I have a anaconda installation with which I configured python 3.6, tensorflow 1.13 on windows 10. The installation works fine except when I try invoking cudnn code i.e conv2d and similar functions when I greeted with the following error: " E tensorflow/stream_executor/cuda/cuda_dnn.cc:324] Loaded runtime CuDNN library: 7.3.1 but source was compiled with: 7.4.1. CuDNN library major and minor version needs to match or have higher minor version in case of CuDNN 7.0 or later version. If using a binary install, upgrade your CuDNN library".
Which TensorFlow and CUDA version combinations are compatible?. From this post, I tried installing cudnn 7.4.1 on my desktop but looks like anaconda keeps using the default cudnn 7.3.1 only and does not reference the one pointed to by my path variables. Any help in terms of how I can force conda to use the cudnn 7.4.1 will be much appreciated.
I don't know if the original poster's problem had been resolved. Just in case someone has the same problem (as I did), one should try this first:
https://towardsdatascience.com/tensorflow-gpu-installation-made-easy-use-conda-instead-of-pip-52e5249374bc
Basically, it recommends you to directly conda tensorflow-gpu and let conda handles cudnn and cuda versions. At least afterward if you conda list it would list the updated (or more current) cudnn version.
But in my case, afterwards, I still had the same error. So I dug a little and found out it was from one of my other packages. You can search, under your anaconda folder, all occurrences of cudnn64_7.dll. For me, it's the mxnet-cu100 package that (even with the newest version) still uses the older cudnn dll. I simply copied the newer cudnn64_7.dll to that site-packages folder (with some risk I guess but at this stage, I had nothing to lose) and it solved the problem.
Related
Just checking if anyone knows whether Anaconda Project has released TensorFlow 2.2 and above for macOS users? From my understanding, this version of TF exists for all platforms except Mac.
Thanks for the help
you have the option to check things like this on the anaconda website. Just go to https://anaconda.org/anaconda/tensorflow and you will see:
linux-64 v2.2.0
win-64 v2.1.0
osx-64 v2.0.0
so the newest version available for macOS from the anaconda channel is (currently) 2.0.0
Alternatively you can also always run
conda search tensorflow
to see which versions are available for your platform
Note
From my understanding, this version of TF exists for all platforms except Mac
From anaconda, yes, but you are definately able to find this version on pypi, i.e. use pip install to install it
Just to precise I'm a total beginner in this. I check on the internet and it seems that nothing match to this problem.
My goal is to run this github which is a facial recognition program: https://github.com/anisayari/easy_facial_recognition
So here is the error:
Class RunLoopModeTracker is implemented in both
/Users/pierre/anaconda3/lib/python3.7/site-packages/cv2/.dylibs/QtCore (0x1086267f0)
and /Users/pierre/anaconda3/lib/libQt5Core.5.9.7.dylib (0x122fc0a80).
One of the two will be used. Which one is undefined.
From what I understand, I just have to indicate which one to use but I don't find anything about how to do it. Also, It seems to be a recurrent error on Mac.
I had this error on my Mac too. Apparently opencv's GUI tools were conflicting with the PYQT libraries that were also installed on my system in my anaconda/lib/ folder. So to give opencv just one set of GUI tools and since I wasn't sure what else was using PYQT in my lib, I chose to use the non-GUI version of opencv, aka opencv-python-headless. I previously had opencv-python-headless installed on my system, and maybe the two packages can't co-exist in the new version of OpenCV or with Catalina.
I removed opencv-python
pip uninstall opencv-python
uninstalled and reinstalled the headless version (which has no GUI tools)
pip uninstall opencv-python-headless
pip install opencv-python-headless
I don't know if it was necessary to remove then reinstall the existing headless package, but that's what I did. I wasn't sure that opencv-headless would find the PYQT in my lib but it didn't have any problems.
In the end, created an env to install graph-tool thru the channel vgauthier.
Thinking by now that I might have as well used earlier version of python or installed a whole new python latest version just for this.
But as it is all part of the learning process to creating dev. env.. (at the same time worrying my Mac has too many installations with much too interlinked dependencies)
Someone kindly let me know if this rings any bell or got any suggestion:
Incompatible library version states that _cairo.cpython-36m-darwin.so requires later version of libcairo.2.dylib.
Errors could have been from:
having on .condarc file, wrong priorities btw channels, so having trouble finding the right files
(modified to the order of conda-forge > ~dir/vgauthier > default
my having given up on installing pygobject after googling it is not supported on Mac OS (which was apparently wrong or I just wasnt searching well)
Solution to easy setup of graph-tool for conda-installed python users:
Get the benefit of GUI (seriously a god option) ->
https://medium.com/#ronie/installing-graph-tool-for-python-3-on-anaconda-3f76d9004979
To still continue with command line (still get the idea from GUI)
conda install -c pkgw-forge gtk3
conda install -c https://conda.anaconda.org/fallen pygobject
(if there are cairo blah blah missing, look up setup directions for igraph which has more information and help, having been an older library)
Do these before installing graph-tool, which Im still wondering why I cannot remove from the system though. ;)
For others, https://git.skewed.de/count0/graph-tool/wikis/installation-instructions#compiler-choice-in-macos-x
It seems that the OpenMDAO packages have been removed from binstar? I have tried the following:
conda search -c https://conda.binstar.org/OpenMDAO openmdao
and
conda install -c openmdao openmdao
both resulting in the following:
PackageNotFoundError: Packages missing in current channels:
- openmdao
We have searched for the packages in the following channels:
- https://conda.anaconda.org/openmdao/win-64
- https://conda.anaconda.org/openmdao/noarch
Was OpenMDAO Anaconda support removed or am I missing something?
Edit: Is this an incompatibility with win64 vs win32?
Edit 2: Yes that is the issue as outlined in their blog post:
Caveat: Thanks to a bug in Traits (we’ve put in a bug report and await a new version), we can’t currently support an Anaconda install on a Windows2012_64bit installation unless you’re using a 32-bit Anaconda installed on the 64-bit architecture. We will announce when that problem is resolved for our 64-bit Windows users.
As per Edit2. There is no update on 64-bit support that I can find.
Caveat: Thanks to a bug in Traits (we’ve put in a bug report and await a new version), we can’t currently support an Anaconda install on a Windows2012_64bit installation unless you’re using a 32-bit Anaconda installed on the 64-bit architecture. We will announce when that problem is resolved for our 64-bit Windows users.
I recently upgraded QGIS to 2.0 using Kyngchaos packages only to discover that my kriging scipts which I run in iPython then threw up the following error:
numpy.core.multiarray failed to import
It seems that the pathname variable assigned by SWIG is incompatible with iPython. I hate path management, it seems to be the single greatest cause of errors on UNIX machines.
Anyway, a rather clumsy fix to this is to copy the Python/2.7/site-packages from from version 1.9 of gdal to version 1.10 (renaming the original 1.10 directory to something appropriately rude) This works for iPython at least and QGIS seems to function still.
Can somebody suggest a more elegant solution?
I have QGIS 2.0 and GDAL 1.10 installed from Kyngchaos pre-compiled distributions, and was also having this error.
It seems it was expecting a newer version of numpy (I was running 1.6), and I'm assuming QGIS/GDAL expect the lib to be in a certain path, since it's a pre-compiled version. So, for me, installing the numpy package also available from Kyngchaos, which is version 1.7.1, solved the problem.