library(knitr) # used for slides rendering library(dplyr) library(readr) library(stringr) library(lubridate) library(xts) library(sp) library(CORElearn)
2024-03-18
library(knitr) # used for slides rendering library(dplyr) library(readr) library(stringr) library(lubridate) library(xts) library(sp) library(CORElearn)
set.seed(1024)
Feature engineering is the process of modifying raw data in order to make it more suitable for efficient extraction of wisdom through machine learning approaches.
In R programming, feature engineering can be done through a variety of functions.
dplyr::select()
dplyr::filter()
dplyr::group_by()
City | Temperature | Humidity |
---|---|---|
Ankara | 20 | 55 |
Ankara | 35 | 28 |
Antalya | 30 | 82 |
Antalya | 22 | 88 |
Ankara | 10 | 61 |
Antalya | 41 | 93 |
Ankara | -12 | 42 |
Antalya | 13 | 73 |
Antalya | 33 | 78 |
Ankara | 0 | 52 |
The data as is doesn’t tell much.
spatial_weather <- data.frame( City = c("Ankara", "Ankara", "Antalya", "Antalya", "Ankara", "Antalya", "Ankara", "Antalya", "Antalya", "Ankara"), Temperature = c(20, 35, 30, 22, 10, 41, -12, 13, 33, 0), Humidity = c(55, 28, 82, 88, 61, 93, 42, 73, 78, 52) ) attach(spatial_weather) par(mfrow=c(1,2), mar=c(7,5,0,0)) boxplot(Temperature) boxplot(Humidity)
detach(spatial_weather)
What if we add some context?
attach(spatial_weather) par(mfrow=c(1,2), mar=c(7,5,0,0)) boxplot(Temperature~City) boxplot(Humidity~City)
detach(spatial_weather)
What about relations between variables?
attach(spatial_weather) par(mar=c(7,5,0,0)) plot(Humidity~Temperature) abline(lm(Humidity~Temperature, spatial_weather))
detach(spatial_weather)
With context:
attach(spatial_weather) par(mar=c(7,5,0,0)) plot(Humidity~Temperature, col=factor(City), pch=19) abline(lm(Humidity~Temperature, spatial_weather[spatial_weather$City=="Ankara",]), col="black") abline(lm(Humidity~Temperature, spatial_weather[spatial_weather$City=="Antalya",]), col="red")
detach(spatial_weather)
xts
packagexts
package is used for working with time series datasp500 <- xts(c(1102.94, 1104.49, 1115.71, 1118.31), ymd(c("2010-02-25", "2010-02-26", "2010-03-01", "2010-03-02"))) sp500
## [,1] ## 2010-02-25 1102.94 ## 2010-02-26 1104.49 ## 2010-03-01 1115.71 ## 2010-03-02 1118.31
sp500["2010-03-02"]
## [,1] ## 2010-03-02 1118.31
sp500["2010-03"]
## [,1] ## 2010-03-01 1115.71 ## 2010-03-02 1118.31
sp500["2010-03-01/"]
## [,1] ## 2010-03-01 1115.71 ## 2010-03-02 1118.31
sp500["2010-02-26/2010-03-01"]
## [,1] ## 2010-02-26 1104.49 ## 2010-03-01 1115.71
xts
packagexts
also allows nice plots of time series dataplot(sp500)
data(AirPassengers) AirPassengers
## Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec ## 1949 112 118 132 129 121 135 148 148 136 119 104 118 ## 1950 115 126 141 135 125 149 170 170 158 133 114 140 ## 1951 145 150 178 163 172 178 199 199 184 162 146 166 ## 1952 171 180 193 181 183 218 230 242 209 191 172 194 ## 1953 196 196 236 235 229 243 264 272 237 211 180 201 ## 1954 204 188 235 227 234 264 302 293 259 229 203 229 ## 1955 242 233 267 269 270 315 364 347 312 274 237 278 ## 1956 284 277 317 313 318 374 413 405 355 306 271 306 ## 1957 315 301 356 348 355 422 465 467 404 347 305 336 ## 1958 340 318 362 348 363 435 491 505 404 359 310 337 ## 1959 360 342 406 396 420 472 548 559 463 407 362 405 ## 1960 417 391 419 461 472 535 622 606 508 461 390 432
ap <- as.xts(AirPassengers) head(ap)
## [,1] ## Oca 1949 112 ## Şub 1949 118 ## Mar 1949 132 ## Nis 1949 129 ## May 1949 121 ## Haz 1949 135
head(diff(ap))
## [,1] ## Oca 1949 NA ## Şub 1949 6 ## Mar 1949 14 ## Nis 1949 -3 ## May 1949 -8 ## Haz 1949 14
tail(diff(ap))
## [,1] ## Tem 1960 87 ## Ağu 1960 -16 ## Eyl 1960 -98 ## Eki 1960 -47 ## Kas 1960 -71 ## Ara 1960 42
apRel <- diff(ap) / ap head(apRel)
## [,1] ## Oca 1949 NA ## Şub 1949 0.05084746 ## Mar 1949 0.10606061 ## Nis 1949 -0.02325581 ## May 1949 -0.06611570 ## Haz 1949 0.10370370
tail(apRel)
## [,1] ## Tem 1960 0.13987138 ## Ağu 1960 -0.02640264 ## Eyl 1960 -0.19291339 ## Eki 1960 -0.10195228 ## Kas 1960 -0.18205128 ## Ara 1960 0.09722222
plot(ap)
plot(apRel)
embed
Functionhead(apRel, 10)
## [,1] ## Oca 1949 NA ## Şub 1949 0.05084746 ## Mar 1949 0.10606061 ## Nis 1949 -0.02325581 ## May 1949 -0.06611570 ## Haz 1949 0.10370370 ## Tem 1949 0.08783784 ## Ağu 1949 0.00000000 ## Eyl 1949 -0.08823529 ## Eki 1949 -0.14285714
head(embed(apRel[-1], 5))
## [,1] [,2] [,3] [,4] [,5] ## [1,] 0.10370370 -0.06611570 -0.02325581 0.10606061 0.05084746 ## [2,] 0.08783784 0.10370370 -0.06611570 -0.02325581 0.10606061 ## [3,] 0.00000000 0.08783784 0.10370370 -0.06611570 -0.02325581 ## [4,] -0.08823529 0.00000000 0.08783784 0.10370370 -0.06611570 ## [5,] -0.14285714 -0.08823529 0.00000000 0.08783784 0.10370370 ## [6,] -0.14423077 -0.14285714 -0.08823529 0.00000000 0.08783784
“Everything is related with everything else, but near things are more related than distant things.”
First law of geography, (Tobler, 1970)
sp
is useful in spatial analysis# https://web.cs.dal.ca/~ltorgo/AuxFiles/forestFires.txt ff <- read_csv("forestFires.txt") print(ff, width=70)
## # A tibble: 25,000 × 14 ## FID_ CID ano1991 ano1992 ano1993 ano1994 ano1995 ano1996 ano1997 ## <lgl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> ## 1 NA 1 0 0 0 0 0 0 0 ## 2 NA 2 0 0 0 0 0 0 0 ## 3 NA 3 0 0 0 0 0 0 0 ## 4 NA 4 0 0 0 0 0 0 0 ## 5 NA 5 0 0 0 0 0 0 0 ## 6 NA 6 0 0 0 0 0 0 0 ## 7 NA 7 0 0 0 0 0 0 0 ## 8 NA 8 0 0 0 0 0 0 0 ## 9 NA 9 0 0 0 0 0 0 0 ## 10 NA 10 0 0 0 0 0 0 0 ## # ℹ 24,990 more rows ## # ℹ 5 more variables: ano1998 <dbl>, ano1999 <dbl>, ano2000 <dbl>, ## # x <dbl>, y <dbl>
anoX
: Fire happened in year XspatialCoord <- select(ff, long = x, lat = y) spatialData <- select(ff, Year2000 = ano2000) coordRefSys <- CRS("+proj=longlat +ellps=WGS84") fires <- SpatialPointsDataFrame(spatialCoord, spatialData, proj4string = coordRefSys) head(fires)
## coordinates Year2000 ## 1 (-7.31924, 38.5406) 0 ## 2 (-7.63557, 40.5022) 0 ## 3 (-7.90273, 40.3418) 0 ## 4 (-7.25657, 39.2572) 0 ## 5 (-8.50379, 37.3445) 0 ## 6 (-8.05975, 41.562) 0
bbox(fires)
## min max ## long -9.49174 -6.20743 ## lat 36.98050 42.14360
head(coordinates(fires))
## long lat ## [1,] -7.31924 38.5406 ## [2,] -7.63557 40.5022 ## [3,] -7.90273 40.3418 ## [4,] -7.25657 39.2572 ## [5,] -8.50379 37.3445 ## [6,] -8.05975 41.5620
tm
provides useful functionsColumns
Rows
sample()
data(iris) sampleRate <- 0.7 sampledRows <- sample(1:nrow(iris), nrow(iris) * sampleRate) iris.sample <- iris[sampledRows,]
sample.int()
data(iris) sampleRate <- 0.7 sampledRows <- sample.int(nrow(iris), nrow(iris) * sampleRate) iris.sample <- iris[sampledRows,]
A pseudocode for very large dataset sampling
Potential problems:
ncol() > nrow()
Correlation - simple
Information theoretic metrics
\(H(Y)=-\sum_{c_i\in\mathcal{Y}}{P(Y=c_i)\times\log{P(Y=c_i)}}\)
\(H(Y|X) = -\sum_{v_i\in \mathcal {X}, c_i\in \mathcal {Y}} P(X=v_i,Y=c_i)\log {\frac {P(X=v_i,Y=c_i)}{p(X=v_i)}}\)
\(IG(X) = H(Y) - H(Y|X)\)
\(GR(X) = \frac{IG(X)} {H(X)}\)
FSelector
and CORElearn
attrEval
is from CORElearn
data(iris) attrEval(Species ~ ., iris, estimator = "GainRatio")
## Sepal.Length Sepal.Width Petal.Length Petal.Width ## 0.5919339 0.3512938 1.0000000 1.0000000
attrEval(Species ~ ., iris, estimator = "InfGain")
## Sepal.Length Sepal.Width Petal.Length Petal.Width ## 0.5572327 0.2831260 0.9182958 0.9182958
attrEval(Species ~ ., iris, estimator = "Gini")
## Sepal.Length Sepal.Width Petal.Length Petal.Width ## 0.2277603 0.1269234 0.3333333 0.3333333
attrEval(Species ~ ., iris, estimator = "Relief")
## Sepal.Length Sepal.Width Petal.Length Petal.Width ## 0.1974074 0.1874074 0.7267797 0.7088889
infoCore(what = "attrEval")
## [1] "ReliefFequalK" "ReliefFexpRank" "ReliefFbestK" ## [4] "Relief" "InfGain" "GainRatio" ## [7] "MDL" "Gini" "MyopicReliefF" ## [10] "Accuracy" "ReliefFmerit" "ReliefFdistance" ## [13] "ReliefFsqrDistance" "DKM" "ReliefFexpC" ## [16] "ReliefFavgC" "ReliefFpe" "ReliefFpa" ## [19] "ReliefFsmp" "GainRatioCost" "DKMcost" ## [22] "ReliefKukar" "MDLsmp" "ImpurityEuclid" ## [25] "ImpurityHellinger" "UniformDKM" "UniformGini" ## [28] "UniformInf" "UniformAccuracy" "EqualDKM" ## [31] "EqualGini" "EqualInf" "EqualHellinger" ## [34] "DistHellinger" "DistAUC" "DistAngle" ## [37] "DistEuclid"
infoCore(what = "attrEvalReg")
## [1] "RReliefFequalK" "RReliefFexpRank" "RReliefFbestK" ## [4] "RReliefFwithMSE" "MSEofMean" "MSEofModel" ## [7] "MAEofModel" "RReliefFdistance" "RReliefFsqrDistance"
data(iris) pca <- princomp(iris[,1:4]) loadings(pca)
## ## Loadings: ## Comp.1 Comp.2 Comp.3 Comp.4 ## Sepal.Length 0.361 0.657 0.582 0.315 ## Sepal.Width 0.730 -0.598 -0.320 ## Petal.Length 0.857 -0.173 -0.480 ## Petal.Width 0.358 -0.546 0.754 ## ## Comp.1 Comp.2 Comp.3 Comp.4 ## SS loadings 1.00 1.00 1.00 1.00 ## Proportion Var 0.25 0.25 0.25 0.25 ## Cumulative Var 0.25 0.50 0.75 1.00
new.iris <- data.frame(pca$scores[, 1:2], Species = iris$Species) head(new.iris) %>% kable()
Comp.1 | Comp.2 | Species |
---|---|---|
-2.684126 | 0.3193972 | setosa |
-2.714142 | -0.1770012 | setosa |
-2.888991 | -0.1449494 | setosa |
-2.745343 | -0.3182990 | setosa |
-2.728716 | 0.3267545 | setosa |
-2.280860 | 0.7413304 | setosa |