intro_to_ml/19_b_nystroem_approximation...

204 lines
8.2 KiB
R
Raw Permalink Normal View History

source("04_validation_croisee_code.R")
source("05_d_svd_mca_code.R")
source("15_loocv_code.R")
source("19_nystroem_approximation_code.R")
dataset.housing <-
function()
{
set.seed(1123)
# loading dataset in memory
dat <- read.csv(file="data/housing.csv", header=TRUE)
# transform ocean_proximity into a factor
dat$ocean_proximity <- as.factor(dat$ocean_proximity)
levels(dat$ocean_proximity)[levels(dat$ocean_proximity)=="<1H OCEAN"] <- "O:<1H"
levels(dat$ocean_proximity)[levels(dat$ocean_proximity)=="ISLAND"] <- "O:ISL"
levels(dat$ocean_proximity)[levels(dat$ocean_proximity)=="INLAND"] <- "O:INL"
levels(dat$ocean_proximity)[levels(dat$ocean_proximity)=="NEAR BAY"] <- "O:NB"
levels(dat$ocean_proximity)[levels(dat$ocean_proximity)=="NEAR OCEAN"] <- "O:NO"
# replace total_bedrooms missing values with the median value
dat$total_bedrooms[is.na(dat$total_bedrooms)] <- median(dat$total_bedrooms, na.rm=TRUE)
# remove individuals corresponding to the Island modality of ocean_proximity
dat <- dat[dat$ocean_proximity != "O:ISL", ]
# remove empty modalities (here, only "O:ISL")
dat <- droplevels(dat)
# introduce new variable for number of rooms by households
dat['rooms'] <- dat['total_rooms'] / dat['households']
# introduce new variable for number of bedrooms by households
dat['bedrooms'] <- dat['total_bedrooms'] / dat['households']
# introduce new variable for number of people by households
dat['pop'] <- dat['population'] / dat['households']
# remove individuals with extremely high values of the target
dat <- dat[dat$median_house_value < 500001, ]
# select variables to retain in the dataset
dat <- dat[c('longitude', 'latitude', 'housing_median_age', 'households',
'median_income', 'median_house_value', 'ocean_proximity',
'rooms', 'bedrooms', 'pop')]
# one-hot-enconde categorical variable ocean_proximity
Z <- onehot_enc(dat[c('ocean_proximity')])
dat <- cbind(dat, as.data.frame(Z))
dat <- dat[,!(colnames(dat) %in% c('ocean_proximity'))]
dat.all <- dat
# separate observed variables X from the target Y
2023-01-23 16:04:52 -05:00
X <- dat[,!(colnames(dat) %in% c('median_house_value'))]
Y <- dat[,c('median_house_value')]
names(Y) <- rownames(X)
dat <- list(X = X, Y = Y)
# split the dataset into train and test
split <- splitdata(dat, 0.8)
train <- split$entr
test <- split$test
r <- list( dat=dat.all, train=train, test=test )
return(r)
}
# create a categorical dataset for correspondence analysis of the training data
dataset.housing.mca <-
function(dat=NULL)
{
# loading dataset in memory
if(is.null(dat)) { dat <- read.csv(file="data/housing.csv", header=TRUE) }
# transform ocean_proximity into a factor
dat$ocean_proximity <- as.factor(dat$ocean_proximity)
levels(dat$ocean_proximity)[levels(dat$ocean_proximity)=="<1H OCEAN"] <- "O:<1H"
levels(dat$ocean_proximity)[levels(dat$ocean_proximity)=="ISLAND"] <- "O:ISL"
levels(dat$ocean_proximity)[levels(dat$ocean_proximity)=="INLAND"] <- "O:INL"
levels(dat$ocean_proximity)[levels(dat$ocean_proximity)=="NEAR BAY"] <- "O:NB"
levels(dat$ocean_proximity)[levels(dat$ocean_proximity)=="NEAR OCEAN"] <- "O:NO"
# discretization of longitude
cuts <- kcuts(x = dat$longitude, centers = 4)
dat$c_longitude <- cut(x = dat$longitude, unique(cuts), include.lowest = TRUE)
levels(dat$c_longitude) <- c('LO:W', 'LO:MW', 'LO:ME', 'LO:E')
# discretization of latitude
cuts <- kcuts(x = dat$latitude, centers = 4)
dat$c_latitude <- cut(x = dat$latitude, unique(cuts), include.lowest = TRUE)
levels(dat$c_latitude) <- c('LA:S','LA:MS','LA:MN','LA:N')
# discretization of housing_median_age
cuts <- c(min(dat$housing_median_age), 15, 25, 35, 51, 52)
dat$c_age <- cut(x = dat$housing_median_age, unique(cuts),
include.lowest = TRUE)
levels(dat$c_age) <- c('AG:15]','AG:25]','AG:35]','AG:51]', 'AG:52')
# creation and discretization of rooms
dat$rooms <- dat$total_rooms / dat$households
cuts <- c(min(dat$rooms), 4, 6, 8, max(dat$rooms))
dat$c_rooms <- cut(x = dat$rooms, unique(cuts), include.lowest = TRUE)
levels(dat$c_rooms) <- c('RO:4]','RO:6]','RO:8]', 'RO:>8')
# creation and discretization of bedrooms
dat$bedrooms <- dat$total_bedrooms / dat$households
cuts <- c(min(dat$bedrooms, na.rm = TRUE), 1.1, max(dat$bedrooms, na.rm = TRUE))
dat$c_bedrooms <- cut(x = dat$bedrooms, unique(cuts), include.lowest = TRUE)
levels(dat$c_bedrooms) <- c('BE:1]','BE:>1')
# creation and discretization of pop
dat$pop <- dat$population / dat$households
cuts <- c(min(dat$pop), 2, 3, 4, max(dat$pop))
dat$c_pop <- cut(x = dat$pop, unique(cuts), include.lowest = TRUE)
levels(dat$c_pop) <- c('PO:2]','PO:3]', 'PO:4]', 'PO:>4')
# discretization of households
cuts <- c(min(dat$households), 300, 400, 600, max(dat$households))
dat$c_households <- cut(x = dat$households, cuts, include.lowest = TRUE)
levels(dat$c_households) <- c('HH:3]', 'HH:4]', 'HH:6]', 'HH:>6')
# discretization of median_income
cuts <- quantile(dat$median_income, probs = seq(0,1,1/4))
cuts <- c(cuts[1:length(cuts)-1], 15, max(dat$median_income))
dat$c_income <- cut(x = dat$median_income, cuts, include.lowest = TRUE)
levels(dat$c_income) <- c('IC:L', 'IC:ML', 'IC:MH', 'IC:H', 'IC:>15')
# discretization of median_house_value
cuts <- c(min(dat$median_house_value), 115000, 175000, 250000,
max(dat$median_house_value))
dat$c_house_value <- cut(x = dat$median_house_value, cuts,
include.lowest = TRUE)
levels(dat$c_house_value) <- c('HV:A', 'HV:B', 'HV:C', 'HV:D')
# discretized version of the entire dataset
dat <- dat[c('ocean_proximity', 'c_longitude', 'c_latitude', 'c_age',
'c_rooms', 'c_bedrooms', 'c_pop', 'c_households', 'c_income',
'c_house_value')]
vent <- list()
# ventilation of modality RO:>8 of variable c_rooms
vent$c_rooms <- ventilate(dat$c_rooms, "RO:>8")
dat$c_rooms[vent$c_rooms$sup_i] <- vent$c_rooms$smpl
# ventilation of modality IC:>15 of variable c_income
vent$c_income <- ventilate(dat$c_income, "IC:>15")
dat$c_income[vent$c_income$sup_i] <- vent$c_income$smpl
# ventilation of missing values of variable c_bedrooms
vent$c_bedrooms <- ventilate(dat$c_bedrooms, "NA")
dat$c_bedrooms[vent$c_bedrooms$sup_i] <- vent$c_bedrooms$smpl
# removal of empty modalities following the various ventilations
dat <- droplevels(dat)
# c_house_value, the target, is considered as a supplementary variable
supvar <- which(names(dat) == "c_house_value")
r <- list( vent=vent, dat=dat, supvar=supvar )
return(r)
}
print("load dataset")
dat <- dataset.housing()
print("Nystroem approx ridge regression")
nakrm <- nakr(dat$train$X, dat$train$Y)
nakrm.yh <- predict(nakrm, dat$test$X)
nakrm.mae <- mean(abs(nakrm.yh - dat$test$Y))
print(paste("Test MAE for Nystroem: ", nakrm.mae))
# correspondence analysis of the training set
rawdat <- read.csv(file="data/housing.csv", header=TRUE)
train.rawdat <- rawdat[as.numeric(rownames(dat$train$X)),]
dat.mca <- dataset.housing.mca(train.rawdat)
cam <- mca(dat.mca)
# average of the prediction errors by clusters
nakrm.yh.train <- predict(nakrm, dat$train$X)
nakrm.mae.train <- mean(abs(nakrm.yh.train - dat$train$Y))
nakrm.err.train <- abs(nakrm.yh.train - dat$train$Y)
clst.err <- aggregate(nakrm.err.train, list(cam$clsti$cluster), FUN=mean)
clst.err$Group.1 <- paste0("clst-", as.character(clst.err$Group.1))
clst.tbl <- clstcor.mca(cam)
clst.err <- clst.err[match(rownames(clst.tbl),clst.err$Group.1),]
clst.tbl <- cbind(clst.tbl, clst.err[,2])
colnames(clst.tbl)[ncol(clst.tbl)] <- "ERR"
clst.tbl <- clst.tbl[order(-clst.tbl[,"ERR"]),]
print("training set clusters obtained by correspondence analysis and ordered by amount of error")
print(clst.tbl)
# From these informations, we could try to understand why for some clusters the model commits greater errors...
print("Linear ridge regression")
rm <- ridge(dat$train$X, dat$train$Y)
rm.yh <- predict(rm, dat$test$X)
rm.mae <- mean(abs(rm.yh - dat$test$Y))
print(paste("Test MAE for Ridge: ", rm.mae))
print("Random forest")
library(randomForest)
rfm <- randomForest(dat$train$X, dat$train$Y)
rfm.yh <- predict(rfm, dat$test$X)
rfm.mae <- mean(abs(rfm.yh - dat$test$Y))
print(paste("Test MAE for Random Forest: ", rfm.mae))