7.2. Friedman (1991) introduced several benchmark data sets create by
simulation. One of these simulations used the following nonlinear
equation to create data:
\[y = 10 sin(\pi x_1x_2) + 20(x_3 - 0.5)^2 +
10x_4 + 5x_5 + N(0, \sigma^2)\]
where the x values are random variables uniformly distributed between
[0, 1] (there are also 5 other non-informative variables also created in
the simulation). The package mlbench contains a function called
mlbench.friedman1 that simulates these data:
library(caret)
## Warning: package 'caret' was built under R version 4.4.3
## Loading required package: ggplot2
## Warning: package 'ggplot2' was built under R version 4.4.3
## Loading required package: lattice
# From textbook
library(mlbench)
## Warning: package 'mlbench' was built under R version 4.4.3
set.seed(200)
trainingData <- mlbench.friedman1(200, sd = 1)
##We convert the 'x' data from a matrix to a data frame
##One reason is that this will give the columns names.
trainingData$x <- data.frame(trainingData$x)
##Look at the data using
featurePlot(trainingData$x, trainingData$y)
##or other methods.
##This creates a list with a vector 'y' and a matrix
##of predictors 'x'. Also simulate a large test set to
##estimate the true error rate with good precision:
testData <- mlbench.friedman1(5000, sd = 1)
testData$x <- data.frame(testData$x)
# From textbook
knnModel <- train(x = trainingData$x,
y = trainingData$y,
method = "knn",
preProc = c("center", "scale"),
tuneLength = 10)
knnModel
## k-Nearest Neighbors
##
## 200 samples
## 10 predictor
##
## Pre-processing: centered (10), scaled (10)
## Resampling: Bootstrapped (25 reps)
## Summary of sample sizes: 200, 200, 200, 200, 200, 200, ...
## Resampling results across tuning parameters:
##
## k RMSE Rsquared MAE
## 5 3.466085 0.5121775 2.816838
## 7 3.349428 0.5452823 2.727410
## 9 3.264276 0.5785990 2.660026
## 11 3.214216 0.6024244 2.603767
## 13 3.196510 0.6176570 2.591935
## 15 3.184173 0.6305506 2.577482
## 17 3.183130 0.6425367 2.567787
## 19 3.198752 0.6483184 2.592683
## 21 3.188993 0.6611428 2.588787
## 23 3.200458 0.6638353 2.604529
##
## RMSE was used to select the optimal model using the smallest value.
## The final value used for the model was k = 17.
# From textbook
knnPred <- predict(knnModel, newdata = testData$x)
## The function 'postResample' can be used to get the test set
## perforamnce values
postResample(pred = knnPred, obs = testData$y)
## RMSE Rsquared MAE
## 3.2040595 0.6819919 2.5683461
Which models appear to give the best performance? Does MARS select the informative predictors (those named X1–X5)?
svmModel <- train(x=trainingData$x,
y=trainingData$y,
method = "svmRadial",
preProc = c("center", "scale"),
tuneLength = 10)
svmModel
## Support Vector Machines with Radial Basis Function Kernel
##
## 200 samples
## 10 predictor
##
## Pre-processing: centered (10), scaled (10)
## Resampling: Bootstrapped (25 reps)
## Summary of sample sizes: 200, 200, 200, 200, 200, 200, ...
## Resampling results across tuning parameters:
##
## C RMSE Rsquared MAE
## 0.25 2.545335 0.7804647 2.015121
## 0.50 2.319786 0.7965148 1.830009
## 1.00 2.188349 0.8119636 1.726027
## 2.00 2.103655 0.8241314 1.655842
## 4.00 2.066879 0.8294322 1.631051
## 8.00 2.052681 0.8313929 1.623550
## 16.00 2.049867 0.8318312 1.621820
## 32.00 2.049867 0.8318312 1.621820
## 64.00 2.049867 0.8318312 1.621820
## 128.00 2.049867 0.8318312 1.621820
##
## Tuning parameter 'sigma' was held constant at a value of 0.06802164
## RMSE was used to select the optimal model using the smallest value.
## The final values used for the model were sigma = 0.06802164 and C = 16.
mars <- train(x=trainingData$x,
y=trainingData$y,
method = "earth",
preProc = c("center", "scale"),
tuneLength = 10)
## Loading required package: earth
## Warning: package 'earth' was built under R version 4.4.3
## Loading required package: Formula
## Loading required package: plotmo
## Warning: package 'plotmo' was built under R version 4.4.3
## Loading required package: plotrix
## Warning: package 'plotrix' was built under R version 4.4.3
mars
## Multivariate Adaptive Regression Spline
##
## 200 samples
## 10 predictor
##
## Pre-processing: centered (10), scaled (10)
## Resampling: Bootstrapped (25 reps)
## Summary of sample sizes: 200, 200, 200, 200, 200, 200, ...
## Resampling results across tuning parameters:
##
## nprune RMSE Rsquared MAE
## 2 4.447386 0.2254125 3.620675
## 3 3.790305 0.4344625 3.058704
## 4 2.801182 0.6884819 2.233531
## 6 2.493135 0.7492201 1.986528
## 7 2.089713 0.8239588 1.645996
## 9 1.816053 0.8673608 1.420333
## 10 1.819611 0.8674028 1.417343
## 12 1.832487 0.8651613 1.426371
## 13 1.845943 0.8632112 1.436005
## 15 1.854557 0.8617322 1.452920
##
## Tuning parameter 'degree' was held constant at a value of 1
## RMSE was used to select the optimal model using the smallest value.
## The final values used for the model were nprune = 9 and degree = 1.
knnperf <- predict(knnModel, newdata=testData$x)
postResample(pred = knnperf, obs =testData$y)
## RMSE Rsquared MAE
## 3.2040595 0.6819919 2.5683461
svmperf <- predict(svmModel, newdata=testData$x)
postResample(pred = svmperf, obs=testData$y)
## RMSE Rsquared MAE
## 2.0864652 0.8236735 1.5854649
marsperf <- predict(mars, newdata = testData$x)
postResample(pred =marsperf, obs=testData$y)
## RMSE Rsquared MAE
## 1.7901760 0.8705315 1.3712537
The MARS model had the lowest RMSE and highest Rsquared indicating that this model had the best performance.
varImp(mars)
## earth variable importance
##
## Overall
## X1 100.00
## X4 82.92
## X2 64.47
## X5 40.67
## X3 28.65
## X6 0.00
The top 5 informative variables by importance are X1-X5.
Exercise 6.3 describes data for a chemical manufacturing process. Use the same data imputation, data splitting, and pre-processing steps as before and train several nonlinear regression models.
# From textbook
library(AppliedPredictiveModeling)
## Warning: package 'AppliedPredictiveModeling' was built under R version 4.4.3
data(ChemicalManufacturingProcess)
# Impute missing values via knn
library(impute)
chemicals <- impute.knn(as.matrix(ChemicalManufacturingProcess), rng.seed = 123)
chemicals <- as.data.frame(chemicals$data)
# Test/train split
training_chemicals <- createDataPartition(chemicals$Yield, p = 0.8, list = FALSE)
train_predictors_chemicals <- chemicals[training_chemicals,-1]
train_yield <- chemicals[training_chemicals, 1]
test_predictors_chemicals <- chemicals[-training_chemicals,-1]
test_yield <- chemicals[-training_chemicals, 1]
suppressWarnings(chemicals_knn <- train(x=train_predictors_chemicals,
y=train_yield,
method = "knn",
preProc = c("center", "scale"),
tuneLength=10))
knn_predict <- predict(chemicals_knn, test_predictors_chemicals)
yield_knn <-data.frame(obs=test_yield, pred=knn_predict)
print(defaultSummary(yield_knn))
## RMSE Rsquared MAE
## 1.4239614 0.4011076 1.1771250
suppressWarnings(chemicals_mars <-train(train_predictors_chemicals,
train_yield,
method='earth',
preProc = c('center', 'scale')))
mars_predict <- predict(chemicals_mars, test_predictors_chemicals)
yield_mars <- data.frame(obs=test_yield, pred=mars_predict)
yield_mars$pred <- yield_mars$y
yield_mars <-yield_mars[,-2]
print(defaultSummary(yield_mars))
## RMSE Rsquared MAE
## 1.402311 0.424434 1.110419
suppressWarnings(chemicals_svm <- train(train_predictors_chemicals,
train_yield,
method = "svmRadial",
preProc = c("center", "scale")))
svm_predict <-predict(chemicals_svm, test_predictors_chemicals)
yield_svm <- data.frame(obs=test_yield, pred=svm_predict)
print(defaultSummary(yield_svm))
## RMSE Rsquared MAE
## 1.2654699 0.5270595 1.0447860
Based on the models above (knn, svm, and mars), SVM had the lowest RSME and highest Rsquared value indicating that this model is the most optimal model for the test set.
varImp(chemicals_svm, scale =FALSE)
## loess r-squared variable importance
##
## only 20 most important variables shown (out of 57)
##
## Overall
## ManufacturingProcess32 0.3704
## BiologicalMaterial06 0.3484
## ManufacturingProcess36 0.3045
## BiologicalMaterial03 0.3010
## ManufacturingProcess13 0.2987
## BiologicalMaterial02 0.2817
## ManufacturingProcess17 0.2812
## ManufacturingProcess09 0.2706
## ManufacturingProcess31 0.2687
## BiologicalMaterial12 0.2574
## ManufacturingProcess06 0.2456
## BiologicalMaterial11 0.2212
## ManufacturingProcess33 0.2203
## BiologicalMaterial04 0.1998
## ManufacturingProcess29 0.1991
## BiologicalMaterial01 0.1690
## BiologicalMaterial08 0.1664
## ManufacturingProcess11 0.1659
## BiologicalMaterial09 0.1514
## ManufacturingProcess30 0.1458
The top 10 predictors in the SVM model are manufacturing processes 32,36,13,17,09,31 and biological material 06,03,02,12. In my previous assignment of predictors in the linear model, the top predictors were manufacturing processes 32,17,13,09,36,06,33 and biological material 06,02,03.
There are many overlapping predictors in the linear and SVM models, including manufacturing processes 32,36,13,17,09 and biological material 02,03,06.
In both models, the manufacturing processes seem to dominate the lists. However there are much fewer biological material variables compared to manufacturing processes so there are actually a higher presence of biological materials in the top predictors lists relative to the number of variables.
ggplot(chemicals, aes(x=ManufacturingProcess32, y=Yield)) + geom_point()
ggplot(chemicals, aes(x=BiologicalMaterial06, y=Yield)) + geom_point()
ggplot(chemicals, aes(x=ManufacturingProcess36, y = Yield)) + geom_point()
The top 3 predictors have been plotted. For manufacturing process 32 and
biological material 06, points further right tend to lead to higher
yield but some points that are far right result in lower yield. In
manufacturing process 36, the yield appears to be higher towards the
left but the furthest left points result in a more blunted effect.