[MSR] MSDN - Analyzing a Large Data Set w/ ScaleR

/*******************************************************************************************************************
-- Title : [MSR] MSDN - Analyzing a Large Data Set w/ ScaleR
-- Reference : microsoft.com
-- Key word : microsoft r scaler revoscaler rxxdfdata rxgetvarinfo rxgetinfo rxreadxdf summary nrow 
                  linear model regression 단순 회귀 분석 회귀분석 회귀모델 회귀 모델 lm system.time
                  rxlinmod rxreadxdf dataframe data frame data.frame rxlineplot sort coef dccoef sprintf
                  dclhcoef 선형 모델 선형모델 intercept 절편 기울기 lm linear model rxresultsdf
*******************************************************************************************************************/

-- MSDN
    * https://msdn.microsoft.com/en-us/microsoft-r/scaler-getting-started-3-analyze-large-data

-- Chart

--  Airline DS 다운로드

    * Download URL :  http://packages.revolutionanalytics.com/datasets/

    * Airline on-time performance : http://stat-computing.org/dataexpo/2009/

-- Microsoft R

# ***********************************************
# -- Processing the airline dataset
# ***********************************************
 
# ------------------------------
# -- Download the airline dataset
# ------------------------------
 
# -- 디렉토리 설정 
bigDataDir <- "C:\\RProject\\MRS\\Data"
 
# -- Airline DS 다운로드
#    Download URL :  http://packages.revolutionanalytics.com/datasets/
#    Airline on-time performance : http://stat-computing.org/dataexpo/2009/
 
# -- 파일 설정
airDataName <- file.path(bigDataDir, "AirOnTime7Pct.xdf")
 
# -- DS 가져오기 
bigAirDS <- RxXdfData( airDataName )
head(bigAirDS)
 
# -- 추가 변수 정보 보기
rxGetVarInfo(bigAirDS)
rxGetInfo(bigAirDS, getVarInfo=TRUE)
 
# ------------------------------
# -- Create Dataframe from DS
# ------------------------------
 
# -- create dataframe
myData <- rxReadXdf(bigAirDS)
head(myData)
 
# -- Export to CSV
write.csv(myData,"e:\\AirOnTime7Pct.csv", row.names=TRUE) 
 
# ------------------------------
# -- Reading a chunk of data
# ------------------------------
 
# -- 데이터프레임으로 가져오기 
testDF <- rxReadXdf(file = bigAirDS,
                    varsToKeep = c("ArrDelay","DepDelay", "DayOfWeek"),
                    startRow = 100000, numRows = 1000)
summary(testDF)
nrow(testDF)
 
# -- 단순 회귀분석 샘플([MSR]에서)
lmObj <- lm(ArrDelay~DayOfWeek, data = testDF)
summary(lmObj)                                                            # Intercept : 절편
 
# -- 회귀모델 생성([R]에서)
result.lm = lm(formula=ArrDelay~DayOfWeek, data=testDF)
result.lm  
 
 
# ***********************************************
# -- Estimating a Linear Model with a Huge Data Set
# ***********************************************
#    대량건을 처리할 경우 "cannot allocate vector" error 발생 가능
#    이 경우 RevoScaleR로 처리 
 
# ------------------------------
# -- 처리 시스템 시간 추정 
# ------------------------------
# -- blocksPerRead = 30
system.time(
    delayArr <- rxLinMod(ArrDelay ~ DayOfWeek, data = bigAirDS,
                         cube = TRUE, blocksPerRead = 30)
)
 
summary(delayArr)
 
# -- blocksPerRead = 300
system.time(
    delayArr <- rxLinMod(ArrDelay ~ DayOfWeek, data = bigAirDS,
                         cube = TRUE, blocksPerRead = 300)
)
 
summary(delayArr)
 
# -- cube = TRUE
#    Condition Number가 변경됨(뭔지 모르겠다).
system.time(
    delayArr <- rxLinMod(ArrDelay ~ DayOfWeek, data = bigAirDS,
                         cube = FALSE, blocksPerRead = 300)
)
 
summary(delayArr)
 
# ------------------------------
# -- 선형 모델 결합 
# ------------------------------
 
# --선형모델 2개 수행
delayArr <- rxLinMod(ArrDelay ~ DayOfWeek, data = bigAirDS,
                     cube = TRUE, blocksPerRead = 300)
 
delayDep <- rxLinMod(DepDelay ~ DayOfWeek, data = bigAirDS,
                     cube = TRUE, blocksPerRead = 30)
 
# -- 데이터프레임으로의 변환(rxResultsDF) 및 결합 
#    c.f) rxReadXdf
cubeResults <- rxResultsDF(delayArr)
head(cubeResults)
class(cubeResults)                                                        # data.frame
 
cubeResults$DepDelay <- rxResultsDF(delayDep)$DepDelay
head(cubeResults)
 
# -- Lineplot 출력 
rxLinePlot( ArrDelay + DepDelay ~ DayOfWeek, data = cubeResults,
            title = 'Average Arrival and Departure Delay by Day of Week')
 
# ------------------------------
# -- 진행 사항 출력 유무 
# ------------------------------
 
# -- 진행사항 생략(reportProgress = 0)
delayDep <- rxLinMod(DepDelay ~ DayOfWeek, data = bigAirDS,
                     cube = TRUE, blocksPerRead = 30, reportProgress = 0)
 
# -- 진행사항 출력 
delayDep <- rxLinMod(DepDelay ~ DayOfWeek, data = bigAirDS,
                     cube = TRUE, blocksPerRead = 30)
 
 
# ***********************************************
# -- Handling Larger Linear Models
# ***********************************************
 
# -- 선형 모델 생성
delayCarrier <- rxLinMod(ArrDelay ~ UniqueCarrier,
                         data = bigAirDS, cube = TRUE, blocksPerRead = 30)
 
summary(delayCarrier)
 
# -- sort the coefficient vector(계수 백터 정렬)
dcCoef <- sort(coef(delayCarrier))
dcCoef
 
# -- 가장 낮은 지연 시간 
head(dcCoef, 10)
 
# -- 가장 높은 지연 시간 
tail(dcCoef, 10)
 
# -- 지연 차이 출력 
sprintf("United's additional delay compared with Hawaiian: %f",
        dcCoef["UniqueCarrier=UA"]-dcCoef["UniqueCarrier=HA"])
 
 
# ***********************************************
# -- 다중 회귀 모델 처리 
# ***********************************************
 
# -- 여러 변수 처리 
delayCarrierLoc <- rxLinMod(ArrDelay ~ UniqueCarrier + Origin+Dest,
                            data = bigAirDS, cube = TRUE, blocksPerRead = 30)
 
dclCoef <- coef(delayCarrierLoc)
 
sprintf(
    "United's additional delay accounting for dep and arr location: %f",
    dclCoef["UniqueCarrier=UA"]- dclCoef["UniqueCarrier=HA"])
 
paste("Number of coefficients estimated: ", length(!is.na(dclCoef)))
 
# -- 비행 시간 변수 추가 
delayCarrierLocHour <- rxLinMod(ArrDelay ~
                                    UniqueCarrier + Origin + Dest + F(CRSDepTime),
                                data = bigAirDS, cube = TRUE, blocksPerRead = 30)
 
dclhCoef <- coef(delayCarrierLocHour)
dclhCoef
 
# -- 결과 요약 
sprintf("United's additional delay compared with Hawaiian: %f",
        dcCoef["UniqueCarrier=UA"]-dcCoef["UniqueCarrier=HA"])
 
paste("Number of coefficients estimated: ", length(!is.na(dcCoef)))
 
sprintf(
    "United's additional delay accounting for dep and arr location: %f",
    dclCoef["UniqueCarrier=UA"]- dclCoef["UniqueCarrier=HA"])
 
paste("Number of coefficients estimated: ", length(!is.na(dclCoef)))
 
sprintf(
    "United's additional delay accounting for location and time: %f",    
    dclhCoef["UniqueCarrier=UA"]-dclhCoef["UniqueCarrier=HA"])
 
paste("Number of coefficients estimated: ", length(!is.na(dclhCoef)))
 
 
# ***********************************************
# -- 항공사 지연 예측 
# ***********************************************
 
# -- 예상 지연을 추정하는 함수 생성 
expectedDelay <- function( carrier = "AA", origin = "SEA",
                           dest = "SFO", deptime = "9")
{
    coeffNames <- c(
        sprintf("UniqueCarrier=%s", carrier),
        sprintf("Origin=%s", origin),
        sprintf("Dest=%s", dest),
        sprintf("F_CRSDepTime=%s", deptime))
    return (sum(dclhCoef[coeffNames]))
}
 
# ------------------------------
# -- 여행 지연 비교
# ------------------------------
 
# Go to JFK (New York) from Seattle at 5 in the afternoon on United
expectedDelay("AA", "SEA", "JFK", "17")
# Go to Newark from Seattle at 5 in the afternoon on United
expectedDelay("UA", "SEA", "EWR", "17")
# Or go to Honolulu from Seattle at 7 am on Hawaiian
expectedDelay("HA", "SEA", "HNL", "7")