R Deep Learning Cookbook

感謝分享

谢谢分享

1. ######################## USING MXNET package (neural Network)
   
2. library(mxnet)
   
3. 
   
4. # load the data
   
5. # Load the occupancy data
   
6. occupancy_train <- read.csv("/occupation_detection/datatraining.txt",stringsAsFactors = T)
   
7. occupancy_test <- read.csv("/occupation_detection/datatest.txt",stringsAsFactors = T)
   
8. 
   
9. # Define input (x) and output (y) variables"
   
10. x = c("Temperature", "Humidity", "Light", "CO2", "HumidityRatio")
    
11. y = "Occupancy"
    
12. 
    
13. # convert the train data into matrix
    
14. occupancy_train.x <- data.matrix(occupancy_train[,x])
    
15. occupancy_train.y <- occupancy_train$Occupancy
    
16. 
    
17. # convert the test data into matrix
    
18. occupancy_test.x <- data.matrix(occupancy_test[,x])
    
19. occupancy_test.y <- occupancy_test$Occupancy
    
20. 
    
21. # Train a multilayered perceptron
    
22. 
    
23. #set seed to reproduce results
    
24. mx.set.seed(1234567)
    
25. 
    
26. # create NN structure
    
27. smb.data <- mx.symbol.Variable("data")
    
28. smb.fc <- mx.symbol.FullyConnected(smb.data, num_hidden=5)
    
29. smb.soft <- mx.symbol.SoftmaxOutput(smb.fc)
    
30. 
    
31. # Train the network
    
32. nnmodel <- mx.model.FeedForward.create(symbol = smb.soft,
    
33.                                        X = occupancy_train.x,
    
34.                                        y = occupancy_train.y,
    
35.                                        ctx = mx.cpu(),
    
36.                                        num.round = 100,
    
37.                                        eval.metric = mx.metric.accuracy,
    
38.                                        array.batch.size = 100,
    
39.                                        learning.rate = 0.01)
    
40. 
    
41. # Train accuracy (AUC)
    
42. train_pred <- predict(nnmodel,occupancy_train.x)
    
43. train_yhat <- max.col(t(train_pred))-1
    
44. roc_obj <- pROC::roc(c(occupancy_train.y), c(train_yhat))
    
45. pROC::auc(roc_obj)
    
46. 
    
47. # Predict on the test data
    
48. test_pred <- predict(nnmodel,occupancy_test.x)
    
49. test_yhat <- max.col(t(test_pred))-1
    
50. 
    
51. #Test accuracy (AUC)
    
52. roc_obj <- pROC::roc(c(occupancy_test.y), c(test_yhat))
    
53. pROC::auc(roc_obj)

复制代码

1. ######################## USING TENSORFLOW (neural network)
   
2. library(tensorflow)
   
3. 
   
4. np <- import("numpy")
   
5. tf <- import("tensorflow")
   
6. 
   
7. # Loading input and test data
   
8. xFeatures = c("Temperature", "Humidity", "Light", "CO2", "HumidityRatio")
   
9. yFeatures = "Occupancy"
   
10. occupancy_train <-read.csv("/occupation_detection/datatraining.txt",stringsAsFactors = T)
    
11. occupancy_test <- read.csv("/occupation_detection/datatest.txt",stringsAsFactors = T)
    
12. 
    
13. # subset features for modeling and transform to numeric values
    
14. occupancy_train<-apply(occupancy_train[, c(xFeatures, yFeatures)], 2, FUN=as.numeric)
    
15. occupancy_test<-apply(occupancy_test[, c(xFeatures, yFeatures)], 2, FUN=as.numeric)
    
16. 
    
17. # Data dimensions
    
18. nFeatures<-length(xFeatures)
    
19. nRow<-nrow(occupancy_train)
    
20. 
    
21. # Reset the graph
    
22. tf$reset_default_graph()
    
23. # Starting session as interactive session
    
24. sess<-tf$InteractiveSession()
    
25. 
    
26. # Network Parameters
    
27. n_hidden_1 = 5L # 1st layer number of features
    
28. n_hidden_2 = 5L # 2nd layer number of features
    
29. n_input = 5L    # 5 attributes
    
30. n_classes = 1L  # Binary class
    
31. 
    
32. # Model Parameters
    
33. learning_rate = 0.001
    
34. training_epochs = 10000
    
35. 
    
36. # Graph input
    
37. x = tf$constant(unlist(occupancy_train[,xFeatures]), shape=c(nRow, n_input), dtype=np$float32)
    
38. y = tf$constant(unlist(occupancy_train[,yFeatures]), dtype="float32", shape=c(nRow, 1L))
    
39. 
    
40. # Create model
    
41. multilayer_perceptron <- function(x, weights, biases){
    
42.   # Hidden layer with RELU activation
    
43.   layer_1 = tf$add(tf$matmul(x, weights[["h1"]]), biases[["b1"]])
    
44.   layer_1 = tf$nn$relu(layer_1)
    
45.   # Hidden layer with RELU activation
    
46.   layer_2 = tf$add(tf$matmul(layer_1, weights[["h2"]]), biases[["b2"]])
    
47.   layer_2 = tf$nn$relu(layer_2)
    
48.   # Output layer with linear activation
    
49.   out_layer = tf$matmul(layer_2, weights[["out"]]) + biases[["out"]]
    
50.   return(out_layer)
    
51. }
    
52. 
    
53. # Initialises and store hidden layer's weight & bias
    
54. weights = list(
    
55.   "h1" = tf$Variable(tf$random_normal(c(n_input, n_hidden_1))),
    
56.   "h2" = tf$Variable(tf$random_normal(c(n_hidden_1, n_hidden_2))),
    
57.   "out" = tf$Variable(tf$random_normal(c(n_hidden_2, n_classes)))
    
58. )
    
59. biases = list(
    
60.   "b1" =  tf$Variable(tf$random_normal(c(1L,n_hidden_1))),
    
61.   "b2" = tf$Variable(tf$random_normal(c(1L,n_hidden_2))),
    
62.   "out" = tf$Variable(tf$random_normal(c(1L,n_classes)))
    
63. )
    
64. 
    
65. # Construct model
    
66. pred = multilayer_perceptron(x, weights, biases)
    
67. 
    
68. # Define loss and optimizer
    
69. cost = tf$reduce_mean(tf$nn$sigmoid_cross_entropy_with_logits(logits=pred, labels=y))
    
70. optimizer = tf$train$AdamOptimizer(learning_rate=learning_rate)$minimize(cost)
    
71. 
    
72. # Initializing the global variables
    
73. init = tf$global_variables_initializer()
    
74. sess$run(init)
    
75. 
    
76. # Training cycle
    
77. for(epoch in 1:training_epochs){
    
78.     sess$run(optimizer)
    
79.   if (epoch %% 20== 0)
    
80.     cat(epoch, "-", sess$run(cost), "\n")                                   
    
81. }
    
82. 
    
83. # Performance on Train
    
84. library(pROC)
    
85. ypred <- sess$run(tf$nn$sigmoid(multilayer_perceptron(x, weights, biases)))
    
86. roc_obj <- roc(occupancy_train[, yFeatures], as.numeric(ypred))
    
87. 
    
88. # Performance on Test
    
89. nRowt<-nrow(occupancy_test)
    
90. xt <- tf$constant(unlist(occupancy_test[, xFeatures]), shape=c(nRowt, nFeatures), dtype=np$float32) #
    
91. ypredt <- sess$run(tf$nn$sigmoid(multilayer_perceptron(xt, weights, biases)))
    
92. roc_objt <- roc(occupancy_test[, yFeatures], as.numeric(ypredt))
    
93. 
    
94. plot.roc(roc_obj, col = "green", lty=2, lwd=2)
    
95. plot.roc(roc_objt, add=T, col="red", lty=4, lwd=2)

复制代码

thx for sharing