[BUG] Results for bagging different for opencl #3606
Description
The bagging example has an accuracy of 91.45% for CPU and CUDA. For OpenCL, the accuracy is only 27.06%
Description
The result should be independent from used platform (except rounding errors).
Reproducible Code and/or Steps
System Information
All run in Debug mode, explaining the long prediction times.
ArrayFire v3.9.0 (CPU, 64-bit Windows, build f4157374e)
[0] AMD: AMD A10-7870K Radeon R7, 12 Compute Cores 4C+8G
Accuracy on testing data: 91.45
Prediction time: 1829.4632
D:\github\willyborn\arrayfire\build\examples\machine_learning\Debug\bagging_cpu.exe (process 13640) exited with code 0.
------------------
ArrayFire v3.9.0 (CUDA, 64-bit Windows, build f4157374e)
Platform: CUDA Runtime 11.8, Driver: 12060
[0] NVIDIA GeForce GTX 750 Ti, 2048 MB, CUDA Compute 5.0
Accuracy on testing data: 91.45
Prediction time: 668.5332
D:\github\willyborn\arrayfire\build\examples\machine_learning\Debug\bagging_cuda.exe (process 11996) exited with code 0.
------------------
ArrayFire v3.9.0 (OpenCL, 64-bit Windows, build f4157374e)
[0] AMD: Spectre, 7211 MB -- OpenCL 2.0 AMD-APP (3224.5) -- Device driver 3224.5 -- FP64 Support: True
-1- NVIDIA: NVIDIA GeForce GTX 750 Ti, 2047 MB -- OpenCL 3.0 CUDA -- Device driver 560.81 -- FP64 Support: True
Accuracy on testing data: 27.06
Prediction time: 15.0551
D:\github\willyborn\arrayfire\build\examples\machine_learning\Debug\bagging_opencl.exe (process 12884) exited with code 0.
Checklist
- Using the latest available ArrayFire release
- GPU drivers are up to date