Visualization of neural network with R
Neural network visualization package in R
The neuralnet function can visualize the calculation graph by the plot () function as standard. Note below how to visualize the computational graph when using other neural network packages that do not have features like the neuralnet function.
- plot.nn function
- plotnet function
Preparation
Sample data uses iris
Creating a learner
d = iris d $ Species <-as.factor (d $ Species) #train_test_split set.seed (0) sample <-sample.int (n = nrow (d), size = floor (0.80 * nrow (d)), replace = F) train <-d [sample,] test <-d [-sample,] summary (train) #nnet library (nnet) nn1 = nnet (Species ~., Size = 5, data = train) pred_nn1 <- predict (nn1, test, type = "class") table (test $ Species, pred_nn1)
Visualize nnet
In each case, the color indicates the positive or negative, and the thickness indicates the magnitude of the numerical value.
plot.nn function
source ("http://hosho.ees.hokudai.ac.jp/~kubo/log/2007/img07/plot.nn.txt") plot.nn (nn1)
plot.nnet function
install.packages ("NeuralNetTools") library (NeuralNetTools) plotnet (nn1)
By the way, in the neuralnet function
library (caret) tmp <-dummyVars (~., Data = train) dummy <-as.data.frame (predict (tmp, train)) library ("neuralnet") f = Species.setosa + Species.versicolor + Species. virginica ~ Sepal.Length + Sepal.Width + Petal.Length + Petal.Width nn2 <-neuralnet (formula = f, data = dummy) plot (nn2)
When there are many variables, it is easier to see the visualization with the horizontal plotnet function. The plotnet function is easy to use because it can visualize not only nnet but also neural networks created with RSNNS and caret and has a wide range of applications.
In-Depth Discussions
Comment list
There are not any comments yet