2": output_2}
  model = Model(inputs, output)
  
  x = tf.zeros(shape=[1, 784])
  input_values = {"input_1": x, "input_2": x}
  
  # model 실행 : __call__() => numpy 
  print(model.predict(input_values))
  
  # model 실행 : Model.predict() => tensor 
  print(model(input_values))
  
  

Optional

model 에 input 입력시 dictionary 로 입력 가능

output = layer1 + layer2
inputs = {"input_1": input_1, "input_2": input_2}
model = Model(inputs, output)

model 에 input 입력시 list 로 입력 가능

output = layer1 + layer2
inputs = [input_1, input_2]
model = Model(inputs, output)

Reference

offical tensorflow guide