MXNetSharp icon indicating copy to clipboard operation
MXNetSharp copied to clipboard

DQN example with MKL build error

Open kevmal opened this issue 5 years ago • 1 comments

Specifically the python MKL build of libmxnet (mxnet-cu101mkl==1.6.0b20191125) errors on copyToRun after execRun has been used. To reproduce click Use DQN and update the model. Whether it updates while in keyboard mode or DQN mode it will throw:

"MXNDArrayWaitAll: [16:59:32] C:\Jenkins\workspace\mxnet\mxnet\src\ndarray\ndarray.cc:1132: Check failed: to.IsDefaultData(): "
"   at MXNetSharp.Interop.Helper.throwOnError(String call, Int32 returnCode) in E:\profile\Projects\repo\MXNetSharp\MXNetSharp\interop.fs:line 194
   at MXNetSharp.Interop.MXNDArray.waitAll() in E:\profile\Projects\repo\MXNetSharp\MXNetSharp\interop.fs:line 1170
   at MXNetSharp.NDArray.WaitAll() in E:\profile\Projects\repo\MXNetSharp\MXNetSharp\ndarray.fs:line 230
   at [email protected](Unit _arg1) in E:\profile\Projects\repo\MXNetSharp\Examples\DQN\DQN.fsx:line 695
   at FSI_0004.copyToRun() in E:\profile\Projects\repo\MXNetSharp\Examples\DQN\DQN.fsx:line 685
   at FSI_0004.updateModel(TrainState ts) in E:\profile\Projects\repo\MXNetSharp\Examples\DQN\DQN.fsx:line 767
   at [email protected](TrainState ts) in E:\profile\Projects\repo\MXNetSharp\Examples\DQN\DQN.fsx:line 781
   at [email protected](Int32 frameMove, GameBoard s, TrainState ts) in E:\profile\Projects\repo\MXNetSharp\Examples\DQN\DQN.fsx:line 615
   at [email protected](FSharpOption`1 _arg2) in E:\profile\Projects\repo\MXNetSharp\Examples\DQN\DQN.fsx:line 644
   at Microsoft.FSharp.Control.AsyncPrimitives.CallThenInvokeNoHijackCheck[a,b](AsyncActivation`1 ctxt, FSharpFunc`2 userCode, b result1) in F:\workspace\_work\1\s\src\fsharp\FSharp.Core\async.fs:line 416
   at <StartupCode$FSharp-Core>[email protected](AsyncActivation`1 ctxt) in F:\workspace\_work\1\s\src\fsharp\FSharp.Core\mailbox.fs:line 313
   at Microsoft.FSharp.Control.Trampoline.Execute(FSharpFunc`2 firstAction) in F:\workspace\_work\1\s\src\fsharp\FSharp.Core\async.fs:line 109"

mxnet-cu101==1.6.0b20191125 works fine

kevmal avatar Jan 24 '20 00:01 kevmal

Most likely an issue with multithreading in MXNet

See: https://github.com/apache/incubator-mxnet/issues/16431

kevmal avatar Jan 29 '20 00:01 kevmal