malmaud
Windows support?
Now that TensorFlow supports Python 3.5 on Windows, is it possible for TensorFlow.jl to be supported on Windows?
I would say this is a low priority for me personally, but the kind of thing I'd welcome contributions on. Perhaps in a month or two I can look into it myself.
It is also a low priority for me.
One thing I would like to try out, is if this works inside the Linux Subsystem for Windows 10. Julia does (infact I hear it works a lot better than julia does in windows generally). and I have heard CUDA does. If anyone tries it out, I would appreciate a report back. Can mention it in the docs (as to if it works, or does not).
Only have win7 unfortunately, so can't try this unfortunately. Would be happy to do testing otherwise.
It at least works in CPU-only mode on the Linux Subsystem. Haven't tried GPU yet.
On Sun, Apr 23, 2017 at 2:21 AM niczky12 [email protected] wrote:
Only have win7 unfortunately, so can't try this unfortunately. Would be happy to do testing otherwise.
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/malmaud/TensorFlow.jl/issues/204#issuecomment-296422629, or mute the thread https://github.com/notifications/unsubscribe-auth/AA8SvUo76qDzodhjAPYoeornfNeMqYsrks5ryu3wgaJpZM4M633c .
Can someone point me the direction of some scripts which show how I could use PyCall to run tensorflow from Julia? I'm having troubles replicating the usual with graph.as_default(): block of the Python call within Julia.
https://github.com/malmaud/TensorFlow.jl/blob/master/src/py.jl#L19 shows how to use Python's with from PyCall.
Does anyone have any idea how difficult an issue this might be? Anyone looked at it at all?
It's probably beyond my time and (current) knowledge but I would like to see it work also.
Windows 7 64-bit, Julia 1.0
(It's will harm the adoption of Julia if lots of issues make Windows a 2nd-class citizen on Julia -- not a criticism of the developers and maintainers of this package or any other specifically but as a group it's important to maintain mind share across all platforms.
I can switch to Linux but not everyone can, and I don't really want to do that.)
To really look at this though I am going to need to get a dev environment tested first probably.
Check #302. Getting that far was a days work.
However, the segfaults are mm mysterious. Which on the one hand means hard to fix. But on the other hand we are now many versions later in terms of libtensorflow, So there is a chance it basically might just work. Since we don't know why it was failings.
In theory all that needed to be done is to setup do download libtensorfow.DLL like in #302. Because PyCall is now python 3 by default.
The following might not be relevant, but I started in with the TensorFlow install/update and then the test offered on the TensorFlow site the resulting warning explicitly references "segmentation faults":
> The HDF5 header files used to compile this application do not match
> the version used by the HDF5 library to which this application is linked.
> Data corruption or **segmentation faults** may occur if the application continues.
While lots of things may cause segmentation faults, it is interesting that this is the main problem mentioned above.
More complete message:
<# C:\ #> python -c "import tensorflow as tf; print(tf.version)" C:\ProgramData\Anaconda3\lib\site-packages\h5py_init_.py:36: FutureWarning: Conversion of the second argument of issu bdtype from
floattonp.floatingis deprecated. In future, it will be treated asnp.float64 == np.dtype(float).type. from ._conv import register_converters as _register_converters Warning! HDF5 library version mismatched error The HDF5 header files used to compile this application do not match the version used by the HDF5 library to which this application is linked. Data corruption or segmentation faults may occur if the application continues. This can happen when an application was compiled by one version of HDF5 but linked with a different version of static or shared HDF5 library. You should recompile the application or check your shared library related settings such as 'LD_LIBRARY_PATH'. You can, at your own risk, disable this warning by setting the environment variable 'HDF5_DISABLE_VERSION_CHECK' to a value of '1'. Setting it to 2 or higher will suppress the warning messages totally. Headers are 1.10.1, library is 1.10.2 SUMMARY OF THE HDF5 CONFIGURATION =================================
Not related. I think. HDF5 is not related to the core of TensorFlow, (in python, C, or julia).
Any updates on this issue? I tried to install tensorflow but I got the following error:
Error: Error building TensorFlow:
│ ERROR: LoadError: TensorFlow.jl does not support Windows. Please see https://github.com/malmaud/TensorFlow.jl/issues/204
│ Stacktrace:
│ [1] error(::String) at .\error.jl:33
│ [2] top-level scope at C:\Users\Sannan.julia\packages\TensorFlow\q9pY2\deps\build.jl:13
│ [3] include(::String) at .\client.jl:457
│ [4] top-level scope at none:5
│ in expression starting at C:\Users\Sannan.julia\packages\TensorFlow\q9pY2\deps\build.jl:12
└ @ Pkg.Operations C:\Users\Sannan\AppData\Local\Programs\Julia 1.5.0\share\julia\stdlib\v1.5\Pkg\src\Operations.jl:942
Any updates on this issue? I tried to install tensorflow but I got the following error:
Error: Error building
TensorFlow: │ ERROR: LoadError: TensorFlow.jl does not support Windows. Please see #204 │ Stacktrace: │ [1] error(::String) at .\error.jl:33 │ [2] top-level scope at C:\Users\Sannan.julia\packages\TensorFlow\q9pY2\deps\build.jl:13 │ [3] include(::String) at .\client.jl:457 │ [4] top-level scope at none:5 │ in expression starting at C:\Users\Sannan.julia\packages\TensorFlow\q9pY2\deps\build.jl:12 └ @ Pkg.Operations C:\Users\Sannan\AppData\Local\Programs\Julia 1.5.0\share\julia\stdlib\v1.5\Pkg\src\Operations.jl:942
I get the same error lately
@DrKrar I think this library is outdated. The last patch was four years ago, and TensorFlow changes frequently.