728x90
반응형
# fashionmnist
from tensorflow.keras.datasets.fashion_mnist import load_data
(x_train, y_train), (x_test, y_test) = load_data()
print(x_train.shape)
# (60000, 28, 28)
y_train[:10]
# array([9, 0, 0, 3, 0, 2, 7, 2, 5, 5], dtype=uint8)
class_names = ['T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat', 'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle boot']
# 임의의 데이터를 이미지로 출력
import matplotlib.pyplot as plt
import numpy as np
sample_size = 9
# 난수 생성 0 ~ 59999 중 9개
random_idx = np.random.randint(60000, size = sample_size)
plt.figure(figsize = (5, 5))
for i, idx in enumerate(random_idx) :
plt.subplot(3,3, i + 1)
plt.xticks([])
plt.yticks([])
plt.imshow(x_train[idx], cmap = 'gray')
plt.xlabel(class_names[y_train[idx]])
plt.show()
# 학습데이터 테스트 데이터 정규화 0~1
x_train[:10]
x_train = x_train / 255
x_test = x_test / 255
from tensorflow.keras.utils import to_categorical
y_train = to_categorical(y_train)
y_test = to_categorical(y_test)
from sklearn.model_selection import train_test_split
X_train, X_val, Y_train, Y_val = train_test_split(x_train, y_train, test_size=0.3, random_state = 777)
print('훈련데이터 :',X_train.shape,', 레이블 :',Y_train.shape)
print('검증데이터 :',X_val.shape,', 레이블 :',Y_val.shape)
# 훈련데이터 : (42000, 28, 28) , 레이블 : (42000, 10)
# 검증데이터 : (18000, 28, 28) , 레이블 : (18000, 10)
Model 1 : 입력층 64, 은닉층 32 출력층 10
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Flatten
model1 = Sequential()
model1.add(Flatten(input_shape=(28, 28))) # 2차원을 1차원으로 변경, 입력층 vs .reshape으로 1차원으로 바꿀필요없음
model1.add(Dense(64, activation = 'relu'))
model1.add(Dense(32, activation = 'relu'))
model1.add(Dense(10, activation = 'softmax'))
model1.summary() # 입력층
# 첫번째 은닉층 64
# 출력층 32
# 출력층 10
Model: "sequential_3"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
flatten_3 (Flatten) (None, 784) 0
_________________________________________________________________
dense_9 (Dense) (None, 64) 50240
_________________________________________________________________
dense_10 (Dense) (None, 32) 2080
_________________________________________________________________
dense_11 (Dense) (None, 10) 330
=================================================================
Total params: 52,650
Trainable params: 52,650
Non-trainable params: 0
_________________________________________________________________
model1.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['acc'])
history1 = model1.fit(x_train, y_train, epochs=30, batch_size=128, validation_data=(X_val, Y_val))
Epoch 1/30
469/469 [==============================] - 1s 3ms/step - loss: 0.8711 - acc: 0.6973 - val_loss: 0.4393 - val_acc: 0.8425
Epoch 2/30
469/469 [==============================] - 1s 2ms/step - loss: 0.4209 - acc: 0.8495 - val_loss: 0.3694 - val_acc: 0.8717
Epoch 3/30
469/469 [==============================] - 1s 3ms/step - loss: 0.3815 - acc: 0.8622 - val_loss: 0.3443 - val_acc: 0.8754
Epoch 4/30
469/469 [==============================] - 1s 2ms/step - loss: 0.3505 - acc: 0.8727 - val_loss: 0.3208 - val_acc: 0.8841
Epoch 5/30
469/469 [==============================] - 1s 2ms/step - loss: 0.3327 - acc: 0.8785 - val_loss: 0.3102 - val_acc: 0.8882
Epoch 6/30
469/469 [==============================] - 1s 2ms/step - loss: 0.3199 - acc: 0.8824 - val_loss: 0.2903 - val_acc: 0.8952
Epoch 7/30
469/469 [==============================] - 1s 2ms/step - loss: 0.3057 - acc: 0.8892 - val_loss: 0.2904 - val_acc: 0.8952
Epoch 8/30
469/469 [==============================] - 1s 2ms/step - loss: 0.2986 - acc: 0.8901 - val_loss: 0.2779 - val_acc: 0.8995
Epoch 9/30
469/469 [==============================] - 1s 2ms/step - loss: 0.2860 - acc: 0.8940 - val_loss: 0.2606 - val_acc: 0.9059
Epoch 10/30
469/469 [==============================] - 1s 2ms/step - loss: 0.2795 - acc: 0.8974 - val_loss: 0.2608 - val_acc: 0.9048
Epoch 11/30
469/469 [==============================] - 1s 2ms/step - loss: 0.2768 - acc: 0.8970 - val_loss: 0.2661 - val_acc: 0.9025
Epoch 12/30
469/469 [==============================] - 1s 2ms/step - loss: 0.2607 - acc: 0.9033 - val_loss: 0.2498 - val_acc: 0.9096
Epoch 13/30
469/469 [==============================] - 1s 2ms/step - loss: 0.2628 - acc: 0.9039 - val_loss: 0.2418 - val_acc: 0.9116
Epoch 14/30
469/469 [==============================] - 1s 2ms/step - loss: 0.2509 - acc: 0.9073 - val_loss: 0.2443 - val_acc: 0.9102
Epoch 15/30
469/469 [==============================] - 1s 2ms/step - loss: 0.2412 - acc: 0.9120 - val_loss: 0.2269 - val_acc: 0.9183
Epoch 16/30
469/469 [==============================] - 1s 2ms/step - loss: 0.2372 - acc: 0.9133 - val_loss: 0.2412 - val_acc: 0.9125
Epoch 17/30
469/469 [==============================] - 1s 2ms/step - loss: 0.2391 - acc: 0.9112 - val_loss: 0.2172 - val_acc: 0.9212
Epoch 18/30
469/469 [==============================] - 1s 2ms/step - loss: 0.2262 - acc: 0.9171 - val_loss: 0.2078 - val_acc: 0.9261
Epoch 19/30
469/469 [==============================] - 1s 2ms/step - loss: 0.2258 - acc: 0.9153 - val_loss: 0.2031 - val_acc: 0.9273
Epoch 20/30
469/469 [==============================] - 1s 2ms/step - loss: 0.2245 - acc: 0.9169 - val_loss: 0.2010 - val_acc: 0.9285
Epoch 21/30
469/469 [==============================] - 1s 2ms/step - loss: 0.2173 - acc: 0.9191 - val_loss: 0.2006 - val_acc: 0.9281
Epoch 22/30
469/469 [==============================] - 1s 2ms/step - loss: 0.2162 - acc: 0.9197 - val_loss: 0.2037 - val_acc: 0.9268
Epoch 23/30
469/469 [==============================] - 1s 2ms/step - loss: 0.2087 - acc: 0.9242 - val_loss: 0.1952 - val_acc: 0.9304
Epoch 24/30
469/469 [==============================] - 1s 2ms/step - loss: 0.2038 - acc: 0.9250 - val_loss: 0.1937 - val_acc: 0.9313
Epoch 25/30
469/469 [==============================] - 1s 2ms/step - loss: 0.2038 - acc: 0.9245 - val_loss: 0.2027 - val_acc: 0.9263
Epoch 26/30
469/469 [==============================] - 1s 2ms/step - loss: 0.2032 - acc: 0.9229 - val_loss: 0.2045 - val_acc: 0.9246
Epoch 27/30
469/469 [==============================] - 1s 3ms/step - loss: 0.1944 - acc: 0.9291 - val_loss: 0.1839 - val_acc: 0.9336
Epoch 28/30
469/469 [==============================] - 1s 2ms/step - loss: 0.1925 - acc: 0.9286 - val_loss: 0.1966 - val_acc: 0.9277
Epoch 29/30
469/469 [==============================] - 1s 2ms/step - loss: 0.1818 - acc: 0.9326 - val_loss: 0.1905 - val_acc: 0.9293
Epoch 30/30
469/469 [==============================] - 1s 2ms/step - loss: 0.1872 - acc: 0.9304 - val_loss: 0.1950 - val_acc: 0.9287
model1.evaluate(x_test,y_test)
313/313 [==============================] - 0s 1ms/step - loss: 0.3881 - acc: 0.8763
[0.3880555033683777, 0.8762999773025513]
results = model1.predict(x_test)
# 혼돈행렬 조회하기
from sklearn.metrics import classification_report, confusion_matrix
import numpy as np
# np.argmax(y_test, axis = -1) y_test의 최대값 인덱스 리턴, 인덱스 숫자
cm = confusion_matrix(np.argmax(y_test, axis = -1), np.argmax(results, axis = -1))
cm
array([[891, 2, 14, 11, 4, 2, 72, 0, 4, 0],
[ 10, 971, 1, 10, 4, 0, 4, 0, 0, 0],
[ 22, 1, 787, 12, 91, 0, 87, 0, 0, 0],
[ 37, 8, 13, 869, 37, 1, 31, 0, 4, 0],
[ 4, 0, 99, 23, 815, 1, 57, 0, 1, 0],
[ 0, 0, 0, 1, 0, 952, 0, 20, 1, 26],
[159, 1, 77, 25, 77, 0, 657, 0, 4, 0],
[ 0, 0, 0, 0, 0, 20, 0, 884, 2, 94],
[ 10, 1, 4, 4, 12, 3, 11, 4, 951, 0],
[ 0, 0, 0, 0, 0, 5, 1, 8, 0, 986]], dtype=int64)
import matplotlib.pyplot as plt
import seaborn as sns
plt.figure(figsize = (7,7))
sns.heatmap(cm, annot = True, fmt = 'd', cmap = 'Blues')
plt.xlabel('predicted label', fontsize = 15)
plt.ylabel('true label', fontsize = 15)
plt.show()
df.value_counts().sort_index()
0 1000
1 1000
2 1000
3 1000
4 1000
5 1000
6 1000
7 1000
8 1000
9 1000
dtype: int64
print(classification_report(np.argmax(y_test, axis = -1), np.argmax(results, axis = -1)))
precision recall f1-score support
0 0.79 0.89 0.84 1000
1 0.99 0.97 0.98 1000
2 0.79 0.79 0.79 1000
3 0.91 0.87 0.89 1000
4 0.78 0.81 0.80 1000
5 0.97 0.95 0.96 1000
6 0.71 0.66 0.68 1000
7 0.97 0.88 0.92 1000
8 0.98 0.95 0.97 1000
9 0.89 0.99 0.94 1000
accuracy 0.88 10000
macro avg 0.88 0.88 0.88 10000
weighted avg 0.88 0.88 0.88 10000
Model 2 구성하기 ,입력층 출력 128, 은닉층 64, 은닉층 32
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Flatten
model2 = Sequential()
model2.add(Flatten(input_shape=(28, 28))) # 2차원을 1차원으로 변경, 입력층 vs .reshape으로 1차원으로 바꿀필요없음
model2.add(Dense(128, activation = 'relu'))
model2.add(Dense(64, activation = 'relu'))
model2.add(Dense(32, activation = 'relu'))
model2.add(Dense(10, activation = 'softmax'))
model2.summary()
Model: "sequential_2"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
flatten_2 (Flatten) (None, 784) 0
_________________________________________________________________
dense_6 (Dense) (None, 128) 100480
_________________________________________________________________
dense_7 (Dense) (None, 64) 8256
_________________________________________________________________
dense_8 (Dense) (None, 32) 2080
_________________________________________________________________
dense_9 (Dense) (None, 10) 330
=================================================================
Total params: 111,146
Trainable params: 111,146
Non-trainable params: 0
_________________________________________________________________
model2.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['acc'])
histroy2 = model2.fit(x_train, y_train, epochs=30, batch_size=128, validation_data=(X_val, Y_val))
Epoch 1/30
469/469 [==============================] - 2s 3ms/step - loss: 0.8033 - acc: 0.7328 - val_loss: 0.3923 - val_acc: 0.8634
Epoch 2/30
469/469 [==============================] - 1s 3ms/step - loss: 0.4000 - acc: 0.8577 - val_loss: 0.3466 - val_acc: 0.8776
Epoch 3/30
469/469 [==============================] - 1s 3ms/step - loss: 0.3499 - acc: 0.8721 - val_loss: 0.3355 - val_acc: 0.8779
Epoch 4/30
469/469 [==============================] - 1s 3ms/step - loss: 0.3283 - acc: 0.8814 - val_loss: 0.3091 - val_acc: 0.8868
Epoch 5/30
469/469 [==============================] - 1s 3ms/step - loss: 0.3073 - acc: 0.8877 - val_loss: 0.2801 - val_acc: 0.8997
Epoch 6/30
469/469 [==============================] - 1s 3ms/step - loss: 0.2897 - acc: 0.8933 - val_loss: 0.2657 - val_acc: 0.9041
Epoch 7/30
469/469 [==============================] - 1s 3ms/step - loss: 0.2751 - acc: 0.8974 - val_loss: 0.2663 - val_acc: 0.9022
Epoch 8/30
469/469 [==============================] - 1s 3ms/step - loss: 0.2692 - acc: 0.8995 - val_loss: 0.2529 - val_acc: 0.9076
Epoch 9/30
469/469 [==============================] - 1s 3ms/step - loss: 0.2634 - acc: 0.9024 - val_loss: 0.2363 - val_acc: 0.9133
Epoch 10/30
469/469 [==============================] - 1s 3ms/step - loss: 0.2438 - acc: 0.9089 - val_loss: 0.2339 - val_acc: 0.9136
Epoch 11/30
469/469 [==============================] - 1s 3ms/step - loss: 0.2448 - acc: 0.9083 - val_loss: 0.2393 - val_acc: 0.9090
Epoch 12/30
469/469 [==============================] - 1s 3ms/step - loss: 0.2355 - acc: 0.9112 - val_loss: 0.2355 - val_acc: 0.9114
Epoch 13/30
469/469 [==============================] - 1s 3ms/step - loss: 0.2243 - acc: 0.9148 - val_loss: 0.2158 - val_acc: 0.9202
Epoch 14/30
469/469 [==============================] - 1s 3ms/step - loss: 0.2173 - acc: 0.9186 - val_loss: 0.2080 - val_acc: 0.9226
Epoch 15/30
469/469 [==============================] - 1s 3ms/step - loss: 0.2168 - acc: 0.9197 - val_loss: 0.1917 - val_acc: 0.9295
Epoch 16/30
469/469 [==============================] - 1s 3ms/step - loss: 0.2055 - acc: 0.9223 - val_loss: 0.1866 - val_acc: 0.9332
Epoch 17/30
469/469 [==============================] - 1s 3ms/step - loss: 0.1978 - acc: 0.9242 - val_loss: 0.1891 - val_acc: 0.9302
Epoch 18/30
469/469 [==============================] - 1s 3ms/step - loss: 0.2004 - acc: 0.9248 - val_loss: 0.1708 - val_acc: 0.9376
Epoch 19/30
469/469 [==============================] - 1s 3ms/step - loss: 0.1906 - acc: 0.9285 - val_loss: 0.1878 - val_acc: 0.9294
Epoch 20/30
469/469 [==============================] - 1s 3ms/step - loss: 0.1884 - acc: 0.9303 - val_loss: 0.1774 - val_acc: 0.9341
Epoch 21/30
469/469 [==============================] - 1s 3ms/step - loss: 0.1841 - acc: 0.9300 - val_loss: 0.1714 - val_acc: 0.9340
Epoch 22/30
469/469 [==============================] - 1s 3ms/step - loss: 0.1739 - acc: 0.9330 - val_loss: 0.1576 - val_acc: 0.9419
Epoch 23/30
469/469 [==============================] - 1s 3ms/step - loss: 0.1723 - acc: 0.9344 - val_loss: 0.1664 - val_acc: 0.9353
Epoch 24/30
469/469 [==============================] - 1s 3ms/step - loss: 0.1663 - acc: 0.9377 - val_loss: 0.1593 - val_acc: 0.9411
Epoch 25/30
469/469 [==============================] - 1s 3ms/step - loss: 0.1593 - acc: 0.9401 - val_loss: 0.1602 - val_acc: 0.9388
Epoch 26/30
469/469 [==============================] - 1s 3ms/step - loss: 0.1598 - acc: 0.9394 - val_loss: 0.1670 - val_acc: 0.9378
Epoch 27/30
469/469 [==============================] - 1s 3ms/step - loss: 0.1606 - acc: 0.9389 - val_loss: 0.1579 - val_acc: 0.9382
Epoch 28/30
469/469 [==============================] - 1s 3ms/step - loss: 0.1464 - acc: 0.9441 - val_loss: 0.1777 - val_acc: 0.9341
Epoch 29/30
469/469 [==============================] - 1s 3ms/step - loss: 0.1471 - acc: 0.9452 - val_loss: 0.1376 - val_acc: 0.9475
Epoch 30/30
469/469 [==============================] - 1s 3ms/step - loss: 0.1513 - acc: 0.9430 - val_loss: 0.1364 - val_acc: 0.9483
model2.evaluate(x_test,y_test)
313/313 [==============================] - 0s 1ms/step - loss: 0.3984 - acc: 0.8906
[0.398380845785141, 0.8906000256538391]
results = model2.predict(x_test)
# 혼돈행렬 조회하기
from sklearn.metrics import classification_report, confusion_matrix
import numpy as np
# np.argmax(y_test, axis = -1) y_test의 최대값 인덱스 리턴, 인덱스 숫자
cm = confusion_matrix(np.argmax(y_test, axis = -1), np.argmax(results, axis = -1))
cm
array([[860, 1, 11, 26, 5, 1, 89, 0, 7, 0],
[ 6, 973, 0, 15, 4, 0, 2, 0, 0, 0],
[ 21, 1, 793, 14, 89, 1, 78, 0, 3, 0],
[ 21, 3, 7, 900, 40, 0, 22, 0, 7, 0],
[ 1, 0, 78, 21, 856, 1, 41, 0, 2, 0],
[ 0, 0, 0, 1, 0, 979, 0, 14, 1, 5],
[131, 1, 60, 42, 76, 0, 677, 0, 13, 0],
[ 0, 0, 0, 0, 0, 27, 0, 930, 0, 43],
[ 6, 0, 4, 4, 2, 2, 4, 3, 974, 1],
[ 0, 0, 0, 0, 0, 15, 1, 20, 0, 964]], dtype=int64)
import matplotlib.pyplot as plt
import seaborn as sns
plt.figure(figsize = (7,7))
sns.heatmap(cm, annot = True, fmt = 'd', cmap = 'Blues')
plt.xlabel('predicted label', fontsize = 15)
plt.ylabel('true label', fontsize = 15)
plt.show()
print(classification_report(np.argmax(y_test, axis = -1), np.argmax(results, axis = -1)))
precision recall f1-score support
0 0.82 0.86 0.84 1000
1 0.99 0.97 0.98 1000
2 0.83 0.79 0.81 1000
3 0.88 0.90 0.89 1000
4 0.80 0.86 0.83 1000
5 0.95 0.98 0.97 1000
6 0.74 0.68 0.71 1000
7 0.96 0.93 0.95 1000
8 0.97 0.97 0.97 1000
9 0.95 0.96 0.96 1000
accuracy 0.89 10000
macro avg 0.89 0.89 0.89 10000
weighted avg 0.89 0.89 0.89 10000
모델 비교하기
import numpy as np
import matplotlib.pyplot as plt
def draw_loss_acc(history_1, history_2, epochs) :
his_dict_1 = history_1.history
his_dict_2 = history_2.history
keys = list(his_dict_1.keys()) # ['loss', 'acc', 'val_loss', 'val_acc']
epochs = range(1, epochs)
fig = plt.figure(figsize = (10,10))
ax = fig.add_subplot(1,1,1) # 여러개를 하나의 그래프로
ax.spines['top'].set_color('none') # c로 바꾸면 외곽선이 그려짐
ax.spines['bottom'].set_color('none')
ax.spines['left'].set_color('none')
ax.spines['right'].set_color('none')
ax.tick_params(labelcolor='w', top=False, bottom=False, left=False, right=False)
for i in range(len(his_dict_1)) :
temp_ax = fig.add_subplot(2,2, i + 1)
temp = keys[i%2]
val_temp = keys[(i+2)%2 +2]
temp_history = his_dict_1 if i < 2 else his_dict_2
temp_ax.plot(epochs, temp_history[temp][1:], color = 'blue', label ='train_'+temp) # 손실함수 그래프
temp_ax.plot(epochs, temp_history[val_temp][1:], color = 'orange', label = val_temp)
if(i==1 or i==3) : # 정확도인 경우
start, end = temp_ax.get_ylim()
temp_ax.yaxis.set_ticks(np.arange(np.round(start, 2), end, 0.01))
temp_ax.legend()
ax.set_ylabel('loss', size = 20)
ax.set_xlabel('Epochs', size = 20)
plt.tight_layout()
plt.show()
draw_loss_acc(history1, history2, 30)
history1.history.keys()
history2.history.keys()
# dict_keys(['loss', 'acc', 'val_loss', 'val_acc'])
반응형
'Data_Science > Data_Analysis_Py' 카테고리의 다른 글
| 44. 보스턴주택가격 회귀예측 || kfold (0) | 2021.11.25 |
|---|---|
| 43. 보스턴주택가격 딥러닝 회귀 예측 || hold-out (0) | 2021.11.25 |
| 41. MNIST 딥러닝 예측 (0) | 2021.11.25 |
| 40. Tensorflow를 통한 논리구현 (0) | 2021.11.25 |
| 39. Tensorflow 구현 (0) | 2021.11.25 |