I am interested in classify binary data using a Random Forest method. My data is numeric tensor and label of each class represented by 1.0 or 0.0. The size of the training data is ~16 million examples, and the size of the test data is around 500 million examples.

rfST = Classify[traningData, 
  Method -> {"RandomForest", "TreeNumber" -> 500}, 
  TrainingProgressReporting -> "Print", TargetDevice -> {"GPU", 1}]

predictionRes = 
    Values[rfST[testData, "Probabilities", TargetDevice -> {"GPU", 1}]]][[

To speed up the training and test, I am using TargetDevice -> "GPU." When I am run the prediction I "feel" that a GPU doesn't take part in computation, however, the CPU is a computation engine. Please see the attached figures. enter image description here enter image description here

Any suggestion on what to check to speed up the calculation Note: I am using Win 10, and WL 11.3 and the GPU is GTX 1080 (I update the driver)

  • 1
    $\begingroup$ try to install CUDA drivers manually $\endgroup$ – Fortsaint Jan 11 '19 at 16:40
  • $\begingroup$ I did it, but I don't see improvement. $\endgroup$ – Kiril Danilchenko Jan 11 '19 at 16:49
  • 1
    $\begingroup$ TargetDevice -> {"GPU", 2} ? $\endgroup$ – Fortsaint Jan 11 '19 at 17:05
  • $\begingroup$ similar performance, also I try 'TargetDevice -> {"GPU",{1,2}}' $\endgroup$ – Kiril Danilchenko Jan 11 '19 at 17:52
  • 1
    $\begingroup$ @KirilDanilchenko You cannot use GPU because RandomForest method uses Intel DAAL library under the hood. To speed up training, use option PerformanceGoal -> "DirectTraining". $\endgroup$ – Alexey Golyshev Jan 12 '19 at 12:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.