[theano-users] Shared variable not freeing memory after each code run

2016-08-25 Thread Michael Klachko
I tried running this simple code which copies two large arrays to GPU memory for dot product computation. import numpy as np import theano import theano.tensor as T a = np.asarray(np.random.uniform(-1,1, (1,4)), dtype=np.float32) b = np.asarray(np.random.uniform(-1,1, (4,2)), dt

[theano-users] Strange Momentum Implementation in the Tutorial

2016-09-08 Thread Michael Klachko
http://nbviewer.jupyter.org/github/craffel/theano-tutorial/blob/master/Theano%20Tutorial.ipynb (section 24): updates = [] for param in params: param_update = theano.shared(param.get_value()*0., broadcastable=param. broadcastable) updates.append((param, param - learning_rate*param_update))

[theano-users] How to find Float64 Variable?

2016-10-24 Thread Michael Klachko
I enabled warn_float64 flag in theano config, and now when I'm running my neural net code, I'm getting the following: c:\Theano\theano\gof\graph.py:447: UserWarning: You are creating a TensorVariable with float64 dtype. You requested an action via the Theano flag warn_float64={ignore,warn,raise

[theano-users] Re: How to find Float64 Variable?

2016-10-24 Thread Michael Klachko
. I added dtype=theano.config.floatX and the warning disappeared. On Monday, October 24, 2016 at 8:23:01 AM UTC-7, Michael Klachko wrote: > > I enabled warn_float64 flag in theano config, and now when I'm running my > neural net code, I'm getting the following: >

[theano-users] Re: Is CuDnn 5105 supported ?

2016-10-24 Thread Michael Klachko
Yes, it's supported, I'm using it right now (CUDA 8.0 on Ubuntu 14.04): >>> import theano Using gpu device 0: TITAN X (Pascal) (CNMeM is enabled with initial size: 30.0% of memory, cuDNN 5105) >>> print theano.__version__ 0.9.0dev3.dev-20fd30a38d34687e9d944140042762ca9fca6276 On Saturday, Oct

Re: [theano-users] Re: Is CuDnn 5105 supported ?

2016-11-09 Thread Michael Klachko
y, October 24, 2016 at 9:38:17 AM UTC-7, nouiz wrote: >>> >>> What errors do you have? Delete your Theano cache, just in case and be >>> sure to use Theano dev version. The last release don't support it I think. >>> >>> Fred >>> >>> O

Re: [theano-users] How to reproduce training results?

2016-11-09 Thread Michael Klachko
Andre, can you please post your theano config and your convnet parameters? To debug this I would first try running on CPU only, then MLP, then convnet with no pooling and tanh, with CuDNN disabled, then partially enable one thing at a time. Also, check the obvious things like using the same initial

Re: [theano-users] Re: Is CuDnn 5105 supported ?

2016-11-09 Thread Michael Klachko
t; tutilization with cudnn 5105. If I use cuda backend, I am only getting >>> about 35% utilization. >>> Anyidea why this might be so ? >>> >>> On Monday, October 24, 2016 at 9:38:17 AM UTC-7, nouiz wrote: >>>> >>>> What errors do you have?

Re: [theano-users] Re: Is CuDnn 5105 supported ?

2016-11-10 Thread Michael Klachko
s fast as it can ? > > On Wednesday, November 9, 2016 at 5:36:14 PM UTC-7, Michael Klachko wrote: >> >> Ragav, so when GPU is 98% utilized, is the training faster than when it's >> 35% utilized? Have you timed it? >> >> On Wed, Nov 9, 2016 at 4:09 PM, Ragav Venk

Re: [theano-users] different results for different runs on same GPU

2016-11-11 Thread Michael Klachko
This is an old issue, see: https://groups.google.com/forum/#!topic/theano-users/Q9tD4Af_7ho On Friday, November 11, 2016 at 10:07:22 AM UTC-8, Amin Farajian wrote: > > Hi Fred, > I just followed your suggestion and hard coded the changes in my Theano > package, and ran multiple experiments with

Re: [theano-users] Re: Is CuDnn 5105 supported ?

2016-11-11 Thread Michael Klachko
allowgc > =True. > > On Thursday, November 10, 2016 at 4:47:38 PM UTC-7, Michael Klachko wrote: >> >> Yes. It depends on the size of your network/input - the smaller it is, >> the harder it is to keep 3k cores busy all the time. >> Regarding timing, you don't need to

Re: [theano-users] Re: Is CuDnn 5105 supported ?

2016-11-12 Thread Michael Klachko
ow do I find that out ? > > On Friday, November 11, 2016 at 10:00:39 PM UTC-7, Michael Klachko wrote: >> >> Do both versions use CuDNN? If gpu0 version didn't use it, that would >> explain the difference. Also, look at CPU usage for gpu0 version - it could >>

Re: [theano-users] Re: Is CuDnn 5105 supported ?

2016-11-14 Thread Michael Klachko
esan >> > wrote: >>> >>>> Both are using CUdNNs.. I am wondering if some ops are running on the >>>> CPU, how do I find that out ? >>>> >>>> On Friday, November 11, 2016 at 10:00:39 PM UTC-7, Michael Klachko >>>> wrote: &g

[theano-users] Re: How to quantize theano variable to arbitrary resolution?

2017-02-02 Thread Michael Klachko
I'm also interested in how to do that efficiently. Currently, when I want to quantize weights, I pull them from GPU using get_value(), quantize them in Python, and them import them back to GPU with set_value(). But, of course, this is very slow. For binary quantization, I can use Theano functio

Re: [theano-users] Using multiple gpus on windows using theano,keras

2017-03-16 Thread Michael Klachko
Hi Fred, are multiple GPUs on Windows supported if I use each GPU to train a separate network? For example, can I launch two theano programs, where one is using device=gpu0, and the other device=gpu1? On Monday, March 13, 2017 at 3:15:04 PM UTC-7, nouiz wrote: > > Hi, > > For many reasons, mul

[theano-users] Re: Theano 0.9.0rc1 on Windows 10 x64 and Ubuntu 16.04 have same pygpu errors

2017-03-23 Thread Michael Klachko
I have the same error. I did both conda commands to install libgpuarray and pygpu, and both seemed to be installed successfully. However, I get this: In [8]: import theano ERROR:theano.gpuarray:Could not initialize pygpu, support disabled Traceback (most recent call last): File "c:\Theano\t

[theano-users] Re: Theano 0.9.0rc1 on Windows 10 x64 and Ubuntu 16.04 have same pygpu errors

2017-03-23 Thread Michael Klachko
I updated my CUDA from 7.5 to 8.0, and the error changed to: In [1]: import theano ERROR:theano.gpuarray:Could not initialize pygpu, support disabled Traceback (most recent call last): File "c:\Theano\theano\gpuarray\__init__.py", line 164, in use(config.device) File "c:\Theano\theano\

Re: [theano-users] Re: How to find Float64 Variable?

2017-03-25 Thread Michael Klachko
I just updated to gpuarray backend, and suddenly I started seeing float64 warnings. I printed the graph (float64 shown in bold red below), then set it up so that float64 triggers pdb. However, I still can't figure out which variable it is. By using UP in pdb I eventually got to my training funct

Re: [theano-users] Re: How to find Float64 Variable?

2017-03-25 Thread Michael Klachko
Oh, looks like my message got truncated. Here's the complete graph: C:\ProgramData\Miniconda2\lib\site-packages\theano\gof\type.py:405: UserWarning: You are creating a TensorVariable with float64 dtype. You requested an action via the Theano flag warn_float64={ignore,warn,raise,pdb}. return u

[theano-users] Re: theano_alexnet "train.py"

2017-03-28 Thread Michael Klachko
Has anyone here trained Alexnet successfully on Imagenet dataset? I'm trying to decide if it's worth it trying to make it work with Theano, or switch to Tensorflow... -- --- You received this message because you are subscribed to the Google Groups "theano-users" group. To unsubscribe from th

[theano-users] Re: Theano 0.9.0rc1 on Windows 10 x64 and Ubuntu 16.04 have same pygpu errors

2017-03-28 Thread Michael Klachko
In my case, the problem was related to the old version of mingw64 that I installed in the past. I removed it, installed mingw toolchain with conda, and updated the path to it in theanorc, now everything works. On Thursday, March 23, 2017 at 8:35:04 PM UTC-7, Michael Klachko wrote: >

[theano-users] Re: The architecture of Theano

2017-03-28 Thread Michael Klachko
You can install everything using conda (use mingw's gcc instead of Visual Studio): http://deeplearning.net/software/theano/install_windows.html#requirements-installation-through-conda-recommended Here's the summary of the process: 1. install CUDA 2. copy the CuDNN files 3. install miniconda 4. i

Re: [theano-users] Re: theano_alexnet "train.py"

2017-03-28 Thread Michael Klachko
idation images from ImageNet > standard dataset. Later I will post the errors. But I think that for some > people AlexNet with theano works. > > Goffredo > > Il 28/Mar/2017 19:14, "Michael Klachko" > ha scritto: > >> Has anyone here trained Alexnet success

Re: [theano-users] Re: How to find Float64 Variable?

2017-04-11 Thread Michael Klachko
stien < frederic.bast...@gmail.com> wrote: > Just an updated, we fixed many such useless warning. Update Theano to the > dev version. If you still see them, tell us. > > Fred > > On Sat, Mar 25, 2017 at 4:22 AM Michael Klachko > wrote: > >> Oh, looks like my message

Re: [theano-users] Re: How to find Float64 Variable?

2017-04-27 Thread Michael Klachko
.9.0-py27_0\Lib\site-packages\theano >> folder? >> >> On Tue, Apr 11, 2017 at 8:02 AM, Frédéric Bastien < >> frederic.bast...@gmail.com> wrote: >> >> Just an updated, we fixed many such useless warning. Update Theano to the >> dev version. If you st

Re: [theano-users] Re: How to find Float64 Variable?

2017-04-27 Thread Michael Klachko
Theano > > I added "-U" > > On Thu, Apr 27, 2017 at 5:45 PM Michael Klachko > wrote: > >> It does not want to update: >> >> C:\Users\Michael\>pip install --no-deps git+https://github.com/Theano/ >> Theano.git#egg=Theano >> Requirement al

Re: [theano-users] Re: RuntimeError: Mixed dnn version. The header is version 5105 while the library is version 5110.

2017-04-27 Thread Michael Klachko
I had the same problem, and I solved it by overwriting all versions of the files I could find, with the latest version. Also, I used CuDNN v6 with the latest bleeding edge Theano, and it seems to work fine. On Thursday, April 20, 2017 at 3:07:38 PM UTC-7, Robert wrote: > > I come from a Windows

Re: [theano-users] Re: Theano 0.9.0rc1 on Windows 10 x64 and Ubuntu 16.04 have same pygpu errors

2017-06-05 Thread Michael Klachko
be imported successfully, but implementing > pygpu.test() results in those errors. > > Can you post you workable .theanorc file? > > > On Tuesday, 28 March 2017 13:20:44 UTC-4, Michael Klachko wrote: > >> In my case, the problem was related to the old version of min

[theano-users] Grouped Convolution Error

2017-07-24 Thread Michael Klachko
I'm trying the new grouped convolutions feature in the latest Theano version, so I ran a simple convnet with CIFAR-10: 32x32 RGB input images (batch size = 128), and the first convolutional layer has 9 feature maps. I want to have 3 feature maps per color, so if I understand it correctly, I sho

Re: [theano-users] How to change the learning rate dynamically

2017-10-03 Thread Michael Klachko
The learning rate schedule is defined in this line: updates.append((l_r,0.95*l_r)), and is used in this line: train_model = theano.function([index], cost, updates=updates ... If you don't understand what's going on, read about theano functions. On Thursday, September 7, 2017 at 6:58:34 AM UTC

[theano-users] Could not initialize pygpu error with CUDA 9.0 and latest Theano

2018-05-08 Thread Michael Klachko
I have CUDA 9.0 and CuDNN 7.0.5 on my Ubuntu 16.04, and Tensorflow works fine. In order to install theano, I first installed miniconda, then ran "conda install theano pygpu" and it seemed to have installed fine. However, here's what I get: $ python Python 3.6.5 |Anaconda, Inc.| (default, A

Re: [theano-users] Could not initialize pygpu error with CUDA 9.0 and latest Theano

2018-05-10 Thread Michael Klachko
may be an error message from something else. > > First things first, since you are on cuda 9.0, I would recommend that you > update your driver to 384.111 or 390.*. If that doesn't help, then I'll > need some help reproducing the problem since I don't get that in any o

Re: [theano-users] Could not initialize pygpu error with CUDA 9.0 and latest Theano

2018-05-10 Thread Michael Klachko
e cuda version installed to have TF working. > > Le jeu. 10 mai 2018 16:28, Michael Klachko > a écrit : > >> After struggling with this error for a day, I decided to upgrade CUDA to >> 9.1 and CuDNN to 7.1. After that I got "your driver might be too old" >> er