[julia-users] Re: ANN: CUDAdrv.jl, and CUDA.jl deprecation
Good news! I had wished there's would be some integration in several CUDA packages. By the way, is there's any plan for 'standard' GPU array type, such as https://github.com/JuliaGPU/GPUArrays.jl ? CUDArt, CUDAdrv has its own CUDA array type and there's package such as ArrayFire.jl For example, if I add package wrapping new NVIDIA library such as cuRAND, which GPU array type should I support in that package? Best, Kyunghun
[julia-users] Re: Julia Reference Card
Thank you for great work! I have to print it out and place on my desk. By the way, PDF file has title 'Python 2.5 Reference Card'. (not file name, my acrobat reader shows that title on the window title. maybe that is in the PDF file metadata?)
[julia-users] Re: Announcing TensorFlow.jl, an interface to Google's TensorFlow machine learning library
Wonderful jobs, Jonathan!
I'd better try this version rather than use TensorFlow in python.
Does it based on PyCall package?
-Kyunghun
2016년 9월 1일 목요일 오전 7시 31분 58초 UTC+9, Jonathan Malmaud 님의 말:
>
> Hello,
> I'm pleased to announce the release of TensorFlow.jl, enabling modern
> GPU-accelerated deep learning for Julia. Simply run Pkg.add("TensorFlow")
> to install and then read through the documentation at
> https://malmaud.github.io/tfdocs/index.html to get started. Please file
> any issues you encounter at https://github.com/malmaud/TensorFlow.jl.
>
> TensorFlow.jl offers a convenient Julian interface to Google's TensorFlow
> library. It includes functionality for building up a computation graph that
> encodes a deep-learning model and automatically minimizing an arbitrary
> loss function with respect to the model parameters. Support is included for
> convolutional networks, recurrent networks with LSTMs, the Adam
> optimization algorithm, loading images, and checkpointing model parameters
> to disk during training
>
> I'm hopeful that this package will ensure Julia remain a first-class
> citizen in world of modern machine learning and look forward to the
> community's help in getting it to match or exceed the capabilities of the
> official Python TensorFlow API.
>
> -Jon
>
Re: [julia-users] Performance issue of signal processing code compared to MATLAB
Thank you for kind answer. Always I am afraid of unintended copy of buffers because I'm a such a C/C++ guy. I have read the performance tips when it is early 0.3 version. I surprised that there were lots of updates since then. (julia is developing so fast!) I will try the profiling functions and find unintended allocations. By the way, my another wish is about performance of loading packages. In my code, Image.jl takes ~5 sec for just loading libraries, (new) Interpolation.jl takes even ~20 sec. (maybe cause of metaprogramming codes I think) I heard there will be some package pre-loading or cache feature in v0.4. Is there any updates about that? 2015년 2월 12일 목요일 오전 12시 34분 12초 UTC+9, Tim Holy 님의 말: A[:, indx] currently makes a copy. Try replacing it with slice(A, :, indx) (if you're on julia 0.4) or use ArrayViews if you're on julia 0.3. For performance questions, if you aren't using the tools advertised at http://docs.julialang.org/en/release-0.3/manual/performance-tips/ you will likely find them to be a big help. --Tim On Wednesday, February 11, 2015 07:25:38 AM Kyunghun Kim wrote: Hi, all. I am sorry that I am writing repeating these questions again. (performance compared to ~) I have some signal processing code written in MATLAB, and rewriting the code with Julia. The signal processing function take about 1024 x 1024 floating number array as input called in loop about 100~1000 times. Here is core function of algorithm: MATLAB version: https://gist.github.com/moon6pence/3e60772943f206842d31 0.16 sec per each call Julia version: https://gist.github.com/moon6pence/4b43c63cb4240b31ea10 1.4 sec per each call Not only julia code is unusually slow, but MATLAB code is also unusually fast. (Naive C++ implementation of this code takes 100~200 sec, maybe MATLAB JIT compiler is doing very well in SIMD) I will dig up this julia code line-by-line to find which line takes much time. But before then, I want get this code checked if there is any mistake on code for performance. (or give me tips for vectorized code) Thanks.
[julia-users] Performance issue of signal processing code compared to MATLAB
Hi, all. I am sorry that I am writing repeating these questions again. (performance compared to ~) I have some signal processing code written in MATLAB, and rewriting the code with Julia. The signal processing function take about 1024 x 1024 floating number array as input called in loop about 100~1000 times. Here is core function of algorithm: MATLAB version: https://gist.github.com/moon6pence/3e60772943f206842d31 0.16 sec per each call Julia version: https://gist.github.com/moon6pence/4b43c63cb4240b31ea10 1.4 sec per each call Not only julia code is unusually slow, but MATLAB code is also unusually fast. (Naive C++ implementation of this code takes 100~200 sec, maybe MATLAB JIT compiler is doing very well in SIMD) I will dig up this julia code line-by-line to find which line takes much time. But before then, I want get this code checked if there is any mistake on code for performance. (or give me tips for vectorized code) Thanks.
[julia-users] Re: dlopen could not load 32 bit module on Win7 64 bits?
You can run 32 bit applications in 64 bit OS, but you have to use only 64
bit DLL/libraries in 64 bit applications.
It is not available because they use different size of memory address.
2015년 2월 5일 목요일 오후 2시 46분 22초 UTC+9, 进陆 님의 말:
I am using Julia Version 0.3.4 (2014-12-26 10:42 UTC)( Official release,
x86_64-w64-mingw32) coming with Juno bundle on Win7 64 bits
I have both Anaconda-python27(which is 32 bits) and
python-3.4.2.amd64(which is 64 bits)
But I can't let PyCall run, so I traced it and found that dlopen could not
load 32 bit module on Win7 64 bits.
But why a 32bit application can run 64bit OS no problem? Can dlopen
supports both 32 and 64 bits module?
Thanks.
[quote]
julia isfile(E:\\prg\\py\\Anaconda\\python27.dll)
true
julia dlopen(E:\\prg\\py\\Anaconda\\python27.dll,
RTLD_LAZY|RTLD_DEEPBIND|RTLD_GLOBAL) #this is 32bits DLL
ERROR: could not load module E:\prg\py\Anaconda\python27.dll: The
specified module could not be found.
julia dlopen(E:\\prg\\py\\python-3.4.2.amd64\\python34.dll,
RTLD_LAZY|RTLD_DEEPBIND|RTLD_GLOBAL)#this is 64 bits DLL
Ptr{Void} @0x0737a500
[/quote]
