Cifar-10 據書上說是由深度學習大師 Geoffrey Hinton 教授與其在加拿大多倫多大學的學生 Alex Krixhevsky 與 Vinoid Nair 所整理之影像資料集, 包含 6 萬筆 32*32 低解析度之彩色圖片, 其中 5 萬筆為訓練集; 1 萬筆為測試集, 是機器學習中常用的圖片辨識資料集。Cifar-10 的所有圖片被分為 10 個類別 (以 0~9 數字作為 Label 之編碼):

A caption

載入資料集：

資料預處理

library(keras)

cifar<-dataset_cifar10()

#Training Data
train_x<-cifar$train$x / 255
train_y<-to_categorical(cifar$train$y,num_classes = 10)

#Test Data
test_x<-cifar$test$x/255
test_y<-to_categorical(cifar$test$y,num_classes = 10)

顯示資料結構（shape）

dim(train_x)

## [1] 50000    32    32     3

dim(test_x)

## [1] 10000    32    32     3

dim(train_y)

## [1] 50000    10

dim(test_y)

## [1] 10000    10

cat("No of training samples\t--",dim(train_x)[[1]] ,
    "\tNo of test samples\t--",dim(test_x)[[1]])#checking the

## No of training samples   -- 50000    No of test samples  -- 10000

建立模型：

model<-keras_model_sequential()

#建立線性堆疊模型, 加入兩層卷積 (丟棄 25% 神經元) + 池化層 :
model %>%
  layer_conv_2d(filter=32,kernel_size=c(3,3),padding="same",
                input_shape=c(32,32,3)) %>%
  layer_activation("relu") %>%
  layer_dropout(0.25) %>%
  layer_max_pooling_2d(pool_size=c(2,2)) %>%
  layer_conv_2d(filter=64,kernel_size=c(3,3),padding="same") %>%
  layer_activation("relu") %>%
  layer_dropout(0.25) %>%
  layer_max_pooling_2d(pool_size=c(2,2)) %>%

  #建立分類模型 (MLP) : 平坦層 + 隱藏層 (1024 神經元) + 輸出層 (10 神經元)
  layer_flatten() %>%
  layer_dropout(0.25) %>%
  layer_dense(1024) %>%
  layer_activation("relu") %>%
  layer_dropout(0.25) %>%
  layer_dense(10) %>%
  layer_activation("softmax")

檢視模型摘要

summary(model)

## ___________________________________________________________________________
## Layer (type)                     Output Shape                  Param #     
## ===========================================================================
## conv2d_1 (Conv2D)                (None, 32, 32, 32)            896         
## ___________________________________________________________________________
## activation_1 (Activation)        (None, 32, 32, 32)            0           
## ___________________________________________________________________________
## dropout_1 (Dropout)              (None, 32, 32, 32)            0           
## ___________________________________________________________________________
## max_pooling2d_1 (MaxPooling2D)   (None, 16, 16, 32)            0           
## ___________________________________________________________________________
## conv2d_2 (Conv2D)                (None, 16, 16, 64)            18496       
## ___________________________________________________________________________
## activation_2 (Activation)        (None, 16, 16, 64)            0           
## ___________________________________________________________________________
## dropout_2 (Dropout)              (None, 16, 16, 64)            0           
## ___________________________________________________________________________
## max_pooling2d_2 (MaxPooling2D)   (None, 8, 8, 64)              0           
## ___________________________________________________________________________
## flatten_1 (Flatten)              (None, 4096)                  0           
## ___________________________________________________________________________
## dropout_3 (Dropout)              (None, 4096)                  0           
## ___________________________________________________________________________
## dense_1 (Dense)                  (None, 1024)                  4195328     
## ___________________________________________________________________________
## activation_3 (Activation)        (None, 1024)                  0           
## ___________________________________________________________________________
## dropout_4 (Dropout)              (None, 1024)                  0           
## ___________________________________________________________________________
## dense_2 (Dense)                  (None, 10)                    10250       
## ___________________________________________________________________________
## activation_4 (Activation)        (None, 10)                    0           
## ===========================================================================
## Total params: 4,224,970
## Trainable params: 4,224,970
## Non-trainable params: 0
## ___________________________________________________________________________

編譯與訓練模型：

#Compiling the Model
model %>% compile(
  loss = "categorical_crossentropy",
  optimizer = 'adam',
  metrics = "accuracy"
)

#Fit the Model
history <- model %>% fit(
  train_x, train_y, 
  epochs = 10, batch_size = 128, 
  validation_split = 0.2,verbose = 2
)

訓練十次後模型在training data的準確率達到0.84，在test data的準確率達到0.74。

繪製訓練結果 :

plot(history)

評估預測準確率 :

model %>% evaluate(test_x, test_y)

## $loss
## [1] 0.7644399
## 
## $acc
## [1] 0.7384

proj5：cnn for image classification on cifar-10

group2 （zoey han）

2018/5/29