CH3. 비지도 그래프 학습(오토인코더)

graph
Author

김보람

Published

April 6, 2023

ref

오토인코더(AutoEncoder)

  • 입력과 출력이 기본적으로 동일, 은닉충에서 적은 수의 유닛 사용

  • 상당히 적은 수의 변수 및 자유도를 사용해 입력을 재구성하도록 학습하는 신경망

  • 입력과 출력의 차이 최소화하여 학습

  • 손실함수: 평균 제곱 오차, 평균 절대 오차, 교차 엔트로피, KL발산 등

- 구성

  • 인코더 네트워크: 하나 이상의 단위를 통해 입력을 처리하고, 입력의 차원을 줄이고 희소성을 제한하는 인코딩된 표현으로 매핑

  • 디코더 네트워크: 중간 계층의 인코딩된 표현에서 입력 신호 재구성

Load Dataset

import tensorflow as tf
from tensorflow.keras.datasets import fashion_mnist
2023-04-06 18:27:51.222436: I tensorflow/core/platform/cpu_feature_guard.cc:193] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  AVX2 FMA
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
(x_train, y_train), (x_test, y_test) = fashion_mnist.load_data()
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/train-labels-idx1-ubyte.gz
29515/29515 [==============================] - 0s 0us/step
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/train-images-idx3-ubyte.gz
26421880/26421880 [==============================] - 0s 0us/step
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/t10k-labels-idx1-ubyte.gz
5148/5148 [==============================] - 0s 0us/step
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/t10k-images-idx3-ubyte.gz
4422102/4422102 [==============================] - 0s 0us/step
x_train = x_train.astype('float32') / 255. #0~1사이로 정규화 하기 위하여
x_test = x_test.astype('float32') / 255.

print (x_train.shape)
print (x_test.shape)
(60000, 28, 28)
(10000, 28, 28)
from matplotlib import pyplot as plt
classes = {
    0:"T-shirt/top",
    1: "Trouser",
    2: "Pullover",
    3: "Dress",
    4: "Coat",
    5: "Sandal",
    6: "Shirt",
    7: "Sneaker",
    8: "Bag",
    9: "Ankle boot", 
}
n = 10
plt.figure(figsize=(20, 4))
for i in range(n):
    # display original
    ax = plt.subplot(1, n, i + 1)
    plt.imshow(x_test[i])
    plt.title(classes[y_test[i]])
    plt.gray()
    ax.get_xaxis().set_visible(False)
    ax.get_yaxis().set_visible(False)

plt.show()
# plt.savefig("TrainingSet.png")

Create Autoencoder

from tensorflow.keras.layers import Flatten, Conv2D, Dropout, MaxPooling2D, UpSampling2D, Input
from tensorflow.keras import Model

인코더/디코더

input_img = Input(shape=(28, 28, 1))

#인코딩
x = Conv2D(16, (3, 3), activation='relu', padding='same')(input_img)      # 출력값 28x28x16
x = MaxPooling2D((2, 2), padding='same')(x)                               # 출력값 14x14x16
x = Conv2D(8, (3, 3), activation='relu', padding='same')(x)               # 출력값 14x14x8
x = MaxPooling2D((2, 2), padding='same')(x)                               # 출력값 7x7x8
x = Conv2D(8, (3, 3), activation='relu', padding='same')(x)               # 출력값 7x7x8
encoded = MaxPooling2D((2, 2), padding='same')(x)                         # 출력값 4x4x8

# at this point the representation is (4, 4, 8) i.e. 128-dimensional

#디코딩
x = Conv2D(8, (3, 3), activation='relu', padding='same')(encoded)         # 출력값 4x4x8
x = UpSampling2D((2, 2))(x)                                               # 출력값 8x8x8
x = Conv2D(8, (3, 3), activation='relu', padding='same')(x)
x = UpSampling2D((2, 2))(x)
x = Conv2D(16, (3, 3), activation='relu')(x)
x = UpSampling2D((2, 2))(x)
decoded = Conv2D(1, (3, 3), activation='sigmoid', padding='same')(x)

autoencoder = Model(input_img, decoded)
Model(input_img, encoded).summary()
Model: "model_2"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 input_2 (InputLayer)        [(None, 28, 28, 1)]       0         
                                                                 
 conv2d_7 (Conv2D)           (None, 28, 28, 16)        160       
                                                                 
 max_pooling2d_3 (MaxPooling  (None, 14, 14, 16)       0         
 2D)                                                             
                                                                 
 conv2d_8 (Conv2D)           (None, 14, 14, 8)         1160      
                                                                 
 max_pooling2d_4 (MaxPooling  (None, 7, 7, 8)          0         
 2D)                                                             
                                                                 
 conv2d_9 (Conv2D)           (None, 7, 7, 8)           584       
                                                                 
 max_pooling2d_5 (MaxPooling  (None, 4, 4, 8)          0         
 2D)                                                             
                                                                 
=================================================================
Total params: 1,904
Trainable params: 1,904
Non-trainable params: 0
_________________________________________________________________

손실함수/옵티마이저

autoencoder.compile(optimizer='adam', loss='binary_crossentropy')

학습

from tensorflow.keras.callbacks import TensorBoard
autoencoder.fit(x_train, x_train,
                epochs=50,
                batch_size=128,
                shuffle=True,
                validation_data=(x_test, x_test),
                callbacks=[TensorBoard(log_dir='/tmp/autoencoder')])
Epoch 1/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0399 - val_loss: 0.0086
Epoch 2/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0085 - val_loss: 0.0085
Epoch 3/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0085 - val_loss: 0.0085
Epoch 4/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0084
Epoch 5/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0084
Epoch 6/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0083
Epoch 7/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0083 - val_loss: 0.0083
Epoch 8/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0082 - val_loss: 0.0083
Epoch 9/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0082 - val_loss: 0.0082
Epoch 10/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0081 - val_loss: 0.0081
Epoch 11/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0081 - val_loss: 0.0081
Epoch 12/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0081 - val_loss: 0.0081
Epoch 13/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0081 - val_loss: 0.0081
Epoch 14/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0080 - val_loss: 0.0080
Epoch 15/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0080 - val_loss: 0.0080
Epoch 16/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0080 - val_loss: 0.0080
Epoch 17/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0080 - val_loss: 0.0080
Epoch 18/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0080 - val_loss: 0.0080
Epoch 19/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0080 - val_loss: 0.0080
Epoch 20/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0080 - val_loss: 0.0080
Epoch 21/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0080 - val_loss: 0.0080
Epoch 22/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0080 - val_loss: 0.0080
Epoch 23/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0080 - val_loss: 0.0080
Epoch 24/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0080 - val_loss: 0.0080
Epoch 25/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0080 - val_loss: 0.0080
Epoch 26/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0080 - val_loss: 0.0080
Epoch 27/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0080 - val_loss: 0.0080
Epoch 28/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0080 - val_loss: 0.0080
Epoch 29/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0080 - val_loss: 0.0080
Epoch 30/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0080 - val_loss: 0.0080
Epoch 31/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0080 - val_loss: 0.0080
Epoch 32/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0080 - val_loss: 0.0080
Epoch 33/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0079 - val_loss: 0.0080
Epoch 34/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0079 - val_loss: 0.0080
Epoch 35/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0079 - val_loss: 0.0080
Epoch 36/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0079 - val_loss: 0.0080
Epoch 37/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0079 - val_loss: 0.0080
Epoch 38/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0079 - val_loss: 0.0080
Epoch 39/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0079 - val_loss: 0.0080
Epoch 40/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0079 - val_loss: 0.0080
Epoch 41/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0079 - val_loss: 0.0080
Epoch 42/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0079 - val_loss: 0.0080
Epoch 43/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0079 - val_loss: 0.0080
Epoch 44/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0079 - val_loss: 0.0080
Epoch 45/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0079 - val_loss: 0.0080
Epoch 46/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0079 - val_loss: 0.0080
Epoch 47/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0079 - val_loss: 0.0080
Epoch 48/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0079 - val_loss: 0.0080
Epoch 49/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0079 - val_loss: 0.0080
Epoch 50/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0079 - val_loss: 0.0080
<keras.callbacks.History at 0x7f27e4110970>
autoencoder.save("./data/Batch50.p")
WARNING:absl:Found untraced functions such as _jit_compiled_convolution_op, _jit_compiled_convolution_op, _jit_compiled_convolution_op, _jit_compiled_convolution_op, _jit_compiled_convolution_op while saving (showing 5 of 8). These functions will not be directly callable after loading.
INFO:tensorflow:Assets written to: ./data/Batch50.p/assets
INFO:tensorflow:Assets written to: ./data/Batch50.p/assets

예측

예시1


from tensorflow.keras.models import load_model
autoencoder_first = load_model("./data/Batch50.p")
decoded_imgs = autoencoder_first.predict(x_test)

n = 6
plt.figure(figsize=(20, 7))
for i in range(1, n + 1):
    # Display original
    ax = plt.subplot(2, n, i)
    plt.imshow(x_test[i].reshape(28, 28))
    plt.gray()
    ax.get_xaxis().set_visible(False)
    ax.get_yaxis().set_visible(False)

    # Display reconstruction
    ax = plt.subplot(2, n, i + n)
    plt.imshow(decoded_imgs[i].reshape(28, 28))
    plt.gray()
    ax.get_xaxis().set_visible(False)  #X축 숨기기
    ax.get_yaxis().set_visible(False)
plt.show()
313/313 [==============================] - 0s 1ms/step

  • 첫 줄 우너본 이미지, 두번째 줄 재구성된 이미지

예시2

from tensorflow.keras.optimizers import Adam
autoencoder.compile(optimizer=Adam(learning_rate=0.0005), loss='binary_crossentropy')
autoencoder.fit(x_train, x_train,
                epochs=10,
                batch_size=128,
                shuffle=True,
                validation_data=(x_test, x_test),
                callbacks=[TensorBoard(log_dir='/tmp/autoencoder')])
Epoch 1/10
469/469 [==============================] - 11s 23ms/step - loss: 0.0079 - val_loss: 0.0079
Epoch 2/10
469/469 [==============================] - 11s 23ms/step - loss: 0.0079 - val_loss: 0.0079
Epoch 3/10
469/469 [==============================] - 11s 23ms/step - loss: 0.0079 - val_loss: 0.0080
Epoch 4/10
469/469 [==============================] - 11s 23ms/step - loss: 0.0079 - val_loss: 0.0079
Epoch 5/10
469/469 [==============================] - 11s 23ms/step - loss: 0.0079 - val_loss: 0.0079
Epoch 6/10
469/469 [==============================] - 11s 23ms/step - loss: 0.0079 - val_loss: 0.0079
Epoch 7/10
469/469 [==============================] - 11s 23ms/step - loss: 0.0079 - val_loss: 0.0079
Epoch 8/10
469/469 [==============================] - 11s 23ms/step - loss: 0.0079 - val_loss: 0.0079
Epoch 9/10
469/469 [==============================] - 11s 23ms/step - loss: 0.0079 - val_loss: 0.0079
Epoch 10/10
469/469 [==============================] - 11s 23ms/step - loss: 0.0079 - val_loss: 0.0079
<keras.callbacks.History at 0x7f272ba11550>

epochs 위에건 50.. 넘 오래걸려서 10개로 줄여보자

autoencoder.save("./data/Batch100.p")
WARNING:absl:Found untraced functions such as _jit_compiled_convolution_op, _jit_compiled_convolution_op, _jit_compiled_convolution_op, _jit_compiled_convolution_op, _jit_compiled_convolution_op while saving (showing 5 of 8). These functions will not be directly callable after loading.
INFO:tensorflow:Assets written to: ./data/Batch100.p/assets
INFO:tensorflow:Assets written to: ./data/Batch100.p/assets
decoded_imgs = autoencoder.predict(x_test)

n = 10
plt.figure(figsize=(20, 4))
for i in range(1, n + 1):
    # Display original
    ax = plt.subplot(2, n, i)
    plt.imshow(x_test[i].reshape(28, 28))
    plt.gray()
    ax.get_xaxis().set_visible(False)
    ax.get_yaxis().set_visible(False)

    # Display reconstruction
    ax = plt.subplot(2, n, i + n)
    plt.imshow(decoded_imgs[i].reshape(28, 28))
    plt.gray()
    ax.get_xaxis().set_visible(False)
    ax.get_yaxis().set_visible(False)

plt.show()
313/313 [==============================] - 1s 2ms/step

아담으로 5번 한거랑 우에서 50번한거랑 비슷하게 나왔땅

Embeddings(T-SNE사용)

- T-SNE: 2차원 평면에서 이미지 인코딩된 버전 표현

embeddings = Model(input_img, Flatten()(encoded)).predict(x_test)
313/313 [==============================] - 0s 815us/step
from sklearn.manifold import TSNE
import numpy as np
tsne = TSNE(n_components=2)
emb2d = tsne.fit_transform(embeddings)
x,y = np.squeeze(emb2d[:, 0]), np.squeeze(emb2d[:, 1])

  • np.squeeze(emb2d[:, 0])는 emb2d의 모든 행에서 첫 번째 열만 선택하여 1차원 배열로 만드는 함수. 이를 통해 2차원 배열의 차원을 축소시키고, x좌표 값을 얻을 수 있다.

  • np.squeeze(emb2d[:, 1])는 emb2d의 모든 행에서 두 번째 열만 선택하여 1차원 배열로 만드는 함수. 이를 통해 2차원 배열의 차원을 축소시키고, y좌표 값을 얻을 수 있다.

import pandas as pd
from matplotlib.cm import tab10
summary =  pd.DataFrame({"x": x, "y": y, "target": y_test, "size": 10})

plt.figure(figsize=(10,8))

for key, sel in summary.groupby("target"):
    plt.scatter(sel["x"], sel["y"], s=10, color=tab10.colors[key], label=classes[key])
    
plt.legend()
plt.axis("off")

  • 샘플이 속한 클래스에 따라서 색상이 지정

  • 서로 다른 의류의 클러스터링 명확

단점: 학습된 이미지를 정확히 재생성하고 일반화하지 않는 경향

Denoising

- 노이즈 제거 오토인코더 : 다양한 강도의 노이즈를 사용해 손상된 입력에 대해 노이즈 없는 답안 사용

from tensorflow.keras.layers import GaussianNoise
input_img = Input(shape=(28, 28, 1))

noisy_input = GaussianNoise(0.1)(input_img)

x = Conv2D(16, (3, 3), activation='relu', padding='same')(noisy_input)
x = MaxPooling2D((2, 2), padding='same')(x)
x = Conv2D(8, (3, 3), activation='relu', padding='same')(x)
x = MaxPooling2D((2, 2), padding='same')(x)
x = Conv2D(8, (3, 3), activation='relu', padding='same')(x)
encoded = MaxPooling2D((2, 2), padding='same')(x)

# at this point the representation is (4, 4, 8) i.e. 128-dimensional

x = Conv2D(8, (3, 3), activation='relu', padding='same')(encoded)
x = UpSampling2D((2, 2))(x)
x = Conv2D(8, (3, 3), activation='relu', padding='same')(x)
x = UpSampling2D((2, 2))(x)
x = Conv2D(16, (3, 3), activation='relu')(x)
x = UpSampling2D((2, 2))(x)
decoded = Conv2D(1, (3, 3), activation='sigmoid', padding='same')(x)

noisy_autoencoder = Model(input_img, decoded)

  • 손상된 입력을 사용해 네트워크 훈련

  • 출력에는 노이즈 없는 이미지 사용

  • GaussianNoise 학습 중에 확률적 노이즈 추가

Model(input_img, decoded).summary()
Model: "model_5"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 input_3 (InputLayer)        [(None, 28, 28, 1)]       0         
                                                                 
 gaussian_noise (GaussianNoi  (None, 28, 28, 1)        0         
 se)                                                             
                                                                 
 conv2d_14 (Conv2D)          (None, 28, 28, 16)        160       
                                                                 
 max_pooling2d_6 (MaxPooling  (None, 14, 14, 16)       0         
 2D)                                                             
                                                                 
 conv2d_15 (Conv2D)          (None, 14, 14, 8)         1160      
                                                                 
 max_pooling2d_7 (MaxPooling  (None, 7, 7, 8)          0         
 2D)                                                             
                                                                 
 conv2d_16 (Conv2D)          (None, 7, 7, 8)           584       
                                                                 
 max_pooling2d_8 (MaxPooling  (None, 4, 4, 8)          0         
 2D)                                                             
                                                                 
 conv2d_17 (Conv2D)          (None, 4, 4, 8)           584       
                                                                 
 up_sampling2d_6 (UpSampling  (None, 8, 8, 8)          0         
 2D)                                                             
                                                                 
 conv2d_18 (Conv2D)          (None, 8, 8, 8)           584       
                                                                 
 up_sampling2d_7 (UpSampling  (None, 16, 16, 8)        0         
 2D)                                                             
                                                                 
 conv2d_19 (Conv2D)          (None, 14, 14, 16)        1168      
                                                                 
 up_sampling2d_8 (UpSampling  (None, 28, 28, 16)       0         
 2D)                                                             
                                                                 
 conv2d_20 (Conv2D)          (None, 28, 28, 1)         145       
                                                                 
=================================================================
Total params: 4,385
Trainable params: 4,385
Non-trainable params: 0
_________________________________________________________________
noisy_autoencoder.compile(optimizer='adam', loss='binary_crossentropy')
noisy_autoencoder.fit(x_train, x_train,
                epochs=50,
                batch_size=128,
                shuffle=True,
                validation_data=(x_test, x_test),
                callbacks=[TensorBoard(log_dir='/tmp/noisy_autoencoder')])
Epoch 1/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0201
Epoch 2/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0193
Epoch 3/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0180
Epoch 4/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0162
Epoch 5/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0157
Epoch 6/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0150
Epoch 7/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0135
Epoch 8/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0129
Epoch 9/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0123
Epoch 10/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0111
Epoch 11/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0109
Epoch 12/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0103
Epoch 13/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0098
Epoch 14/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0096
Epoch 15/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0094
Epoch 16/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0094
Epoch 17/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0090
Epoch 18/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0088
Epoch 19/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0086
Epoch 20/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0088
Epoch 21/50
469/469 [==============================] - 11s 24ms/step - loss: 0.0084 - val_loss: 0.0087
Epoch 22/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0087
Epoch 23/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0086
Epoch 24/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0086
Epoch 25/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0086
Epoch 26/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0086
Epoch 27/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0085
Epoch 28/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0085
Epoch 29/50
469/469 [==============================] - 11s 24ms/step - loss: 0.0084 - val_loss: 0.0086
Epoch 30/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0085
Epoch 31/50
469/469 [==============================] - 11s 24ms/step - loss: 0.0084 - val_loss: 0.0085
Epoch 32/50
469/469 [==============================] - 11s 24ms/step - loss: 0.0084 - val_loss: 0.0085
Epoch 33/50
469/469 [==============================] - 11s 24ms/step - loss: 0.0084 - val_loss: 0.0085
Epoch 34/50
469/469 [==============================] - 11s 24ms/step - loss: 0.0084 - val_loss: 0.0085
Epoch 35/50
469/469 [==============================] - 11s 24ms/step - loss: 0.0084 - val_loss: 0.0085
Epoch 36/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0085
Epoch 37/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0085
Epoch 38/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0085
Epoch 39/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0085
Epoch 40/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0085
Epoch 41/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0085
Epoch 42/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0085
Epoch 43/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0084
Epoch 44/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0085
Epoch 45/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0084
Epoch 46/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0085
Epoch 47/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0084
Epoch 48/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0085
Epoch 49/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0084
Epoch 50/50
469/469 [==============================] - 11s 23ms/step - loss: 0.0084 - val_loss: 0.0085
<keras.callbacks.History at 0x7f27900e2670>
autoencoder.save("./data/DenoisingAutoencoder.p")
WARNING:absl:Found untraced functions such as _jit_compiled_convolution_op, _jit_compiled_convolution_op, _jit_compiled_convolution_op, _jit_compiled_convolution_op, _jit_compiled_convolution_op while saving (showing 5 of 8). These functions will not be directly callable after loading.
INFO:tensorflow:Assets written to: ./data/DenoisingAutoencoder.p/assets
INFO:tensorflow:Assets written to: ./data/DenoisingAutoencoder.p/assets
noise_factor = 0.1
x_train_noisy = x_train + noise_factor * np.random.normal(loc=0.0, scale=1.0, size=x_train.shape) 
x_test_noisy = x_test + noise_factor * np.random.normal(loc=0.0, scale=1.0, size=x_test.shape) 

x_train_noisy = np.clip(x_train_noisy, 0., 1.)
x_test_noisy = np.clip(x_test_noisy, 0., 1.)
  • np.clip 데이터 0과 1사이의 값으로 클리핑
decoded_imgs = autoencoder.predict(x_test_noisy)

decoded_imgs_denoised = noisy_autoencoder.predict(x_test_noisy)

n = 6
plt.figure(figsize=(20, 10))
for i in range(1, n + 1):
    # Display original
    ax = plt.subplot(3, n, i)
    plt.imshow(x_test_noisy[i].reshape(28, 28))
    plt.gray()
    ax.get_xaxis().set_visible(False)
    if i==0:
        plt.ylabel("Original")
    else:
        ax.get_yaxis().set_visible(False)
        
    # Display reconstruction
    ax = plt.subplot(3, n, i + n)
    plt.imshow(decoded_imgs[i].reshape(28, 28))
    plt.gray()
    ax.get_xaxis().set_visible(False)
    if i==0:
        plt.ylabel("Vanilla Autoencoder")
    else:
        ax.get_yaxis().set_visible(False)
     
    ax = plt.subplot(3, n, i + 2*n)
    plt.imshow(decoded_imgs_denoised[i].reshape(28, 28))
    plt.gray()
    ax.get_xaxis().set_visible(False)
    if i==0:
        plt.ylabel("Denoising Autoencoder")
    else:
        ax.get_yaxis().set_visible(False)
    
        
plt.show()
313/313 [==============================] - 0s 1ms/step
313/313 [==============================] - 0s 1ms/step

decoded_imgs = noisy_autoencoder.predict(x_test_noisy)

n = 10
plt.figure(figsize=(20, 4))
for i in range(1, n + 1):
    # Display original
    ax = plt.subplot(2, n, i)
    plt.imshow(x_test_noisy[i].reshape(28, 28))
    plt.gray()
    ax.get_xaxis().set_visible(False)
    ax.get_yaxis().set_visible(False)

    # Display reconstruction
    ax = plt.subplot(2, n, i + n)
    plt.imshow(decoded_imgs[i].reshape(28, 28))
    plt.gray()
    ax.get_xaxis().set_visible(False)
    ax.get_yaxis().set_visible(False)
plt.show()
313/313 [==============================] - 0s 1ms/step