none
rxLogisticRegression error RRS feed

  • Question

  • I'm running the latest version of Microsoft Machine Learning Server 9.3 on a Linux Machine with CentOS 7 for testing purposes. I tried to run some of the samples from GitHub (https://github.com/Microsoft/microsoft-r/blob/master/microsoft-ml/Microsoft ML Tutorial/Microsoft ML Tutorial)


    When running the MNIST.R code in Revo64 console, the system throws an error message when executing  the rxLogisticRegressionn function

    ...
    model <- formula(paste("Label ~", paste(xVars, collapse = " + ")))
    rxLogisticRegressionFit <- rxLogisticRegression(model, dataTrain, type = "multiClass")
    ...

    The error message says:
    exception while shutting down RxClientPipe: Write error: expected 8 got 4294967295


    The same code is running well on my local machine (Windows 10) using the Microsoft R Client v3.4.3.

    Any idea how I can fix this issue?

    The same error occurs when executing the sample code from the function's documentation page  (https://docs.microsoft.com/en-us/machine-learning-server/r-reference/microsoftml/rxlogisticregression)

    using the iris dataset:

    # Multi-class logistic regression

    testObs <- rnorm(nrow(iris)) > 0

    testIris <- iris[testObs,]

    trainIris <- iris[!testObs,]

    multiLogit <- rxLogisticRegression( formula = Species~Sepal.Length + Sepal.Width + Petal.Length + Petal.Width, type = "multiClass", data = trainIris)

    # Score the model score

    MultiDF <- rxPredict(multiLogit, data = testIris, extraVarsToWrite = "Species") 

    # Print the first rows of the data frame with scores 

    head(scoreMultiDF) 

     # Look at confusion matrix

    table(scoreMultiDF$Species, scoreMultiDF$PredictedLabel) 

     # Look at the observations with incorrect predictions

    badPrediction = scoreMultiDF$Species != scoreMultiDF$PredictedLabel

    scoreMultiDF[badPrediction,]

    Error message:  exception while shutting down RxClientPipe: Write error: expected 8 got 4294967295

    I'll be happy to receive your suggestions.



    • Edited by 011gwerc01 Tuesday, May 29, 2018 5:32 PM Missing Links
    Tuesday, May 29, 2018 5:27 PM