Quickstart: Run R code in R Client and Machine Learning Server

Artículo
02/16/2018

Important

This content is being retired and may not be updated in the future. The support for Machine Learning Server will end on July 1, 2022. For more information, see What's happening to Machine Learning Server?

Applies to: R Client 3.x, R Server 9.x, Machine Learning Server 9.x

Learn how to predict flight delays in R locally using R Client or Machine Learning Server. The example in this article uses historical on-time performance and weather data to predict whether the arrival of a scheduled passenger flight is delayed by more than 15 minutes. We approach this problem as a classification problem, predicting two classes -- whether the flight is delayed or on-time.

In machine learning and statistics, classification is the task of identifying the class or category to which an observation belongs based on a training dataset containing observations with known categories. Classification is generally a supervised learning problem. This quick start is a binary classification task with two classes.

In this example, you train a model using many examples from historic flight data, along with an outcome measure that indicates the appropriate category or class for each example. The two classes are '0' for on-time flights and '1' for flights delayed longer than 15 minutes.

Time estimate

If you have completed the prerequisites, this task takes approximately 10 minutes to complete.

Prerequisites

This quickstart assumes that you have:

An installed instance of Microsoft R Client or Machine Learning Server
R running on the command line or in an R integrated development environment (IDE). Read the article Get Started with Microsoft R Client for more information.
An internet connection to get sample data in the RTVS GitHub repository.

Example code

This article walks through some R code you can use to predict whether a flight will be delayed. Here is the entire R code for the example that we walk through in the sections.

       #############################################
       ##           ENTIRE EXAMPLE SCRIPT          ##
       #############################################

#Step 1: Prep and Import Data
#Initialize some variables to specify the data sets.
   github <- "https://raw.githubusercontent.com/Microsoft/RTVS-docs/master/examples/MRS_and_Machine_Learning/Datasets/"
   inputFileFlightURL <- paste0(github, "Flight_Delays_Sample.csv")
   inputFileWeatherURL <- paste0(github, "Weather_Sample.csv")

#Create a temporary directory to store the intermediate XDF files. 
   td <- tempdir()
   outFileFlight <- paste0(td, "/flight.xdf")
   outFileWeather <- paste0(td, "/weather.xdf")
   outFileOrigin <- paste0(td, "/originData.xdf")
   outFileDest <- paste0(td, "/destData.xdf")
   outFileFinal <- paste0(td, "/finalData.xdf")

#Import the flight data.
    flight_mrs <- rxImport(
      inData = inputFileFlightURL, outFile = outFileFlight,
      missingValueString = "M", stringsAsFactors = FALSE,
      # Remove columns that are possible target leakers from the flight data.
      varsToDrop = c("DepDelay", "DepDel15", "ArrDelay", "Cancelled", "Year"),
      # Define "Carrier" as categorical.
      colInfo = list(Carrier = list(type = "factor")),
      # Round down scheduled departure time to full hour.
      transforms = list(CRSDepTime = floor(CRSDepTime/100)),  
      overwrite = TRUE
    )

#Review the first six rows of flight data.
    head(flight_mrs)

#Summarize the flight data.
    rxSummary(~., data = flight_mrs, blocksPerRead = 2)

#Import the weather data.
    xform <- function(dataList) {
      # Create a function to normalize some numerical features.
      featureNames <- c(
        "Visibility", 
        "DryBulbCelsius", 
        "DewPointCelsius", 
        "RelativeHumidity", 
        "WindSpeed", 
        "Altimeter"
      )
      dataList[featureNames] <- lapply(dataList[featureNames], scale)
      return(dataList)
    }

    weather_mrs <- rxImport(
      inData = inputFileWeatherURL, outFile = outFileWeather,
      missingValueString = "M", stringsAsFactors = FALSE,
      # Eliminate some features due to redundance.
      varsToDrop = c("Year", "Timezone", 
                     "DryBulbFarenheit", "DewPointFarenheit"),
      # Create a new column "DestAirportID" in weather data.
      transforms = list(DestAirportID = AirportID),
      # Apply the normalization function.
      transformFunc = xform,  
      transformVars = c(
        "Visibility", 
        "DryBulbCelsius", 
        "DewPointCelsius", 
        "RelativeHumidity", 
        "WindSpeed", 
        "Altimeter"
      ),
      overwrite = TRUE
    )

#Review the variable information for the weather data.
    rxGetVarInfo(weather_mrs)


#Step 2: Pre-process Data
#Prepare for a merge by renaming some columns in the weather data.
    newVarInfo <- list(
      AdjustedMonth = list(newName = "Month"),
      AdjustedDay = list(newName = "DayofMonth"),
      AirportID = list(newName = "OriginAirportID"),
      AdjustedHour = list(newName = "CRSDepTime")
    )
    rxSetVarInfo(varInfo = newVarInfo, data = weather_mrs)

#Concatenate/Merge flight records and weather data.
##Join flight records and weather data at origin of the flight `OriginAirportID`.
      originData_mrs <- rxMerge(
        inData1 = flight_mrs, inData2 = weather_mrs, outFile = outFileOrigin,
        type = "inner", autoSort = TRUE, 
        matchVars = c("Month", "DayofMonth", "OriginAirportID", "CRSDepTime"),
        varsToDrop2 = "DestAirportID",
        overwrite = TRUE
      )

##Join flight records and weather data using the destination of the flight `DestAirportID`.
      destData_mrs <- rxMerge(
        inData1 = originData_mrs, inData2 = weather_mrs, outFile = outFileDest,
        type = "inner", autoSort = TRUE, 
        matchVars = c("Month", "DayofMonth", "DestAirportID", "CRSDepTime"),
        varsToDrop2 = c("OriginAirportID"),
        duplicateVarExt = c("Origin", "Destination"),
        overwrite = TRUE
      )

##Call the rxFactors() function to convert `OriginAirportID` and `DestAirportID` as categorical.
      rxFactors(inData = destData_mrs, outFile = outFileFinal, sortLevels = TRUE,
                factorInfo = c("OriginAirportID", "DestAirportID"),
                overwrite = TRUE)


#Step 3: Prepare Training and Test Datasets
#Randomly split data (80% for training, 20% for testing).
   rxSplit(inData = outFileFinal,
           outFilesBase = paste0(td, "/modelData"),
           outFileSuffixes = c("Train", "Test"),
           splitByFactor = "splitVar",
           overwrite = TRUE,
           transforms = list(
             splitVar = factor(sample(c("Train", "Test"),
                                       size = .rxNumRows,
                                       replace = TRUE,
                                       prob = c(.80, .20)),
                                levels = c("Train", "Test"))),
            rngSeed = 17,
            consoleOutput = TRUE)

#Point to the XDF files for each set.
   train <- RxXdfData(paste0(td, "/modelData.splitVar.Train.xdf"))
   test <- RxXdfData(paste0(td, "/modelData.splitVar.Test.xdf"))


#Step 4: Predict using Logistic Regression
#Choose and apply the Logistic Regression learning algorithm.

   #Build the formula.
   modelFormula <- formula(train, depVars = "ArrDel15",
                           varsToDrop = c("RowNum", "splitVar"))

   #Fit a Logistic Regression model.
   logitModel_mrs <- rxLogit(modelFormula, data = train)

   #Review the model results.
   summary(logitModel_mrs)

#Predict using new data.
    #Predict the probability on the test dataset.
    rxPredict(logitModel_mrs, data = test,
              type = "response",
              predVarNames = "ArrDel15_Pred_Logit",
              overwrite = TRUE)

    #Calculate Area Under the Curve (AUC).
    paste0("AUC of Logistic Regression Model:",
           rxAuc(rxRoc("ArrDel15", "ArrDel15_Pred_Logit", test)))

    #Plot the ROC curve.
    rxRocCurve("ArrDel15", "ArrDel15_Pred_Logit", data = test,
               title = "ROC curve - Logistic regression")

#Step 5: Predict using Decision Tree
#Choose and apply the Decision Tree learning algorithm.
    #Build a decision tree model.
    dTree1_mrs <- rxDTree(modelFormula, data = train, reportProgress = 1)

    #Find the Best Value of cp for Pruning rxDTree Object.
    treeCp_mrs <- rxDTreeBestCp(dTree1_mrs)

    #Prune a decision tree created by rxDTree and return the smaller tree.
    dTree2_mrs <- prune.rxDTree(dTree1_mrs, cp = treeCp_mrs)

#Predict using new data.
   #Predict the probability on the test dataset.
   rxPredict(dTree2_mrs, data = test, 
               overwrite = TRUE)

   #Calculate Area Under the Curve (AUC).
   paste0("AUC of Decision Tree Model:",
               rxAuc(rxRoc("ArrDel15", "ArrDel15_Pred", test)))

   #Plot the ROC curve.
   rxRocCurve("ArrDel15",
               predVarNames = c("ArrDel15_Pred", "ArrDel15_Pred_Logit"),
               data = test,
               title = "ROC curve - Logistic regression")

Step 1. Prepare and import data

Initialize some variables to specify the data sets.

github <- "https://raw.githubusercontent.com/Microsoft/RTVS-docs/master/examples/MRS_and_Machine_Learning/Datasets/"
inputFileFlightURL <- paste0(github, "Flight_Delays_Sample.csv")
inputFileWeatherURL <- paste0(github, "Weather_Sample.csv")

Create a temporary directory to store the intermediate XDF files. The External Data Frame (XDF) file format is a high-performance, binary file format for storing big data sets for use with RevoScaleR. This file format has an R interface and optimizes rows and columns for faster processing and analysis. Learn more
```
td <- tempdir()
outFileFlight <- paste0(td, "/flight.xdf")
outFileWeather <- paste0(td, "/weather.xdf")
outFileOrigin <- paste0(td, "/originData.xdf")
outFileDest <- paste0(td, "/destData.xdf")
outFileFinal <- paste0(td, "/finalData.xdf")
```

Import the flight data.

 flight_mrs <- rxImport(
   inData = inputFileFlightURL, outFile = outFileFlight,
   missingValueString = "M", stringsAsFactors = FALSE,
   # Remove columns that are possible target leakers from the flight data.
   varsToDrop = c("DepDelay", "DepDel15", "ArrDelay", "Cancelled", "Year"),
   # Define "Carrier" as categorical.
   colInfo = list(Carrier = list(type = "factor")),
   # Round down scheduled departure time to full hour.
   transforms = list(CRSDepTime = floor(CRSDepTime/100)),  
   overwrite = TRUE
 )

Review the first six rows of flight data.
```
head(flight_mrs)
```

Summarize the flight data.

rxSummary(~., data = flight_mrs, blocksPerRead = 2)

Import the weather data.

 xform <- function(dataList) {
   # Create a function to normalize some numerical features.
   featureNames <- c(
     "Visibility", 
     "DryBulbCelsius", 
     "DewPointCelsius", 
     "RelativeHumidity", 
     "WindSpeed", 
     "Altimeter"
   )
   dataList[featureNames] <- lapply(dataList[featureNames], scale)
   return(dataList)
 }

 weather_mrs <- rxImport(
   inData = inputFileWeatherURL, outFile = outFileWeather,
   missingValueString = "M", stringsAsFactors = FALSE,
   # Eliminate some features due to redundance.
   varsToDrop = c("Year", "Timezone", 
                  "DryBulbFarenheit", "DewPointFarenheit"),
   # Create a new column "DestAirportID" in weather data.
   transforms = list(DestAirportID = AirportID),
   # Apply the normalization function.
   transformFunc = xform,  
   transformVars = c(
     "Visibility", 
     "DryBulbCelsius", 
     "DewPointCelsius", 
     "RelativeHumidity", 
     "WindSpeed", 
     "Altimeter"
   ),
   overwrite = TRUE
 )

Review the variable information for the weather data.
```
 rxGetVarInfo(weather_mrs)
```

Step 2. Pre-process data

Rename some column names in the weather data to prepare it for merging.

 newVarInfo <- list(
   AdjustedMonth = list(newName = "Month"),
   AdjustedDay = list(newName = "DayofMonth"),
   AirportID = list(newName = "OriginAirportID"),
   AdjustedHour = list(newName = "CRSDepTime")
 )
 rxSetVarInfo(varInfo = newVarInfo, data = weather_mrs)

Concatenate/Merge flight records and weather data.

Join flight records and weather data at origin of the flight OriginAirportID.

originData_mrs <- rxMerge(
  inData1 = flight_mrs, inData2 = weather_mrs, outFile = outFileOrigin,
  type = "inner", autoSort = TRUE, 
  matchVars = c("Month", "DayofMonth", "OriginAirportID", "CRSDepTime"),
  varsToDrop2 = "DestAirportID",
  overwrite = TRUE
)

Join flight records and weather data using the destination of the flight DestAirportID.

destData_mrs <- rxMerge(
  inData1 = originData_mrs, inData2 = weather_mrs, outFile = outFileDest,
  type = "inner", autoSort = TRUE, 
  matchVars = c("Month", "DayofMonth", "DestAirportID", "CRSDepTime"),
  varsToDrop2 = c("OriginAirportID"),
  duplicateVarExt = c("Origin", "Destination"),
  overwrite = TRUE
)

Call the rxFactors() function to convert OriginAirportID and DestAirportID as categorical.

rxFactors(inData = destData_mrs, outFile = outFileFinal, sortLevels = TRUE,
          factorInfo = c("OriginAirportID", "DestAirportID"),
          overwrite = TRUE)

Step 3. Prepare training and test datasets

Randomly split data (80% for training, 20% for testing).

rxSplit(inData = outFileFinal,
        outFilesBase = paste0(td, "/modelData"),
        outFileSuffixes = c("Train", "Test"),
        splitByFactor = "splitVar",
        overwrite = TRUE,
        transforms = list(
          splitVar = factor(sample(c("Train", "Test"),
                                    size = .rxNumRows,
                                    replace = TRUE,
                                    prob = c(.80, .20)),
                             levels = c("Train", "Test"))),
         rngSeed = 17,
         consoleOutput = TRUE)

Point to the XDF files for each set.

train <- RxXdfData(paste0(td, "/modelData.splitVar.Train.xdf"))
test <- RxXdfData(paste0(td, "/modelData.splitVar.Test.xdf"))

Step 4. Predict using logistic regression

Choose and apply the Logistic Regression learning algorithm.

# Build the formula.
modelFormula <- formula(train, depVars = "ArrDel15",
                        varsToDrop = c("RowNum", "splitVar"))

# Fit a Logistic Regression model.
logitModel_mrs <- rxLogit(modelFormula, data = train)

# Review the model results.
summary(logitModel_mrs)

Predict using new data.

# Predict the probability on the test dataset.
 rxPredict(logitModel_mrs, data = test,
           type = "response",
           predVarNames = "ArrDel15_Pred_Logit",
           overwrite = TRUE)

# Calculate Area Under the Curve (AUC).
 paste0("AUC of Logistic Regression Model:",
        rxAuc(rxRoc("ArrDel15", "ArrDel15_Pred_Logit", test)))

# Plot the ROC curve.
 rxRocCurve("ArrDel15", "ArrDel15_Pred_Logit", data = test,
            title = "ROC curve - Logistic regression")

Step 5. Predict using decision tree

Choose and apply the Decision Tree learning algorithm.

# Build a decision tree model.
dTree1_mrs <- rxDTree(modelFormula, data = train, reportProgress = 1)

# Find the Best Value of cp for Pruning rxDTree Object.
treeCp_mrs <- rxDTreeBestCp(dTree1_mrs)

# Prune a decision tree created by rxDTree and return the smaller tree.
dTree2_mrs <- prune.rxDTree(dTree1_mrs, cp = treeCp_mrs)

Predict using new data.

#Predict the probability on the test dataset.
rxPredict(dTree2_mrs, data = test, 
            overwrite = TRUE)

#Calculate Area Under the Curve (AUC).
paste0("AUC of Decision Tree Model:",
            rxAuc(rxRoc("ArrDel15", "ArrDel15_Pred", test)))

#Plot the ROC curve.
rxRocCurve("ArrDel15",
            predVarNames = c("ArrDel15_Pred", "ArrDel15_Pred_Logit"),
            data = test,
            title = "ROC curve - Logistic regression")

Next steps

Now that you've tried this example, you can start developing your own solutions using the RevoScaleR R package functions, MicrosoftML R package functions, and APIs. When ready, you can run that R code using R Client or even send those R commands to a remote R Server for execution.

Learn More

You can learn more with these guides:

Compartir a través de