Test Feature Interaction library

########################################################################
## install packages if not exist
#install.packages("mlbench")
#install.packages("caret")
#install.packages("caret")
#install.packages("ggplot2")
#install.packages("forcats")     
#install.packages("RColorBrewer")
# We assume all related files are in "D:\rworks"
setwd("D:/rworks")
source("f_interact_lib.R")
library(mlbench)
library(caret)
########################################################################
## Regression
## Prepare data
data("BostonHousing")
ds <- BostonHousing[,-14]
cl <- BostonHousing[,14]    # price of a house    
## Fit a model
fitControl <- trainControl(method = "repeatedcv",
                           number = 5,
                           repeats = 2, returnResamp="all")
model.lm <- train(x=ds,y=cl, trControl=fitControl, method="lm") 
pred <- predict(model.lm, ds)
rmse <- RMSE(pred,cl)
rmse    # rmse of a whole model

[1] 4.679191

## Printf feature impact of each feature
for(i in 1:ncol(ds)) {
  cat(names(ds)[i], "\t",f.impact(model.lm, i), "\n")
}

crim     0.1825532
zn   0.2331253 
indus    0.003786494 
chas     0.09189613 
nox      0.8244577 
rm   1.374748 
age      -2.37516e-05 
dis      1.769364 
rad      1.292213 
tax      0.8304573 
ptratio      0.8302797 
b    0.1488995 
lstat    2.423039 

## Degree of feature interaction between two features
result <- f.interact(model.lm, "crim", "zn")
print.fi(result)

** feature interaction between [ crim ],[ zn ]
Task type                      :  Regression 
Performance of whole dataset   :  4.679191 
Reduced performance by F1      :  0.1825532 
Reduced performance by F2      :  0.256769 
Reduced performance by {F1,F2} :  0.4663873 
Interaction between {F1,F2}    :  0.02706504 

## Create a feature interaction matrix
f.matrix <- fi.matrix(model.lm)

[1] "This function may take long time ..."
1 / 13  end..
2 / 13  end..
3 / 13  end..
4 / 13  end..
5 / 13  end..
6 / 13  end..
7 / 13  end..
8 / 13  end..
9 / 13  end..
10 / 13  end..
11 / 13  end..
12 / 13  end..
13 / 13  end..

print(f.matrix)

                crim           zn         indus          chas          nox           rm           age
crim     0.182553229  0.027065039 -0.0106608895  0.0080191720  0.107880215  0.040389283 -0.0013082538
zn       0.027065039  0.233125343 -0.0165526752 -0.0090711658  0.169439028  0.110001871 -0.0025209938
indus   -0.010660889 -0.016552675  0.0037864936  0.0002806107 -0.038999021 -0.026816095  0.0003823493
chas     0.008019172 -0.009071166  0.0002806107  0.0918961293 -0.043323665  0.008267674  0.0003119213
nox      0.107880215  0.169439028 -0.0389990212 -0.0433236647  0.824457684  0.074450738 -0.0052279108
rm       0.040389283  0.110001871 -0.0268160945  0.0082676736  0.074450738  1.374748235 -0.0021911376
age     -0.001308254 -0.002520994  0.0003823493  0.0003119213 -0.005227911 -0.002191138 -0.0000237516
dis     -0.223775842 -0.427364623  0.0463930828  0.0178351928 -0.927368681 -0.568433136  0.0069948220
rad     -0.287422682 -0.212615358  0.0366221687 -0.0155821499 -0.673049673 -0.509543333  0.0042759030
tax      0.158414111  0.102333506 -0.0381552502 -0.0146905949  0.322154196  0.092302857 -0.0039008408
ptratio  0.065261967  0.128630311 -0.0213886716  0.0178508862  0.015984175  0.142753218 -0.0019165943
b        0.053293555  0.019912453 -0.0080007178  0.0050594436  0.083352024  0.017453131 -0.0008910354
lstat    0.147046326  0.162318568 -0.0471308146 -0.0198857623  0.309456961  0.384614795 -0.0061582880
                 dis          rad          tax      ptratio             b        lstat
crim    -0.223775842 -0.287422682  0.158414111  0.065261967  0.0532935550  0.147046326
zn      -0.427364623 -0.212615358  0.102333506  0.128630311  0.0199124527  0.162318568
indus    0.046393083  0.036622169 -0.038155250 -0.021388672 -0.0080007178 -0.047130815
chas     0.017835193 -0.015582150 -0.014690595  0.017850886  0.0050594436 -0.019885762
nox     -0.927368681 -0.673049673  0.322154196  0.015984175  0.0833520238  0.309456961
rm      -0.568433136 -0.509543333  0.092302857  0.142753218  0.0174531310  0.384614795
age      0.006994822  0.004275903 -0.003900841 -0.001916594 -0.0008910354 -0.006158288
dis      1.769364421  0.262119308 -0.733352922 -0.443354780 -0.1551927970 -1.238700599
rad      0.262119308  1.292213260 -0.941227759 -0.541265550 -0.1851499833 -1.010397838
tax     -0.733352922 -0.941227759  0.830457340  0.188359832  0.0998758032  0.273803614
ptratio -0.443354780 -0.541265550  0.188359832  0.830279659  0.0231617239  0.138010924
b       -0.155192797 -0.185149983  0.099875803  0.023161724  0.1488995070  0.116944685
lstat   -1.238700599 -1.010397838  0.273803614  0.138010924  0.1169446846  2.423039283

## Grid chart of two-way feature interaction
fi.grid.chart(model.lm, "BostonHousing")

[1] "This function may take long time ..."
1 / 13  end..
2 / 13  end..
3 / 13  end..
4 / 13  end..
5 / 13  end..
6 / 13  end..
7 / 13  end..
8 / 13  end..
9 / 13  end..
10 / 13  end..
11 / 13  end..
12 / 13  end..
13 / 13  end..

## Total degree of interaction of each feature
fi.total.plot(f.matrix) 

## Best pair of features (Postive interaction)
fi.best(f.matrix, type="pos")

## Best pair of features (Negative interaction)
fi.best(f.matrix, type="neg")