# R 中的 Keras API 我们在第 3 章中了解了 Keras API。在 R 中，此 API 使用`keras` R 包实现。 `keras` R 软件包实现了 Keras Python 接口的大部分功能，包括顺序 API 和函数式 API。 作为示例，我们提供了 MLP 模型的演练，用于在以下链接中对来自 MNIST 数据集的手写数字进行分类： [https://keras.rstudio.com/articles/examples/mnist_mlp.html](https://keras.rstudio.com/articles/examples/mnist_mlp.html) 。 您可以按照 Jupyter R 笔记本中的代码`ch-17c_Keras_in_R`。 1. 首先，加载库： ```r library(keras) ``` 1. 定义超参数： ```r batch_size <- 128 num_classes <- 10 epochs <- 30 ``` 1. 准备数据： ```r # The data, shuffled and split between train and test sets c(c(x_train, y_train), c(x_test, y_test)) %<-% dataset_mnist() x_train <- array_reshape(x_train, c(nrow(x_train), 784)) x_test <- array_reshape(x_test, c(nrow(x_test), 784)) # Transform RGB values into [0,1] range x_train <- x_train / 255 x_test <- x_test / 255 cat(nrow(x_train), 'train samples\n') cat(nrow(x_test), 'test samples\n') # Convert class vectors to binary class matrices y_train <- to_categorical(y_train, num_classes) y_test <- to_categorical(y_test, num_classes) ``` 注释是不言自明的：数据从 Keras 数据集库加载，然后转换为 2D arrray 并归一化为[0,1]范围。 1. 定义模型： ```r 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 %>% compile( loss = 'categorical_crossentropy', optimizer = optimizer_rmsprop(), metrics = c('accuracy') ) ``` 1. 定义和编译顺序模型。我们得到的模型定义如下： ```r _____________________________________________________ Layer (type) Output Shape Param # ===================================================== dense_26 (Dense) (None, 256) 200960 _____________________________________________________ dropout_14 (Dropout) (None, 256) 0 _____________________________________________________ dense_27 (Dense) (None, 128) 32896 _____________________________________________________ dropout_15 (Dropout) (None, 128) 0 _____________________________________________________ dense_28 (Dense) (None, 10) 1290 ===================================================== Total params: 235,146 Trainable params: 235,146 Non-trainable params: 0 ``` 1. 训练模型： ```r history <- model %>% fit( x_train, y_train, batch_size = batch_size, epochs = epochs, verbose = 1, validation_split = 0.2 ) plot(history) ``` 拟合函数的输出存储在历史对象中，其包含来自训练周期的损失和度量值。绘制历史对象中的数据，结果如下： Training and Validation Accuracy (y-axis) in Epochs (x-axis) 1. 评估模型： ```r score <- model %>% evaluate( x_test, y_test, verbose = 0 ) # Output metrics cat('Test loss:', score[[1]], '\n') cat('Test accuracy:', score[[2]], '\n') ``` 输出如下： ```r Test loss: 0.1128517 Test accuracy: 0.9816 ``` 太酷!! 在以下链接中查找更多关于 Keras in R 的示例：[https://keras.rstudio.com/articles/examples/index.html ](https://keras.rstudio.com/articles/examples/index.html) 有关 Keras R 软件包的更多文档可在以下链接中找到：[https://keras.rstudio.com/reference/index.html](https://keras.rstudio.com/reference/index.html)