行動技術與應用

conda create --name r-tensorflow python=3.7

activate r-tensorflow
pip install --ignore-installed --upgrade tensorflow 
conda install --channel defaults conda python=3.6 --yes
conda update --channel defaults --all --yes
conda install -c conda-forge keras

activate r-tensorflow
python

import tensorflow as tf
hello = tf.constant('Hi, TensorFlow!')
session = tf.Session()
print(session.run(hello))

```{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}
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)
```

model <- keras_model_sequential()

  layer_dense(units = 256, activation = 'relu', input_shape= c(784)) %>%
  layer_dropout(rate = 0.4) %>%

1. 安裝 rstudio/keras, rstudio/tensorflow。
```{r eval=FALSE}
if(!"devtools" %in% installed.packages()) 
  install.packages("devtools");

if(!"keras" %in% installed.packages()) {
  require(devtools);
  install_github("rstudio/keras");
}
```
2. 下載mnist資料集。
```{r eval=FALSE}
# 準備資料集MNIST ：手寫灰階圖片,每張圖28*28=784圖點
library(keras)
mnist <- dataset_mnist()   # 資料集內建於keras內
```
3. mnist資料集已分為訓練集與測試集，先檢視訓練集
```{r eval=FALSE}
# 資料集已分為訓練集/測試集
# $x: a 3-d array (images,width,height) of grayscale values
# $y: an integer vector with values ranging from 0 to 9
x_train_image <-mnist$train$x # 3D陣列,圖點集60000筆
x_train_image <- array_reshape(x_train_image, c(nrow(x_train_image), 784)) #轉成矩陣60000*784 

y_train_label <-mnist$train$y # 整數vector，60000個答案

## 矩陣圖點旋轉函式,m--> matrix; 2-->apply to columns, rev function
rotate <- function(m) t(apply(m, 2, rev))
## sample函式
print_sample <- function(image, label, num) {
  if(num>0) {
    for( i in 1:num) {
      sample <- matrix(image[i,], nrow=28, ncol=28, byrow=TRUE)
      print(image(rotate(sample),sub=sprintf("答案: %s",label[i])))
    }
  }
}
## 列印訓練集前10筆
print_sample(x_train_image, y_train_label, 10)
```

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

if(!"keras" %in% installed.packages()) {
  require(devtools);
  install_github("rstudio/keras");
}
```
下載mnist資料集。
```{r eval=FALSE}
# 準備資料集MNIST ：手寫灰階圖片,每張圖28*28=784圖點
library(keras)
mnist <- dataset_mnist()   # 資料集內建於keras內
```
mnist資料集已分為訓練集與測試集，
```{r eval=FALSE}
# 資料集已分為訓練集/測試集
# $x: a 3-d array (images,width,height) of grayscale values
# $y: an integer vector with values ranging from 0 to 9
x_train_image <-mnist$train$x # 3D陣列,圖點集60000筆
x_train_image <- array_reshape(x_train_image, c(nrow(x_train_image), 784)) #轉成矩陣60000*784 
x_test_image <-mnist$test$x
x_test_image <- array_reshape(x_test_image, c(nrow(x_test_image), 784))

y_train_label <-mnist$train$y # 整數vector，60000個答案
y_test_label <-mnist$test$y

```{r eval=FALSE}
# rescale: 將0～255數值轉成 0~1之間的整數
print(x_train_image[1,])
x_train_image <- x_train_image / 255
x_test_image <- x_test_image / 255
print(x_train_image[1,])

# 轉成matrix
print(y_train_label[1:10])
y_train_label <- to_categorical(y_train_label, 10)
y_test_label <- to_categorical(y_test_label, 10)
head(y_train_label,10)

```

model <- keras_model_sequential()

model <- keras_model_sequential() #多元感知器
model %>%
  ## 輸入資料784個圖點
  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) %>%
  ## 輸入資料10個數字機率
  layer_dense(units = 10, activation = 'softmax')
summary(model)

layer_dense(units = 256, activation = 'relu', input_shape= c(784))

layer_dense(units = 128, activation = 'relu')

layer_dense(units = 10, activation = 'softmax')

layer_dense(units = 128, activation = 'relu')

layer_dense(units = 10, activation = 'softmax')

## compile the model
model %>% compile(
  loss = 'categorical_crossentropy',
  optimizer = optimizer_rmsprop(),
  metrics = c('accuracy')
)

```{r eval=FALSE}
## 訓練階段
history <- model %>% fit(
  x_train, y_train, 
  epochs = 30, batch_size = 128, 
  validation_split = 0.2
)
plot(history)

```{r eval=FALSE}
# 準備資料集MNIST ：手寫灰階圖片,每張圖28*28=784圖點
library(keras)
mnist <- dataset_mnist()   # 資料集內建於keras內

# 資料集已分為訓練集/測試集
# $x: a 3-d array (images,width,height) of grayscale values
# $y: an integer vector with values ranging from 0 to 9
x_train <-mnist$train$x
y_train <-mnist$train$y
x_test <- mnist$test$x
y_test <- mnist$test$y

rm(mnist)

#reshape： 3D array 轉成 2d matrix (圖1, 784個圖點)
x_train <- array_reshape(x_train, c(nrow(x_train), 784))
x_test <- array_reshape(x_test, c(nrow(x_test), 784))

# rescale: 將0～255數值轉成 0~1之間的整數
x_train <- x_train / 255
x_test <- x_test / 255

# 轉成matrix
y_train <- to_categorical(y_train, 10)
y_test <- to_categorical(y_test, 10)

```

```{r eval=FALSE}
##定義訓練模型
model <- keras_model_sequential()
model %>%
  ## 輸入資料784個圖點
  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) %>%
  ## 輸入資料10個數字機率
  layer_dense(units = 10, activation = 'softmax')
summary(model)

## comple the model
model %>% compile(
  loss = 'categorical_crossentropy',
  optimizer = optimizer_rmsprop(),
  metrics = c('accuracy')
)
  
```
訓練與評估
```{r eval=FALSE}
## 訓練階段
history <- model %>% fit(
  x_train, y_train, 
  epochs = 30, batch_size = 128, 
  validation_split = 0.2
)
plot(history)

##測試階段
model %>% evaluate(x_test, y_test)

## 產生預測結果
model %>% predict_classes(x_test)
```

model <- keras_model_sequential() #多元感知器
model %>%
  ## 輸入資料784個圖點
  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) %>%
  ## 輸入資料10個數字機率
  layer_dense(units = 10, activation = 'softmax')
summary(model)

## comple the model
model %>% compile(
  loss = 'categorical_crossentropy',
  optimizer = optimizer_rmsprop(),
  metrics = c('accuracy')
)

## 訓練階段
history <- model %>% fit(
  x_train, y_train, 
  epochs = 30, batch_size = 128, 
  validation_split = 0.2
)
plot(history)

if(!"tfdeploy" %in% installed.packages()) {
  devtools::install_github("rstudio/tfdeploy")
}

library(tfdeploy)
export_savedmodel(model, "savemodel", remove_learning_phase = FALSE)
tfdeploy::serve_savedmodel("savemodel", browse = TRUE)

export_savedmodel(model, "資料夾名稱", remove_learning_phase = FALSE)

tfdeploy::serve_savedmodel("savemodel", browse = TRUE)

{
  "instances": [
    {
      "image_input": [0.12,0,0.79,...,0,0]
    }
  ]
}

curl -X POST -H "Content-Type: application/json" -d @new_image.json http://localhost:8089/serving_default/predict

{
  "predictions": [
    {
      "prediction": [ 1.3306e-24, 4.9968e-26, 1.8917e-23, 1.7047e-21, 0,
        8.963e-33, 0, 1, 2.3306e-32,2.0314e-22
      ]
    }
  ]
}