In [ ]:
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
In [ ]:
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Flatten
from tensorflow.keras.layers import Conv2D, MaxPool2D
from tensorflow.keras.optimizers import SGD, Adam
from tensorflow.keras.datasets import cifar10
from tensorflow.keras.utils import to_categorical
from tensorflow.keras.preprocessing.image import load_img
1. 讀入CIFAR--10資料集並進行資料預處理¶
CIFAR 10 是包含 10 種類的彩色小圖資料集，每張圖的尺寸為 $32\times32$
10 個類別分別是：飛機、交通工具、鳥、貓、鹿、狗、青蛙、馬、船、卡車
讀取 CIFAR 10 資料集
In [ ]:
class_name = ['飛機', '汽車', '鳥', '貓', '鹿', '狗', '青蛙', '馬', '船', '卡車']
In [ ]:
(x_train, y_train), (x_test, y_test) = cifar10.load_data()
x_train = x_train/255
x_test = x_test/255
y_train = to_categorical(y_train, 10)
y_test = to_categorical(y_test, 10)
In [ ]:
plt.imshow(x_test[5])
print(f"這是{class_name[y_test[5].argmax()]}");
這是青蛙
In [ ]:
y_test[5].flatten()
Out[ ]:
(10000, 1)
In [ ]:
x_train = x_train/255
x_test = x_test/255
y_train = to_categorical(y_train, 10)
y_test = to_categorical(y_test, 10)
2. 第一代：過去標準逐層打造神經網路的方法¶
在這個部分，我們將逐步帶領大家建立經典的 CNN 模型 LeNet-5 的變形。
LeNet-5 分成兩個部分，分別為卷積層與全連接層，兩部份之間是透過扁平層 (Flatten) ，將卷積層最後輸出的 2 維向量壓扁成 1 維向量。
In [ ]:
model = Sequential()
model.add(Conv2D(32, (3, 3), input_shape=(32, 32, 3), padding='same', activation='relu'))
model.add(MaxPool2D())
model.add(Conv2D(128, (3, 3), padding='same', activation='relu'))
model.add(MaxPool2D())
model.add(Conv2D(512, (3, 3), padding='same', activation='relu'))
model.add(MaxPool2D())
model.add(Flatten())
model.add(Dense(256, activation='relu'))
model.add(Dense(10, activation='softmax'))
In [ ]:
model.summary()
Model: "sequential"
___________________________________________________________________________
Layer (type) Output Shape Param #
===========================================================================
conv2d (Conv2D) (None, 32, 32, 32) 896
max_pooling2d (MaxPooling2D) (None, 16, 16, 32) 0
conv2d_1 (Conv2D) (None, 16, 16, 128) 36992
max_pooling2d_1 (MaxPooling2D) (None, 8, 8, 128) 0
conv2d_2 (Conv2D) (None, 8, 8, 512) 590336
max_pooling2d_2 (MaxPooling2D) (None, 4, 4, 512) 0
flatten (Flatten) (None, 8192) 0
dense (Dense) (None, 256) 2097408
dense_1 (Dense) (None, 10) 2570
===========================================================================
Total params: 2,728,202
Trainable params: 2,728,202
Non-trainable params: 0
___________________________________________________________________________
編譯模型: 設定模型訓練時的設定¶
- Optimizer: Stochastic Gradient Descent (SGD)
- Loss: categorical cross-entropy
In [ ]:
model.compile(loss='categorical_crossentropy',
optimizer=Adam(),
metrics=['accuracy'])
訓練模型: 透過訓練來學習分類資料的函數¶
In [ ]:
model.fit(x_train, y_train,
batch_size=128,
epochs=3,
validation_data=(x_test, y_test)
)
Epoch 1/3 391/391 [==============================] - 26s 38ms/step - loss: 1.4611 - accuracy: 0.4717 - val_loss: 1.1093 - val_accuracy: 0.6097 Epoch 2/3 391/391 [==============================] - 14s 36ms/step - loss: 0.9935 - accuracy: 0.6487 - val_loss: 1.0079 - val_accuracy: 0.6478 Epoch 3/3 391/391 [==============================] - 14s 36ms/step - loss: 0.8006 - accuracy: 0.7194 - val_loss: 0.7835 - val_accuracy: 0.7292
Out[ ]:
<keras.callbacks.History at 0x7f5e2c2fca10>
模型預測: 預測資料集的準確率¶
In [ ]:
score_train = model.evaluate(x_train, y_train)
score_test = model.evaluate(x_test, y_test)
1563/1563 [==============================] - 11s 7ms/step - loss: 0.6452 - accuracy: 0.7798 313/313 [==============================] - 2s 7ms/step - loss: 0.7835 - accuracy: 0.7292
In [ ]:
print(f'Train Accuracy: {score_train[1]*100}')
print(f'Test Accuracy: {score_test[1]*100}')
Train Accuracy: 77.97799706459045 Test Accuracy: 72.9200005531311
3. 第二代：用串列將神經網路層一次放進模型中¶
In [ ]:
many_layers = [Conv2D(32, (3, 3), input_shape=(32, 32, 3), padding='same', activation='relu'),
MaxPool2D(),
Conv2D(128, (3, 3), padding='same', activation='relu'),
MaxPool2D(),
Conv2D(512, (3, 3), padding='same', activation='relu'),
MaxPool2D(),
Flatten(),
Dense(256, activation='relu'),
Dense(10, activation='softmax'),
]
In [ ]:
second_model = Sequential(many_layers)
In [ ]:
second_model.summary()
Model: "sequential_1"
___________________________________________________________________________
Layer (type) Output Shape Param #
===========================================================================
conv2d_3 (Conv2D) (None, 32, 32, 32) 896
max_pooling2d_3 (MaxPooling2D) (None, 16, 16, 32) 0
conv2d_4 (Conv2D) (None, 16, 16, 128) 36992
max_pooling2d_4 (MaxPooling2D) (None, 8, 8, 128) 0
conv2d_5 (Conv2D) (None, 8, 8, 512) 590336
max_pooling2d_5 (MaxPooling2D) (None, 4, 4, 512) 0
flatten_1 (Flatten) (None, 8192) 0
dense_2 (Dense) (None, 256) 2097408
dense_3 (Dense) (None, 10) 2570
===========================================================================
Total params: 2,728,202
Trainable params: 2,728,202
Non-trainable params: 0
___________________________________________________________________________
4. 第三代：用兩個串列定義神經網路，再一次放進模型中¶
In [ ]:
CNN_layers = [Conv2D(32, (3, 3), input_shape=(32, 32, 3), padding='same',activation='relu'),
MaxPool2D(),
Conv2D(128, (3, 3), padding='same', activation='relu'),
MaxPool2D(),
Conv2D(512, (3, 3), padding='same', activation='relu'),
MaxPool2D(),
Flatten(),
]
In [ ]:
FC_layers = [Dense(units=256, activation='relu'),
Dense(units=10, activation='softmax'),
]
In [ ]:
third_model = Sequential(CNN_layers+FC_layers)
In [ ]:
third_model.summary()
Model: "sequential_2"
___________________________________________________________________________
Layer (type) Output Shape Param #
===========================================================================
conv2d_6 (Conv2D) (None, 32, 32, 32) 896
max_pooling2d_6 (MaxPooling2D) (None, 16, 16, 32) 0
conv2d_7 (Conv2D) (None, 16, 16, 128) 36992
max_pooling2d_7 (MaxPooling2D) (None, 8, 8, 128) 0
conv2d_8 (Conv2D) (None, 8, 8, 512) 590336
max_pooling2d_8 (MaxPooling2D) (None, 4, 4, 512) 0
flatten_2 (Flatten) (None, 8192) 0
dense_4 (Dense) (None, 256) 2097408
dense_5 (Dense) (None, 10) 2570
===========================================================================
Total params: 2,728,202
Trainable params: 2,728,202
Non-trainable params: 0
___________________________________________________________________________
5. 第四代：可以方便修改部份神經網路，不會動到別人！¶
In [ ]:
new_FC_layers = [Dense(1024, activation='relu'),
Dense(128, activation='relu'),
Dense(10, activation='softmax'),
]
In [ ]:
fourth_model = Sequential(CNN_layers+new_FC_layers)
In [ ]:
fourth_model.summary()
Model: "sequential_3"
___________________________________________________________________________
Layer (type) Output Shape Param #
===========================================================================
conv2d_6 (Conv2D) (None, 32, 32, 32) 896
max_pooling2d_6 (MaxPooling2D) (None, 16, 16, 32) 0
conv2d_7 (Conv2D) (None, 16, 16, 128) 36992
max_pooling2d_7 (MaxPooling2D) (None, 8, 8, 128) 0
conv2d_8 (Conv2D) (None, 8, 8, 512) 590336
max_pooling2d_8 (MaxPooling2D) (None, 4, 4, 512) 0
flatten_2 (Flatten) (None, 8192) 0
dense_6 (Dense) (None, 1024) 8389632
dense_7 (Dense) (None, 128) 131200
dense_8 (Dense) (None, 10) 1290
===========================================================================
Total params: 9,150,346
Trainable params: 9,150,346
Non-trainable params: 0
___________________________________________________________________________
In [ ]: