How to Fit Regression Data with CNN Model in R

   CNN (Convolutional Neural Networks) models are mainly useful when we apply them for training a multi-dimensional type of data such as an image. But they are not limited to this purpose only, we can also implement the CNN model for regression data analysis. We saw the CNN model regression with Python in the previous post and in this tutorial, we'll implement the same method in R.
   We use a 1-dimensional convolutional function to apply the CNN model. We need Keras R interface to use the Keras neural network API in R. You need to install it if it is not available on your development environment. The tutorial covers:

1. Preparing the data
2. Defining and fitting the model
3. Predicting and visualizing the results
4. Source code listing

We'll start by loading the required libraries for this tutorial.

library(keras)
library(caret)

Preparing the data

   We use the Boston housing dataset as a target regression data in this tutorial. First, we'll load the dataset and split it into the train and test parts.

set.seed(123)
boston = MASS::Boston
indexes = createDataPartition(boston$medv, p = .85, list = F)
 

train = boston[indexes,]
test = boston[-indexes,]

Next, we'll separate x input and y output parts of both train and test data and convert them into the matrix type. As you may know, the 'medv' is the output, y data in the Boston housing dataset and it is the final column (14) in it. The remaining columns are the x input data.

xtrain = as.matrix(train[,-14])
ytrain = as.matrix(train[,14])
xtest = as.matrix(test[,-14])
ytest = as.matrix(test[, 14])

We'll check the dimensions.

dim(xtrain)
[1] 432  13
 

dim(ytrain)
[1] 432   1

Next, we'll reshape the x input data by adding another one-dimension.

xtrain = array(xtrain, dim = c(nrow(xtrain), 13, 1))
xtest = array(xtest, dim = c(nrow(xtest), 13, 1))
 

dim(xtrain)
[1] 432  13   1
 

dim(xtest)
[1] 74 13  1

Here, we can extract the input dimension for the keras model.

in_dim = c(dim(xtrain)[2:3])
print(in_dim)
[1] 13  1

Defining and fitting the model

   We'll define the Keras sequential model and add a one-dimensional convolutional layer. Input shape becomes as it is defined above (13,1). We'll add Flatten and Dense layers and compile it with 'Adam' optimizer.

model = keras_model_sequential() %>% layer_conv_1d(filters = 64, kernel_size = 2, input_shape = in_dim, activation = "relu") %>% layer_flatten() %>% layer_dense(units = 32, activation = "relu") %>% layer_dense(units = 1, activation = "linear") model %>% compile( loss = "mse", optimizer = "adam") model %>% summary() ________________________________________________________________________ Layer (type) Output Shape Param # ======================================================================== conv1d_2 (Conv1D) (None, 12, 64) 192 ________________________________________________________________________ flatten_2 (Flatten) (None, 768) 0 ________________________________________________________________________ dense_3 (Dense) (None, 32) 24608 ________________________________________________________________________ dense_4 (Dense) (None, 1) 33 ======================================================================== Total params: 24,833 Trainable params: 24,833 Non-trainable params: 0 ________________________________________________________________________

Next, we 'll fit the model with train data.

model %>% fit(xtrain, ytrain, epochs = 100, batch_size=16, verbose = 0)
scores = model %>% evaluate(xtrain, ytrain, verbose = 0)
print(scores)
    loss 
24.20518

Predicting and visualizing the results

Now we can predict the test data with the trained model.

ypred = model %>% predict(xtest)

We'll check the accuracy of prediction through the RMSE metrics.

cat("RMSE:", RMSE(ytest, ypred))
RMSE: 4.935908

Finally, we'll visualize the result in a plot to check the difference.

x_axes = seq(1:length(ypred))
plot(x_axes, ytest, ylim = c(min(ypred), max(ytest)),
     col = "burlywood", type = "l", lwd = 2, ylab = "medv")
lines(x_axes, ypred, col = "red", type = "l", lwd = 2)
legend("topleft", legend = c("y-test", "y-pred"),
       col = c("burlywood", "red"), lty=1, cex=0.7, lwd=2, bty='n')

 

   In this tutorial, we've briefly learned how to fit and predict regression data with the keras CNN model in R. The full source code is listed below.


Source code listing

library(keras)
library(caret)

set.seed(123)
 

boston = MASS::Boston
indexes = createDataPartition(boston$medv, p = .85, list = F)
train = boston[indexes,]
test = boston[-indexes,]
 
xtrain = as.matrix(train[,-14])
ytrain = as.matrix(train[,14])
xtest = as.matrix(test[,-14])
ytest = as.matrix(test[, 14])
 
dim(xtrain)
dim(ytrain)
 

xtrain = array(xtrain, dim = c(nrow(xtrain), 13, 1))
xtest = array(xtest, dim = c(nrow(xtest), 13, 1))
 

dim(xtrain)
dim(xtest)
 
in_dim = c(dim(xtrain)[2:3])
print(in_dim)

model = keras_model_sequential() %>%
  layer_conv_1d(filters = 64, kernel_size = 2,
                input_shape = in_dim, activation = "relu") %>%
  layer_flatten() %>%
  layer_dense(units = 32, activation = "relu") %>%
  layer_dense(units = 1, activation = "linear")

model %>% compile(
  loss = "mse",
  optimizer = "adam")

model %>% summary()

model %>% fit(xtrain, ytrain, epochs = 100, batch_size=16, verbose = 0)
scores = model %>% evaluate(xtrain, ytrain, verbose = 0)
print(scores)
 
ypred = model %>% predict(xtest)
 
cat("RMSE:", RMSE(ytest, ypred))

x_axes = seq(1:length(ypred))
plot(x_axes, ytest, ylim = c(min(ypred), max(ytest)),
     col = "burlywood", type = "l", lwd = 2, ylab = "medv")
lines(x_axes, ypred, col = "red", type = "l", lwd = 2)
legend("topleft", legend = c("y-test", "y-pred"),
       col = c("burlywood", "red"), lty = 1, cex=0.7, lwd=2, bty='n')