CharRNN has a TypeInitializationException about gradient-samples HOT 7 CLOSED

@SnowSnakz this looks like a problem with environment configuration. I suggest to add

GradientSetup.EnsureInitialized();

after GradientEngine.UseEnvironmentFromVariable(); in CharRNNProgram.cs. It should give more meaningful exception.

P.S. it would be easier to help if you can provide the stack trace for the inner exception too (e.g. for the ArgumentNullException).

from gradient-samples.

@SnowSnakz actually, I think the unclear error has been fixed in the upcoming RC0.

You need Python 3.4+ (3.7 recommended) installed with TensorFlow 1.15.0. See instructions.

Alternatively, you can reference prepackaged TensorFlow from NuGet: LostTech.TensorFlow.Python. Disadvantage is that then every TensorFlow project deploys its own runtime, so launching every sample will consume ~1GB of disk space. This is the approach RL-MLAgents sample takes: relevant code.

from gradient-samples.

@lostmsu I am using Python 3.6 with tensorflow==1.15 installed.
I will try running the sample with GradientSetup.EnsureInitialized();

from gradient-samples.

I am getting the same error, no further details...

System.ArgumentNullException
  HResult=0x80004003
  Message=Value cannot be null.
Parameter name: source
  Source=Gradient.Runtime
  StackTrace:
   at Gradient.Tools.OnlyOneOrDefault[T](IEnumerable`1 source)
   at Gradient.GradientEngine.GetEnvironment()
   at Gradient.GradientEngine.Initialize()
   at Gradient.GradientEngine.EnsureInitialized()
   at CharRNN.CharRNNProgram.Main(String[] args) in C:\Users\SnowSnakz\Downloads\Gradient-Samples-master\Gradient-Samples-master\CharRNN\CharRNNProgram.cs:line 23

from gradient-samples.

Ah, I see that I used GradientEngine rather than GradientSetup...
If I use GradientSetup, it starts working

This trial version of Gradient will expire in 98 days on Friday, September 25, 2020
using Python: 3.6-X64 at C:\Users\SnowSnakz\AppData\Local\Programs\Python\Python36\
reading text file

and it continues fine

from gradient-samples.

@SnowSnakz yeah, one of them will have to go or be named better. The GradientEngine is a lower-level thing, that does not care about presence or lack of TensorFlow, it just picks the Python interpreter with the highest version number. In your case it probably picks something, that does not have TensorFlow in it.

from gradient-samples.

Created an issue for Gradient to clear this confusion: losttech/Gradient#31

from gradient-samples.