行動技術與應用

```{r eval=FALSE}
# 安裝'devtools' package：方便從github安裝套件
if(!"devtools" %in% installed.packages())  
  install.packages('devtools')

require(devtools)

# install tensorflow(如果你要tensorflow的話)
devtools::install_github("rstudio/tensorflow") 

# installing keras(如果你要keras的話)
devtools::install_github("rstudio/keras") 
```

```{r eval=FALSE}
# 測試tensorflow: Hello
require(tensorflow)
session = tf$Session()
hello <- tf$constant('Hi, TensorFlow!')
session$run(hello)
```

```{r eval=FALSE}
library(keras)
model <- keras_model_sequential()
model %>%
  layer_dense(units = 256, activation = 'relu', input_shape= c(784)) %>%
  layer_dropout(rate = 0.4) %>%
  layer_dense(units = 128, activation = 'relu') %>%
  layer_dropout(rate = 0.3) %>%
  layer_dense(units = 10, activation = 'softmax')
summary(model)
```

```{r eval=FALSE}
if(!"devtools" %in% installed.packages()) 
  install.packages("devtools");

if(!"keras" %in% installed.packages()) {
  require(devtools);
  install_github("rstudio/keras");
}

library(keras)
install_keras()  # 如需GPU設定，參考文件網址
```

```{r eval=FALSE}
fashion_mnist <- dataset_fashion_mnist()

# 訓練圖片/訓練標籤, 測試圖片/測試標籤
c(train_images, train_labels) %<-% fashion_mnist$train
c(test_images, test_labels) %<-% fashion_mnist$test

class_names = c('T-shirt/top',
                'Trouser',
                'Pullover',
                'Dress',
                'Coat', 
                'Sandal',
                'Shirt',
                'Sneaker',
                'Bag',
                'Ankle boot')
```

```{r eval=FALSE}
dim(train_images)
dim(train_labels)
train_labels[1:20]
dim(test_images)
dim(test_labels)

train_images <- train_images / 255
test_images <- test_images / 255
```

```{r eval=FALSE}
# 建立模型
model <- keras_model_sequential()
model %>%
  layer_flatten(input_shape = c(28, 28)) %>%
  layer_dense(units = 128, activation = 'relu') %>%
  layer_dense(units = 10, activation = 'softmax')

# 編譯模型
model %>% compile(
  optimizer = 'adam', 
  loss = 'sparse_categorical_crossentropy',
  metrics = c('accuracy')
)
```

```{r eval=FALSE}
# 訓練模型
model %>% fit(train_images, train_labels, epochs = 5)
```

```{r eval=FALSE}
# 評估正確性
score <- model %>% evaluate(test_images, test_labels)

cat('Test loss:', score$loss, "\n")
cat('Test accuracy:', score$acc, "\n")
```

```{r eval=FALSE}
# 檢視一筆預測結果
predictions <- model %>% predict(test_images)  # 10000筆測試集套用模型預測
predictions[1, ]                               # 印出第一筆
answer <- which.max(predictions[1, ]) - 1      # 預測機率最高者 - 1
test_labels[1]                                 # 陣列index:1-10,對應test_labels: 0-9

# 印出25筆預測結果
par(mfcol=c(5,5))
par(mar=c(0, 0, 1.5, 0), xaxs='i', yaxs='i')
for (i in 1:25) { 
  img <- test_images[i, , ]
  img <- t(apply(img, 2, rev)) 
  # subtract 1 as labels go from 0 to 9
  predicted_label <- which.max(predictions[i, ]) - 1
  true_label <- test_labels[i]
  if (predicted_label == true_label) {
    color <- '#008800' 
  } else {
    color <- '#bb0000'
  }
  image(1:28, 1:28, img, col = gray((0:255)/255), xaxt = 'n', yaxt = 'n',
        main = paste0(class_names[predicted_label + 1], " (",
                      class_names[true_label + 1], ")"),
        col.main = color)
}
```

```{r eval=FALSE}
# 預測新影像
img <- test_images[1, , , drop = FALSE]  # 此處以測試集第一張替代
dim(img)                                 # 28*28矩陣  0~1數值:灰階
predictions <- model %>% predict(img)    # 以模型進行預測
predictions                              # 分類機率串列(1~10)
# 減1 因為標籤為0～9
which.max(predictions)-1                 # 預測答案   
class_pred <- model %>% predict_classes(img)
class_pred                               # 標準答案
```

---
title: "Fashion MINST"
output: html_document
---
## 安裝 Keras R (rstudio/keras)
```{r eval=FALSE}
if(!"devtools" %in% installed.packages()) 
  install.packages("devtools");

if(!"keras" %in% installed.packages()) {
  require(devtools);
  install_github("rstudio/keras");
}

if(!"ggplot2" %in% installed.packages()) {
  install.packages("ggplot2");
}
```


```{r eval=FALSE}
## 1.載入Keras
library(keras)
# install_keras()
```

```{r eval=FALSE}
## 2.準備資料：載入fashion mnist資料庫: 含訓練集/測試
fashion_mnist <- dataset_fashion_mnist()

# 訓練圖片/訓練標籤, 測試圖片/測試標籤
c(train_images, train_labels) %<-% fashion_mnist$train
c(test_images, test_labels) %<-% fashion_mnist$test

class_names = c('T-shirt/top',
                'Trouser',
                'Pullover',
                'Dress',
                'Coat', 
                'Sandal',
                'Shirt',
                'Sneaker',
                'Bag',
                'Ankle boot')
```

```{r eval=FALSE}
## 3.準備資料：檢視資料/資料前處理
dim(train_images)
dim(train_labels)
train_labels[1:20]
dim(test_images)
dim(test_labels)

train_images <- train_images / 255
test_images <- test_images / 255
```

```{r eval=FALSE}
# 建立模型
model <- keras_model_sequential()
model %>%
  layer_flatten(input_shape = c(28, 28)) %>%
  layer_dense(units = 128, activation = 'relu') %>%
  layer_dense(units = 10, activation = 'softmax')

# 編譯模型
model %>% compile(
  optimizer = 'adam', 
  loss = 'sparse_categorical_crossentropy',
  metrics = c('accuracy')
)
```

```{r eval=FALSE}
# 訓練模型
model %>% fit(train_images, train_labels, epochs = 5)

# 評估正確性
score <- model %>% evaluate(test_images, test_labels)

cat('Test loss:', score$loss, "\n")
cat('Test accuracy:', score$acc, "\n")
```

```{r eval=FALSE}
# 檢視一筆預測結果
predictions <- model %>% predict(test_images)  # 10000筆測試集套用模型預測
predictions[1, ]                               # 印出第一筆
answer <- which.max(predictions[1, ]) - 1      # 預測機率最高者 - 1
test_labels[1]                                 # 陣列index:1-10,對應test_labels: 0-9

# 印出25筆預測結果
par(mfcol=c(5,5))
par(mar=c(0, 0, 1.5, 0), xaxs='i', yaxs='i')
for (i in 1:25) { 
  img <- test_images[i, , ]
  img <- t(apply(img, 2, rev)) 
  # subtract 1 as labels go from 0 to 9
  predicted_label <- which.max(predictions[i, ]) - 1
  true_label <- test_labels[i]
  if (predicted_label == true_label) {
    color <- '#008800' 
  } else {
    color <- '#bb0000'
  }
  image(1:28, 1:28, img, col = gray((0:255)/255), xaxt = 'n', yaxt = 'n',
        main = paste0(class_names[predicted_label + 1], " (",
                      class_names[true_label + 1], ")"),
        col.main = color)
}
```

```{r eval=FALSE}
# 預測新影像
img <- test_images[1, , , drop = FALSE]  # 此處以測試集第一張替代
dim(img)                                 # 28*28矩陣  0~1數值:灰階
predictions <- model %>% predict(img)    # 以模型進行預測
predictions                              # 分類機率串列(1~10)
# 減1 因為標籤為0～9
which.max(predictions)-1                 # 預測答案   
class_pred <- model %>% predict_classes(img)
class_pred                               # 標準答案
```

```{r eval=FALSE}
# 建立模型
model <- keras_model_sequential()
model %>%
  layer_flatten(input_shape = c(28, 28)) %>%
  layer_dense(units = 128, activation = 'relu') %>%
  layer_dense(units = 10, activation = 'softmax')
```

```{r eval=FALSE}
# 訓練模型
model %>% fit(train_images, train_labels, epochs = 5)
```

```{r eval=FALSE}
# 編譯模型
model %>% compile(
  optimizer = 'adam', 
  loss = 'sparse_categorical_crossentropy',
  metrics = c('accuracy')
)
```

model %>% compile(
  optimizer = 'adam', 
  loss = 'sparse_categorical_crossentropy',
  metrics = c('accuracy')
)

model <- keras_model_sequential() %>%
  # 卷積層C1
  layer_conv_2d(filters = 32, kernel_size = c(3,3), activation = 'relu',
                input_shape = input_shape) %>% 
  # 卷積層C2
  layer_conv_2d(filters = 64, kernel_size = c(3,3), activation = 'relu') %>% 
  # 池化層S2
  layer_max_pooling_2d(pool_size = c(2, 2)) %>% 
  layer_dropout(rate = 0.25) %>% 
  # 平坦層 12＊12﹡64 節點
  layer_flatten() %>% 
  # 隱藏層
  layer_dense(units = 128, activation = 'relu') %>% 
  layer_dropout(rate = 0.5) %>% 
  # 輸出層
  layer_dense(units = num_classes, activation = 'softmax')

# Compile model
model %>% compile(
  loss = loss_categorical_crossentropy,
  optimizer = optimizer_adadelta(),
  metrics = c('accuracy')
)

# Train model
model %>% fit(
  x_train, y_train,
  batch_size = batch_size,
  epochs = epochs,
  validation_split = 0.2
)

# Data Preparation ------------------------------------
batch_size <- 128   # 分批大小
num_classes <- 10   # 輸出層節點數0~9機率
epochs <- 12        # 訓練回合數

```{r eval=FALSE}
library(keras)

# Data Preparation -----------------------------------------------------

batch_size <- 128   # 分批大小
num_classes <- 10   # 輸出層節點數0~9機率
epochs <- 12        # 訓練回合數

# 圖片圖點數 28*28
img_rows <- 28
img_cols <- 28

# 讀取資料集, 分成訓練集,測試集
mnist <- dataset_mnist()
x_train <- mnist$train$x
y_train <- mnist$train$y
x_test <- mnist$test$x
y_test <- mnist$test$y

# 將原本資料集的三維陣列,轉為矩陣
x_train <- array_reshape(x_train, c(nrow(x_train), img_rows, img_cols, 1))
x_test <- array_reshape(x_test, c(nrow(x_test), img_rows, img_cols, 1))
input_shape <- c(img_rows, img_cols, 1)  ## 輸入圖片 28*28* 1圖層

# 將圖點灰階0~255 轉成 [0,1] range
x_train <- x_train / 255
x_test <- x_test / 255

cat('x_train_shape:', dim(x_train), '\n')
cat(nrow(x_train), 'train samples\n')
cat(nrow(x_test), 'test samples\n')

# 將答案轉成矩陣
y_train <- to_categorical(y_train, num_classes)
y_test <- to_categorical(y_test, num_classes)

# Define Model -----------------------------------------------------------

# Define model
model <- keras_model_sequential() %>%
  layer_conv_2d(filters = 32, kernel_size = c(3,3), activation = 'relu',
                input_shape = input_shape) %>% 
  layer_conv_2d(filters = 64, kernel_size = c(3,3), activation = 'relu') %>% 
  layer_max_pooling_2d(pool_size = c(2, 2)) %>% 
  layer_dropout(rate = 0.25) %>% 
  layer_flatten() %>% 
  layer_dense(units = 128, activation = 'relu') %>% 
  layer_dropout(rate = 0.5) %>% 
  layer_dense(units = num_classes, activation = 'softmax')

# Compile model
model %>% compile(
  loss = loss_categorical_crossentropy,
  optimizer = optimizer_adadelta(),
  metrics = c('accuracy')
)

# Train model
model %>% fit(
  x_train, y_train,
  batch_size = batch_size,
  epochs = epochs,
  validation_split = 0.2
)

scores <- model %>% evaluate(
  x_test, y_test, verbose = 0
)

# Output metrics
cat('Test loss:', scores[[1]], '\n')
cat('Test accuracy:', scores[[2]], '\n')
```

Train on 48000 samples, validate on 12000 samples
Epoch 1/12
48000/48000 [==============================] - 141s 3ms/step - loss: 0.3055 - acc: 0.9050 - val_loss: 0.0725 - val_acc: 0.9803
Epoch 2/12
48000/48000 [==============================] - 148s 3ms/step - loss: 0.0989 - acc: 0.9709 - val_loss: 0.0539 - val_acc: 0.9845
Epoch 3/12
48000/48000 [==============================] - 145s 3ms/step - loss: 0.0737 - acc: 0.9787 - val_loss: 0.0455 - val_acc: 0.9868
Epoch 4/12
48000/48000 [==============================] - 146s 3ms/step - loss: 0.0586 - acc: 0.9823 - val_loss: 0.0427 - val_acc: 0.9881
Epoch 5/12
48000/48000 [==============================] - 148s 3ms/step - loss: 0.0519 - acc: 0.9846 - val_loss: 0.0393 - val_acc: 0.9887
Epoch 6/12
48000/48000 [==============================] - 149s 3ms/step - loss: 0.0458 - acc: 0.9863 - val_loss: 0.0398 - val_acc: 0.9891
Epoch 7/12
48000/48000 [==============================] - 114s 2ms/step - loss: 0.0375 - acc: 0.9879 - val_loss: 0.0396 - val_acc: 0.9880
Epoch 8/12
48000/48000 [==============================] - 113s 2ms/step - loss: 0.0363 - acc: 0.9886 - val_loss: 0.0390 - val_acc: 0.9892
Epoch 9/12
48000/48000 [==============================] - 113s 2ms/step - loss: 0.0336 - acc: 0.9895 - val_loss: 0.0372 - val_acc: 0.9905
Epoch 10/12
48000/48000 [==============================] - 122s 3ms/step - loss: 0.0306 - acc: 0.9907 - val_loss: 0.0375 - val_acc: 0.9899
Epoch 11/12
48000/48000 [==============================] - 138s 3ms/step - loss: 0.0255 - acc: 0.9919 - val_loss: 0.0381 - val_acc: 0.9899
Epoch 12/12
48000/48000 [==============================] - 139s 3ms/step - loss: 0.0264 - acc: 0.9917 - val_loss: 0.0350 - val_acc: 0.9903

> # Output metrics
> cat('Test loss:', scores[[1]], '\n')
Test loss: 0.02830164 
> cat('Test accuracy:', scores[[2]], '\n')
Test accuracy: 0.9909