Naver Project (fashion_MNIST-keras improvement)

import tensorflow as tf
from tensorflow.keras import layers
from tensorflow.keras import callbacks

from sklearn.model_selection import train_test_split

import numpy as np
import matplotlib.pyplot as plt

tf.__version__

#2.15.0

데이터 로드

• TF에서 제공하는 데이터셋을 load해 간단한 전처리 진행

# Load training and eval data from tf.keras
(train_data, train_labels), (test_data, test_labels) = \
    tf.keras.datasets.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

print(train_data.shape, train_labels.shape)

# (60000, 28, 28) (60000,)

train_data, valid_data, train_labels, valid_labels = \
    train_test_split(train_data, train_labels, test_size=0.1, shuffle=True)

print(train_data.shape, train_labels.shape)

# (54000, 28, 28) (54000,)

print(valid_data.shape, valid_labels.shape)

# (6000, 28, 28) (6000,)

index = 5901
print("label = {}".format(valid_labels[index]))
plt.imshow(valid_data[index].reshape(28, 28))
plt.colorbar()
#plt.gca().grid(False)
plt.show()

train_data = train_data / 255.
train_data = train_data.reshape(-1, 28 * 28)
train_data = train_data.astype(np.float32)
train_labels = train_labels.astype(np.int32)

test_data = test_data / 255.
test_data = test_data.reshape(-1, 28 * 28)
test_data = test_data.astype(np.float32)
test_labels = test_labels.astype(np.int32)

valid_data = valid_data / 255.
valid_data = valid_data.reshape(-1, 28 * 28)
valid_data = valid_data.astype(np.float32)
valid_labels = valid_labels.astype(np.int32)


print(train_data.shape, train_labels.shape)
print(test_data.shape, test_labels.shape)
print(valid_data.shape, valid_labels.shape)

#(54000, 784) (54000,)
#(10000, 784) (10000,)
#(6000, 784) (6000,)

학습에 사용할 tf.data.Dataset 구성

• 학습에 잘 적용할 수 있도록 label 처리
• 데이터셋 구성

def one_hot_label(image, label):
  label = tf.one_hot(label, depth=10)
  return image, label

batch_size = 32

# for train
train_dataset = tf.data.Dataset.from_tensor_slices((train_data, train_labels))
train_dataset = train_dataset.shuffle(buffer_size=10000)
train_dataset = train_dataset.map(one_hot_label)
train_dataset = train_dataset.repeat().batch(batch_size=batch_size)
print(train_dataset)

# for test
test_dataset = tf.data.Dataset.from_tensor_slices((test_data, test_labels))
test_dataset = test_dataset.map(one_hot_label)
test_dataset = test_dataset.batch(batch_size=batch_size)
print(test_dataset)

# for test
valid_dataset = tf.data.Dataset.from_tensor_slices((valid_data, valid_labels))
valid_dataset = valid_dataset.map(one_hot_label)
valid_dataset = valid_dataset.batch(batch_size=batch_size)
print(valid_dataset)

#<_BatchDataset element_spec=(TensorSpec(shape=(None, 784), dtype=tf.float32, name=None), TensorSpec(shape=(None, 10), dtype=tf.float32, name=None))>
#<_BatchDataset element_spec=(TensorSpec(shape=(None, 784), dtype=tf.float32, name=None), TensorSpec(shape=(None, 10), dtype=tf.float32, name=None))>
#<_BatchDataset element_spec=(TensorSpec(shape=(None, 784), dtype=tf.float32, name=None), TensorSpec(shape=(None, 10), dtype=tf.float32, name=None))>

모델 구성

# Dense 64 relu
# Dense 32 relu
# Output Layer
# 28 * 28
model = tf.keras.Sequential([
    layers.Dense(64, kernel_initializer=tf.keras.initializers.HeUniform(),
                 kernel_regularizer=tf.keras.regularizers.L2(0.0001)),
    layers.BatchNormalization(),
    layers.Activation("relu"),

    layers.Dropout(0.5),

    layers.Dense(32, kernel_initializer=tf.keras.initializers.HeUniform(),
                 kernel_regularizer=tf.keras.regularizers.L2(0.0001)),
    layers.BatchNormalization(),
    layers.Activation("relu"),

    layers.Dropout(0.5),

    layers.Dense(16, kernel_initializer=tf.keras.initializers.HeUniform(),
                 kernel_regularizer=tf.keras.regularizers.L2(0.0001)),
    layers.BatchNormalization(),
    layers.Activation("relu"),

    layers.Dense(10),
])

model.compile(optimizer=tf.keras.optimizers.Adam(1e-4),
              loss=tf.keras.losses.CategoricalCrossentropy(from_logits=True),
              metrics=['accuracy'])

# without training, just inference a model in eager execution:
predictions = model(train_data[0:1], training=False)
print("Predictions: ", predictions.numpy())

#Predictions:  [[ 0.06548186 -0.08698994  0.17317785 -0.17119434  0.05337192 -0.14751902
#  -0.0592468   0.01507654  0.49552554 -0.04181848]]

model.summary()

Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 dense (Dense)               (1, 64)                   50240     
                                                                 
 batch_normalization (Batch  (1, 64)                   256       
 Normalization)                                                  
                                                                 
 activation (Activation)     (1, 64)                   0         
                                                                 
 dropout (Dropout)           (1, 64)                   0         
                                                                 
 dense_1 (Dense)             (1, 32)                   2080      
                                                                 
 batch_normalization_1 (Bat  (1, 32)                   128       
 chNormalization)                                                
                                                                 
 activation_1 (Activation)   (1, 32)                   0         
                                                                 
 dropout_1 (Dropout)         (1, 32)                   0         
                                                                 
 dense_2 (Dense)             (1, 16)                   528       
                                                                 
 batch_normalization_2 (Bat  (1, 16)                   64        
 chNormalization)                                                
                                                                 
 activation_2 (Activation)   (1, 16)                   0         
                                                                 
 dense_3 (Dense)             (1, 10)                   170       
                                                                 
=================================================================
Total params: 53466 (208.85 KB)
Trainable params: 53242 (207.98 KB)
Non-trainable params: 224 (896.00 Byte)
_________________________________________________________________

모델 학습 구성

check_point_cb = callbacks.ModelCheckpoint('fashion_mnist_model.h5',
                                           save_best_only=True,
                                           verbose=1)
early_stopping_cb = callbacks.EarlyStopping(patience=10,
                                            monitor='val_loss',
                                            restore_best_weights=True,
                                            verbose=1)

max_epochs = 100
# using `tf.data.Dataset`
history = model.fit(train_dataset,
                    epochs=max_epochs,
                    steps_per_epoch=len(train_data) // batch_size,
                    validation_data=valid_dataset,
                    validation_steps=len(valid_data) // batch_size,
                    callbacks=[check_point_cb, early_stopping_cb]
                   )

Epoch 1/100
1686/1687 [============================>.] - ETA: 0s - loss: 1.9787 - accuracy: 0.3221
Epoch 1: val_loss improved from inf to 1.54169, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 18s 8ms/step - loss: 1.9786 - accuracy: 0.3222 - val_loss: 1.5417 - val_accuracy: 0.6832
Epoch 2/100
  19/1687 [..............................] - ETA: 9s - loss: 1.6848 - accuracy: 0.5197/usr/local/lib/python3.10/dist-packages/keras/src/engine/training.py:3103: UserWarning: You are saving your model as an HDF5 file via `model.save()`. This file format is considered legacy. We recommend using instead the native Keras format, e.g. `model.save('my_model.keras')`.
  saving_api.save_model(
1684/1687 [============================>.] - ETA: 0s - loss: 1.5489 - accuracy: 0.5151
Epoch 2: val_loss improved from 1.54169 to 1.13681, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 13s 7ms/step - loss: 1.5485 - accuracy: 0.5153 - val_loss: 1.1368 - val_accuracy: 0.7482
Epoch 3/100
1684/1687 [============================>.] - ETA: 0s - loss: 1.2833 - accuracy: 0.5949
Epoch 3: val_loss improved from 1.13681 to 0.88925, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 12s 7ms/step - loss: 1.2832 - accuracy: 0.5950 - val_loss: 0.8892 - val_accuracy: 0.7834
Epoch 4/100
1685/1687 [============================>.] - ETA: 0s - loss: 1.1196 - accuracy: 0.6378
Epoch 4: val_loss improved from 0.88925 to 0.73826, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 13s 7ms/step - loss: 1.1195 - accuracy: 0.6378 - val_loss: 0.7383 - val_accuracy: 0.8015
Epoch 5/100
1682/1687 [============================>.] - ETA: 0s - loss: 1.0132 - accuracy: 0.6667
Epoch 5: val_loss improved from 0.73826 to 0.65151, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 12s 7ms/step - loss: 1.0129 - accuracy: 0.6668 - val_loss: 0.6515 - val_accuracy: 0.8155
Epoch 6/100
1686/1687 [============================>.] - ETA: 0s - loss: 0.9460 - accuracy: 0.6859
Epoch 6: val_loss improved from 0.65151 to 0.59941, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 12s 7ms/step - loss: 0.9459 - accuracy: 0.6860 - val_loss: 0.5994 - val_accuracy: 0.8254
Epoch 7/100
1687/1687 [==============================] - ETA: 0s - loss: 0.8999 - accuracy: 0.6978
Epoch 7: val_loss improved from 0.59941 to 0.56490, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 13s 8ms/step - loss: 0.8999 - accuracy: 0.6978 - val_loss: 0.5649 - val_accuracy: 0.8260
Epoch 8/100
1681/1687 [============================>.] - ETA: 0s - loss: 0.8703 - accuracy: 0.7097
Epoch 8: val_loss improved from 0.56490 to 0.54167, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 12s 7ms/step - loss: 0.8704 - accuracy: 0.7098 - val_loss: 0.5417 - val_accuracy: 0.8282
Epoch 9/100
1687/1687 [==============================] - ETA: 0s - loss: 0.8425 - accuracy: 0.7179
Epoch 9: val_loss improved from 0.54167 to 0.52763, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 13s 8ms/step - loss: 0.8425 - accuracy: 0.7179 - val_loss: 0.5276 - val_accuracy: 0.8336
Epoch 10/100
1681/1687 [============================>.] - ETA: 0s - loss: 0.8223 - accuracy: 0.7264
Epoch 10: val_loss improved from 0.52763 to 0.51580, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 12s 7ms/step - loss: 0.8222 - accuracy: 0.7265 - val_loss: 0.5158 - val_accuracy: 0.8327
Epoch 11/100
1686/1687 [============================>.] - ETA: 0s - loss: 0.8054 - accuracy: 0.7294
Epoch 11: val_loss improved from 0.51580 to 0.50027, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 12s 7ms/step - loss: 0.8051 - accuracy: 0.7295 - val_loss: 0.5003 - val_accuracy: 0.8374
Epoch 12/100
1685/1687 [============================>.] - ETA: 0s - loss: 0.7895 - accuracy: 0.7349
Epoch 12: val_loss improved from 0.50027 to 0.49493, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 13s 7ms/step - loss: 0.7895 - accuracy: 0.7349 - val_loss: 0.4949 - val_accuracy: 0.8374
Epoch 13/100
1683/1687 [============================>.] - ETA: 0s - loss: 0.7781 - accuracy: 0.7379
Epoch 13: val_loss improved from 0.49493 to 0.48376, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 12s 7ms/step - loss: 0.7782 - accuracy: 0.7381 - val_loss: 0.4838 - val_accuracy: 0.8483
Epoch 14/100
1684/1687 [============================>.] - ETA: 0s - loss: 0.7615 - accuracy: 0.7451
Epoch 14: val_loss improved from 0.48376 to 0.47872, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 13s 7ms/step - loss: 0.7612 - accuracy: 0.7451 - val_loss: 0.4787 - val_accuracy: 0.8427
Epoch 15/100
1685/1687 [============================>.] - ETA: 0s - loss: 0.7548 - accuracy: 0.7482
Epoch 15: val_loss improved from 0.47872 to 0.47074, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 12s 7ms/step - loss: 0.7552 - accuracy: 0.7481 - val_loss: 0.4707 - val_accuracy: 0.8476
Epoch 16/100
1687/1687 [==============================] - ETA: 0s - loss: 0.7464 - accuracy: 0.7511
Epoch 16: val_loss improved from 0.47074 to 0.46636, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 12s 7ms/step - loss: 0.7464 - accuracy: 0.7511 - val_loss: 0.4664 - val_accuracy: 0.8514
Epoch 17/100
1680/1687 [============================>.] - ETA: 0s - loss: 0.7330 - accuracy: 0.7567
Epoch 17: val_loss improved from 0.46636 to 0.46387, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 12s 7ms/step - loss: 0.7330 - accuracy: 0.7567 - val_loss: 0.4639 - val_accuracy: 0.8499
Epoch 18/100
1684/1687 [============================>.] - ETA: 0s - loss: 0.7344 - accuracy: 0.7539
Epoch 18: val_loss improved from 0.46387 to 0.45543, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 12s 7ms/step - loss: 0.7343 - accuracy: 0.7540 - val_loss: 0.4554 - val_accuracy: 0.8486
Epoch 19/100
1679/1687 [============================>.] - ETA: 0s - loss: 0.7278 - accuracy: 0.7569
Epoch 19: val_loss improved from 0.45543 to 0.45203, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 13s 8ms/step - loss: 0.7281 - accuracy: 0.7569 - val_loss: 0.4520 - val_accuracy: 0.8488
Epoch 20/100
1680/1687 [============================>.] - ETA: 0s - loss: 0.7171 - accuracy: 0.7615
Epoch 20: val_loss improved from 0.45203 to 0.44627, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 12s 7ms/step - loss: 0.7175 - accuracy: 0.7613 - val_loss: 0.4463 - val_accuracy: 0.8504
Epoch 21/100
1686/1687 [============================>.] - ETA: 0s - loss: 0.7107 - accuracy: 0.7653
Epoch 21: val_loss improved from 0.44627 to 0.44421, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 12s 7ms/step - loss: 0.7108 - accuracy: 0.7653 - val_loss: 0.4442 - val_accuracy: 0.8511
Epoch 22/100
1686/1687 [============================>.] - ETA: 0s - loss: 0.7080 - accuracy: 0.7634
Epoch 22: val_loss did not improve from 0.44421
1687/1687 [==============================] - 13s 8ms/step - loss: 0.7081 - accuracy: 0.7634 - val_loss: 0.4459 - val_accuracy: 0.8526
Epoch 23/100
1682/1687 [============================>.] - ETA: 0s - loss: 0.7038 - accuracy: 0.7676
Epoch 23: val_loss improved from 0.44421 to 0.44307, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 12s 7ms/step - loss: 0.7035 - accuracy: 0.7676 - val_loss: 0.4431 - val_accuracy: 0.8528
Epoch 24/100
1687/1687 [==============================] - ETA: 0s - loss: 0.7007 - accuracy: 0.7666
Epoch 24: val_loss improved from 0.44307 to 0.43942, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 13s 8ms/step - loss: 0.7007 - accuracy: 0.7666 - val_loss: 0.4394 - val_accuracy: 0.8573
Epoch 25/100
1683/1687 [============================>.] - ETA: 0s - loss: 0.6908 - accuracy: 0.7707
Epoch 25: val_loss improved from 0.43942 to 0.43583, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 12s 7ms/step - loss: 0.6906 - accuracy: 0.7708 - val_loss: 0.4358 - val_accuracy: 0.8571
Epoch 26/100
1685/1687 [============================>.] - ETA: 0s - loss: 0.6857 - accuracy: 0.7716
Epoch 26: val_loss improved from 0.43583 to 0.43411, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 12s 7ms/step - loss: 0.6857 - accuracy: 0.7716 - val_loss: 0.4341 - val_accuracy: 0.8595
Epoch 27/100
1683/1687 [============================>.] - ETA: 0s - loss: 0.6888 - accuracy: 0.7742
Epoch 27: val_loss improved from 0.43411 to 0.43279, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 13s 8ms/step - loss: 0.6885 - accuracy: 0.7743 - val_loss: 0.4328 - val_accuracy: 0.8554
Epoch 28/100
1681/1687 [============================>.] - ETA: 0s - loss: 0.6808 - accuracy: 0.7747
Epoch 28: val_loss improved from 0.43279 to 0.43242, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 12s 7ms/step - loss: 0.6811 - accuracy: 0.7746 - val_loss: 0.4324 - val_accuracy: 0.8553
Epoch 29/100
1684/1687 [============================>.] - ETA: 0s - loss: 0.6812 - accuracy: 0.7752
Epoch 29: val_loss did not improve from 0.43242
1687/1687 [==============================] - 13s 8ms/step - loss: 0.6815 - accuracy: 0.7751 - val_loss: 0.4331 - val_accuracy: 0.8556
Epoch 30/100
1683/1687 [============================>.] - ETA: 0s - loss: 0.6644 - accuracy: 0.7808
Epoch 30: val_loss improved from 0.43242 to 0.43040, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 12s 7ms/step - loss: 0.6645 - accuracy: 0.7807 - val_loss: 0.4304 - val_accuracy: 0.8580
Epoch 31/100
1681/1687 [============================>.] - ETA: 0s - loss: 0.6676 - accuracy: 0.7771
Epoch 31: val_loss did not improve from 0.43040
1687/1687 [==============================] - 13s 8ms/step - loss: 0.6678 - accuracy: 0.7771 - val_loss: 0.4308 - val_accuracy: 0.8598
Epoch 32/100
1687/1687 [==============================] - ETA: 0s - loss: 0.6671 - accuracy: 0.7812
Epoch 32: val_loss did not improve from 0.43040
1687/1687 [==============================] - 13s 8ms/step - loss: 0.6671 - accuracy: 0.7812 - val_loss: 0.4310 - val_accuracy: 0.8573
Epoch 33/100
1685/1687 [============================>.] - ETA: 0s - loss: 0.6650 - accuracy: 0.7790
Epoch 33: val_loss improved from 0.43040 to 0.42394, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 12s 7ms/step - loss: 0.6651 - accuracy: 0.7790 - val_loss: 0.4239 - val_accuracy: 0.8625
Epoch 34/100
1684/1687 [============================>.] - ETA: 0s - loss: 0.6649 - accuracy: 0.7794
Epoch 34: val_loss did not improve from 0.42394
1687/1687 [==============================] - 13s 8ms/step - loss: 0.6647 - accuracy: 0.7795 - val_loss: 0.4246 - val_accuracy: 0.8570
Epoch 35/100
1686/1687 [============================>.] - ETA: 0s - loss: 0.6571 - accuracy: 0.7823
Epoch 35: val_loss improved from 0.42394 to 0.42265, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 12s 7ms/step - loss: 0.6570 - accuracy: 0.7824 - val_loss: 0.4226 - val_accuracy: 0.8593
Epoch 36/100
1684/1687 [============================>.] - ETA: 0s - loss: 0.6544 - accuracy: 0.7816
Epoch 36: val_loss improved from 0.42265 to 0.41752, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 12s 7ms/step - loss: 0.6544 - accuracy: 0.7817 - val_loss: 0.4175 - val_accuracy: 0.8593
Epoch 37/100
1681/1687 [============================>.] - ETA: 0s - loss: 0.6616 - accuracy: 0.7829
Epoch 37: val_loss did not improve from 0.41752
1687/1687 [==============================] - 13s 7ms/step - loss: 0.6613 - accuracy: 0.7830 - val_loss: 0.4178 - val_accuracy: 0.8615
Epoch 38/100
1681/1687 [============================>.] - ETA: 0s - loss: 0.6490 - accuracy: 0.7881
Epoch 38: val_loss did not improve from 0.41752
1687/1687 [==============================] - 12s 7ms/step - loss: 0.6487 - accuracy: 0.7882 - val_loss: 0.4201 - val_accuracy: 0.8611
Epoch 39/100
1687/1687 [==============================] - ETA: 0s - loss: 0.6416 - accuracy: 0.7876
Epoch 39: val_loss improved from 0.41752 to 0.41550, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 14s 8ms/step - loss: 0.6416 - accuracy: 0.7876 - val_loss: 0.4155 - val_accuracy: 0.8616
Epoch 40/100
1687/1687 [==============================] - ETA: 0s - loss: 0.6444 - accuracy: 0.7864
Epoch 40: val_loss did not improve from 0.41550
1687/1687 [==============================] - 12s 7ms/step - loss: 0.6444 - accuracy: 0.7864 - val_loss: 0.4159 - val_accuracy: 0.8601
Epoch 41/100
1682/1687 [============================>.] - ETA: 0s - loss: 0.6465 - accuracy: 0.7845
Epoch 41: val_loss improved from 0.41550 to 0.41065, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 13s 8ms/step - loss: 0.6463 - accuracy: 0.7846 - val_loss: 0.4107 - val_accuracy: 0.8665
Epoch 42/100
1687/1687 [==============================] - ETA: 0s - loss: 0.6415 - accuracy: 0.7893
Epoch 42: val_loss did not improve from 0.41065
1687/1687 [==============================] - 12s 7ms/step - loss: 0.6415 - accuracy: 0.7893 - val_loss: 0.4139 - val_accuracy: 0.8606
Epoch 43/100
1685/1687 [============================>.] - ETA: 0s - loss: 0.6405 - accuracy: 0.7869
Epoch 43: val_loss did not improve from 0.41065
1687/1687 [==============================] - 12s 7ms/step - loss: 0.6405 - accuracy: 0.7869 - val_loss: 0.4120 - val_accuracy: 0.8618
Epoch 44/100
1682/1687 [============================>.] - ETA: 0s - loss: 0.6380 - accuracy: 0.7886
Epoch 44: val_loss did not improve from 0.41065
1687/1687 [==============================] - 13s 8ms/step - loss: 0.6378 - accuracy: 0.7888 - val_loss: 0.4163 - val_accuracy: 0.8633
Epoch 45/100
1682/1687 [============================>.] - ETA: 0s - loss: 0.6377 - accuracy: 0.7878
Epoch 45: val_loss improved from 0.41065 to 0.40604, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 12s 7ms/step - loss: 0.6375 - accuracy: 0.7878 - val_loss: 0.4060 - val_accuracy: 0.8641
Epoch 46/100
1681/1687 [============================>.] - ETA: 0s - loss: 0.6363 - accuracy: 0.7908
Epoch 46: val_loss did not improve from 0.40604
1687/1687 [==============================] - 12s 7ms/step - loss: 0.6362 - accuracy: 0.7908 - val_loss: 0.4166 - val_accuracy: 0.8603
Epoch 47/100
1684/1687 [============================>.] - ETA: 0s - loss: 0.6338 - accuracy: 0.7893
Epoch 47: val_loss did not improve from 0.40604
1687/1687 [==============================] - 12s 7ms/step - loss: 0.6339 - accuracy: 0.7893 - val_loss: 0.4092 - val_accuracy: 0.8640
Epoch 48/100
1683/1687 [============================>.] - ETA: 0s - loss: 0.6337 - accuracy: 0.7905
Epoch 48: val_loss did not improve from 0.40604
1687/1687 [==============================] - 12s 7ms/step - loss: 0.6338 - accuracy: 0.7904 - val_loss: 0.4096 - val_accuracy: 0.8640
Epoch 49/100
1685/1687 [============================>.] - ETA: 0s - loss: 0.6307 - accuracy: 0.7917
Epoch 49: val_loss did not improve from 0.40604
1687/1687 [==============================] - 13s 8ms/step - loss: 0.6308 - accuracy: 0.7917 - val_loss: 0.4065 - val_accuracy: 0.8646
Epoch 50/100
1683/1687 [============================>.] - ETA: 0s - loss: 0.6253 - accuracy: 0.7925
Epoch 50: val_loss improved from 0.40604 to 0.40466, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 12s 7ms/step - loss: 0.6254 - accuracy: 0.7925 - val_loss: 0.4047 - val_accuracy: 0.8636
Epoch 51/100
1687/1687 [==============================] - ETA: 0s - loss: 0.6268 - accuracy: 0.7930
Epoch 51: val_loss improved from 0.40466 to 0.40251, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 13s 8ms/step - loss: 0.6268 - accuracy: 0.7930 - val_loss: 0.4025 - val_accuracy: 0.8655
Epoch 52/100
1685/1687 [============================>.] - ETA: 0s - loss: 0.6265 - accuracy: 0.7952
Epoch 52: val_loss improved from 0.40251 to 0.40179, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 12s 7ms/step - loss: 0.6266 - accuracy: 0.7951 - val_loss: 0.4018 - val_accuracy: 0.8668
Epoch 53/100
1687/1687 [==============================] - ETA: 0s - loss: 0.6264 - accuracy: 0.7926
Epoch 53: val_loss did not improve from 0.40179
1687/1687 [==============================] - 12s 7ms/step - loss: 0.6264 - accuracy: 0.7926 - val_loss: 0.4083 - val_accuracy: 0.8641
Epoch 54/100
1685/1687 [============================>.] - ETA: 0s - loss: 0.6178 - accuracy: 0.7955
Epoch 54: val_loss did not improve from 0.40179
1687/1687 [==============================] - 14s 8ms/step - loss: 0.6177 - accuracy: 0.7956 - val_loss: 0.4056 - val_accuracy: 0.8655
Epoch 55/100
1687/1687 [==============================] - ETA: 0s - loss: 0.6191 - accuracy: 0.7956
Epoch 55: val_loss did not improve from 0.40179
1687/1687 [==============================] - 12s 7ms/step - loss: 0.6191 - accuracy: 0.7956 - val_loss: 0.4059 - val_accuracy: 0.8626
Epoch 56/100
1680/1687 [============================>.] - ETA: 0s - loss: 0.6176 - accuracy: 0.7949
Epoch 56: val_loss improved from 0.40179 to 0.39865, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 13s 8ms/step - loss: 0.6179 - accuracy: 0.7948 - val_loss: 0.3986 - val_accuracy: 0.8688
Epoch 57/100
1687/1687 [==============================] - ETA: 0s - loss: 0.6134 - accuracy: 0.7966
Epoch 57: val_loss did not improve from 0.39865
1687/1687 [==============================] - 12s 7ms/step - loss: 0.6134 - accuracy: 0.7966 - val_loss: 0.4041 - val_accuracy: 0.8656
Epoch 58/100
1679/1687 [============================>.] - ETA: 0s - loss: 0.6156 - accuracy: 0.7962
Epoch 58: val_loss did not improve from 0.39865
1687/1687 [==============================] - 12s 7ms/step - loss: 0.6154 - accuracy: 0.7963 - val_loss: 0.4014 - val_accuracy: 0.8646
Epoch 59/100
1684/1687 [============================>.] - ETA: 0s - loss: 0.6198 - accuracy: 0.7955
Epoch 59: val_loss did not improve from 0.39865
1687/1687 [==============================] - 13s 8ms/step - loss: 0.6200 - accuracy: 0.7955 - val_loss: 0.3991 - val_accuracy: 0.8718
Epoch 60/100
1685/1687 [============================>.] - ETA: 0s - loss: 0.6065 - accuracy: 0.7995
Epoch 60: val_loss did not improve from 0.39865
1687/1687 [==============================] - 12s 7ms/step - loss: 0.6063 - accuracy: 0.7995 - val_loss: 0.4042 - val_accuracy: 0.8676
Epoch 61/100
1682/1687 [============================>.] - ETA: 0s - loss: 0.6096 - accuracy: 0.7983
Epoch 61: val_loss improved from 0.39865 to 0.39738, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 13s 8ms/step - loss: 0.6095 - accuracy: 0.7982 - val_loss: 0.3974 - val_accuracy: 0.8707
Epoch 62/100
1686/1687 [============================>.] - ETA: 0s - loss: 0.6109 - accuracy: 0.7982
Epoch 62: val_loss did not improve from 0.39738
1687/1687 [==============================] - 12s 7ms/step - loss: 0.6110 - accuracy: 0.7982 - val_loss: 0.4011 - val_accuracy: 0.8685
Epoch 63/100
1684/1687 [============================>.] - ETA: 0s - loss: 0.6168 - accuracy: 0.7941
Epoch 63: val_loss did not improve from 0.39738
1687/1687 [==============================] - 13s 8ms/step - loss: 0.6168 - accuracy: 0.7941 - val_loss: 0.4047 - val_accuracy: 0.8666
Epoch 64/100
1683/1687 [============================>.] - ETA: 0s - loss: 0.6083 - accuracy: 0.7989
Epoch 64: val_loss did not improve from 0.39738
1687/1687 [==============================] - 12s 7ms/step - loss: 0.6085 - accuracy: 0.7989 - val_loss: 0.3979 - val_accuracy: 0.8698
Epoch 65/100
1681/1687 [============================>.] - ETA: 0s - loss: 0.6144 - accuracy: 0.7989
Epoch 65: val_loss did not improve from 0.39738
1687/1687 [==============================] - 12s 7ms/step - loss: 0.6149 - accuracy: 0.7987 - val_loss: 0.3985 - val_accuracy: 0.8686
Epoch 66/100
1681/1687 [============================>.] - ETA: 0s - loss: 0.6077 - accuracy: 0.7980
Epoch 66: val_loss improved from 0.39738 to 0.39638, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 13s 8ms/step - loss: 0.6078 - accuracy: 0.7980 - val_loss: 0.3964 - val_accuracy: 0.8698
Epoch 67/100
1679/1687 [============================>.] - ETA: 0s - loss: 0.6067 - accuracy: 0.7998
Epoch 67: val_loss did not improve from 0.39638
1687/1687 [==============================] - 12s 7ms/step - loss: 0.6068 - accuracy: 0.7997 - val_loss: 0.4008 - val_accuracy: 0.8660
Epoch 68/100
1683/1687 [============================>.] - ETA: 0s - loss: 0.6047 - accuracy: 0.8016
Epoch 68: val_loss did not improve from 0.39638
1687/1687 [==============================] - 13s 8ms/step - loss: 0.6046 - accuracy: 0.8015 - val_loss: 0.3990 - val_accuracy: 0.8663
Epoch 69/100
1679/1687 [============================>.] - ETA: 0s - loss: 0.6097 - accuracy: 0.8002
Epoch 69: val_loss did not improve from 0.39638
1687/1687 [==============================] - 12s 7ms/step - loss: 0.6102 - accuracy: 0.8000 - val_loss: 0.4010 - val_accuracy: 0.8653
Epoch 70/100
1685/1687 [============================>.] - ETA: 0s - loss: 0.6059 - accuracy: 0.8001
Epoch 70: val_loss improved from 0.39638 to 0.39021, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 12s 7ms/step - loss: 0.6058 - accuracy: 0.8002 - val_loss: 0.3902 - val_accuracy: 0.8730
Epoch 71/100
1684/1687 [============================>.] - ETA: 0s - loss: 0.5997 - accuracy: 0.8018
Epoch 71: val_loss did not improve from 0.39021
1687/1687 [==============================] - 13s 8ms/step - loss: 0.5997 - accuracy: 0.8018 - val_loss: 0.3953 - val_accuracy: 0.8656
Epoch 72/100
1685/1687 [============================>.] - ETA: 0s - loss: 0.6036 - accuracy: 0.8013
Epoch 72: val_loss did not improve from 0.39021
1687/1687 [==============================] - 12s 7ms/step - loss: 0.6033 - accuracy: 0.8014 - val_loss: 0.4027 - val_accuracy: 0.8635
Epoch 73/100
1682/1687 [============================>.] - ETA: 0s - loss: 0.5982 - accuracy: 0.8021
Epoch 73: val_loss did not improve from 0.39021
1687/1687 [==============================] - 13s 8ms/step - loss: 0.5988 - accuracy: 0.8019 - val_loss: 0.3925 - val_accuracy: 0.8681
Epoch 74/100
1683/1687 [============================>.] - ETA: 0s - loss: 0.5987 - accuracy: 0.7999
Epoch 74: val_loss did not improve from 0.39021
1687/1687 [==============================] - 12s 7ms/step - loss: 0.5985 - accuracy: 0.8000 - val_loss: 0.3909 - val_accuracy: 0.8727
Epoch 75/100
1684/1687 [============================>.] - ETA: 0s - loss: 0.5989 - accuracy: 0.8030
Epoch 75: val_loss did not improve from 0.39021
1687/1687 [==============================] - 13s 7ms/step - loss: 0.5989 - accuracy: 0.8029 - val_loss: 0.3950 - val_accuracy: 0.8676
Epoch 76/100
1687/1687 [==============================] - ETA: 0s - loss: 0.5968 - accuracy: 0.8019
Epoch 76: val_loss did not improve from 0.39021
1687/1687 [==============================] - 13s 8ms/step - loss: 0.5968 - accuracy: 0.8019 - val_loss: 0.3956 - val_accuracy: 0.8703
Epoch 77/100
1683/1687 [============================>.] - ETA: 0s - loss: 0.5910 - accuracy: 0.8051
Epoch 77: val_loss did not improve from 0.39021
1687/1687 [==============================] - 12s 7ms/step - loss: 0.5913 - accuracy: 0.8050 - val_loss: 0.3925 - val_accuracy: 0.8727
Epoch 78/100
1680/1687 [============================>.] - ETA: 0s - loss: 0.5893 - accuracy: 0.8048
Epoch 78: val_loss did not improve from 0.39021
1687/1687 [==============================] - 14s 8ms/step - loss: 0.5890 - accuracy: 0.8049 - val_loss: 0.3943 - val_accuracy: 0.8700
Epoch 79/100
1683/1687 [============================>.] - ETA: 0s - loss: 0.5933 - accuracy: 0.8041
Epoch 79: val_loss improved from 0.39021 to 0.39005, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 12s 7ms/step - loss: 0.5929 - accuracy: 0.8043 - val_loss: 0.3900 - val_accuracy: 0.8728
Epoch 80/100
1687/1687 [==============================] - ETA: 0s - loss: 0.5934 - accuracy: 0.8035
Epoch 80: val_loss did not improve from 0.39005
1687/1687 [==============================] - 13s 7ms/step - loss: 0.5934 - accuracy: 0.8035 - val_loss: 0.3917 - val_accuracy: 0.8723
Epoch 81/100
1687/1687 [==============================] - ETA: 0s - loss: 0.5964 - accuracy: 0.8018
Epoch 81: val_loss did not improve from 0.39005
1687/1687 [==============================] - 13s 8ms/step - loss: 0.5964 - accuracy: 0.8018 - val_loss: 0.3980 - val_accuracy: 0.8688
Epoch 82/100
1679/1687 [============================>.] - ETA: 0s - loss: 0.5947 - accuracy: 0.8015
Epoch 82: val_loss did not improve from 0.39005
1687/1687 [==============================] - 12s 7ms/step - loss: 0.5951 - accuracy: 0.8013 - val_loss: 0.3945 - val_accuracy: 0.8735
Epoch 83/100
1687/1687 [==============================] - ETA: 0s - loss: 0.5892 - accuracy: 0.8050
Epoch 83: val_loss did not improve from 0.39005
1687/1687 [==============================] - 14s 8ms/step - loss: 0.5892 - accuracy: 0.8050 - val_loss: 0.3955 - val_accuracy: 0.8722
Epoch 84/100
1681/1687 [============================>.] - ETA: 0s - loss: 0.5897 - accuracy: 0.8036
Epoch 84: val_loss did not improve from 0.39005
1687/1687 [==============================] - 12s 7ms/step - loss: 0.5900 - accuracy: 0.8034 - val_loss: 0.3904 - val_accuracy: 0.8748
Epoch 85/100
1681/1687 [============================>.] - ETA: 0s - loss: 0.5892 - accuracy: 0.8065
Epoch 85: val_loss did not improve from 0.39005
1687/1687 [==============================] - 13s 8ms/step - loss: 0.5890 - accuracy: 0.8066 - val_loss: 0.3909 - val_accuracy: 0.8710
Epoch 86/100
1685/1687 [============================>.] - ETA: 0s - loss: 0.5895 - accuracy: 0.8046
Epoch 86: val_loss did not improve from 0.39005
1687/1687 [==============================] - 12s 7ms/step - loss: 0.5895 - accuracy: 0.8046 - val_loss: 0.3923 - val_accuracy: 0.8732
Epoch 87/100
1686/1687 [============================>.] - ETA: 0s - loss: 0.5907 - accuracy: 0.8049
Epoch 87: val_loss improved from 0.39005 to 0.38910, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 14s 8ms/step - loss: 0.5907 - accuracy: 0.8049 - val_loss: 0.3891 - val_accuracy: 0.8727
Epoch 88/100
1679/1687 [============================>.] - ETA: 0s - loss: 0.5870 - accuracy: 0.8033
Epoch 88: val_loss improved from 0.38910 to 0.38872, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 12s 7ms/step - loss: 0.5869 - accuracy: 0.8033 - val_loss: 0.3887 - val_accuracy: 0.8740
Epoch 89/100
1680/1687 [============================>.] - ETA: 0s - loss: 0.5916 - accuracy: 0.8038
Epoch 89: val_loss did not improve from 0.38872
1687/1687 [==============================] - 12s 7ms/step - loss: 0.5916 - accuracy: 0.8038 - val_loss: 0.3929 - val_accuracy: 0.8707
Epoch 90/100
1683/1687 [============================>.] - ETA: 0s - loss: 0.5889 - accuracy: 0.8050
Epoch 90: val_loss did not improve from 0.38872
1687/1687 [==============================] - 13s 8ms/step - loss: 0.5891 - accuracy: 0.8049 - val_loss: 0.3921 - val_accuracy: 0.8708
Epoch 91/100
1686/1687 [============================>.] - ETA: 0s - loss: 0.5874 - accuracy: 0.8043
Epoch 91: val_loss improved from 0.38872 to 0.38847, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 12s 7ms/step - loss: 0.5874 - accuracy: 0.8043 - val_loss: 0.3885 - val_accuracy: 0.8702
Epoch 92/100
1680/1687 [============================>.] - ETA: 0s - loss: 0.5843 - accuracy: 0.8048
Epoch 92: val_loss improved from 0.38847 to 0.38727, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 13s 8ms/step - loss: 0.5844 - accuracy: 0.8046 - val_loss: 0.3873 - val_accuracy: 0.8755
Epoch 93/100
1679/1687 [============================>.] - ETA: 0s - loss: 0.5885 - accuracy: 0.8036
Epoch 93: val_loss did not improve from 0.38727
1687/1687 [==============================] - 12s 7ms/step - loss: 0.5886 - accuracy: 0.8035 - val_loss: 0.3893 - val_accuracy: 0.8708
Epoch 94/100
1686/1687 [============================>.] - ETA: 0s - loss: 0.5821 - accuracy: 0.8073
Epoch 94: val_loss did not improve from 0.38727
1687/1687 [==============================] - 13s 8ms/step - loss: 0.5821 - accuracy: 0.8073 - val_loss: 0.3894 - val_accuracy: 0.8723
Epoch 95/100
1682/1687 [============================>.] - ETA: 0s - loss: 0.5850 - accuracy: 0.8055
Epoch 95: val_loss improved from 0.38727 to 0.38255, saving model to fashion_mnist_model.h5
1687/1687 [==============================] - 13s 8ms/step - loss: 0.5850 - accuracy: 0.8054 - val_loss: 0.3825 - val_accuracy: 0.8745
Epoch 96/100
1684/1687 [============================>.] - ETA: 0s - loss: 0.5781 - accuracy: 0.8086
Epoch 96: val_loss did not improve from 0.38255
1687/1687 [==============================] - 12s 7ms/step - loss: 0.5780 - accuracy: 0.8086 - val_loss: 0.3882 - val_accuracy: 0.8732
Epoch 97/100
1686/1687 [============================>.] - ETA: 0s - loss: 0.5875 - accuracy: 0.8031
Epoch 97: val_loss did not improve from 0.38255
1687/1687 [==============================] - 15s 9ms/step - loss: 0.5875 - accuracy: 0.8031 - val_loss: 0.3889 - val_accuracy: 0.8712
Epoch 98/100
1682/1687 [============================>.] - ETA: 0s - loss: 0.5813 - accuracy: 0.8077
Epoch 98: val_loss did not improve from 0.38255
1687/1687 [==============================] - 12s 7ms/step - loss: 0.5811 - accuracy: 0.8077 - val_loss: 0.3937 - val_accuracy: 0.8713
Epoch 99/100
1682/1687 [============================>.] - ETA: 0s - loss: 0.5830 - accuracy: 0.8060
Epoch 99: val_loss did not improve from 0.38255
1687/1687 [==============================] - 15s 9ms/step - loss: 0.5829 - accuracy: 0.8059 - val_loss: 0.3918 - val_accuracy: 0.8727
Epoch 100/100
1685/1687 [============================>.] - ETA: 0s - loss: 0.5868 - accuracy: 0.8051
Epoch 100: val_loss did not improve from 0.38255
1687/1687 [==============================] - 12s 7ms/step - loss: 0.5869 - accuracy: 0.8050 - val_loss: 0.3861 - val_accuracy: 0.8702

# 코드 수정
acc = history.history['accuracy']
val_acc = history.history['val_accuracy']

loss = history.history['loss']
val_loss = history.history['val_loss']

epochs_range = range(len(acc))

plt.figure(figsize=(8, 8))
plt.subplot(1, 2, 1)
plt.plot(epochs_range, acc, label='Training Accuracy')
plt.plot(epochs_range, val_acc, label='Validation Accuracy')
plt.legend(loc='lower right')
plt.title('Training and Valid Accuracy')

plt.subplot(1, 2, 2)
plt.plot(epochs_range, loss, label='Training Loss')
plt.plot(epochs_range, val_loss, label='Validation Loss')
plt.legend(loc='upper right')
plt.title('Training and Valid Loss')
plt.show()