Suggested update to add installation for Ubuntu 22.04.1 LTS #2123
Conversation
|
Thanks @OptogeneticsandNeuralEngineeringCore! I cannot test it, but I just left a minor comment to avoid having to go through all the prompts. |
|
Perhaps the following helps: |
|
I have tried the headless python, to no avail. With 'export QT_DEBUG_PLUGINS=1' I find the most relevant info: Starting GUI... qt.core.plugin.loader: In /home/onecore/anaconda3/envs/DEEPLABCUTe/plugins/platforms/libqeglfs.so: qt.core.plugin.loader: In /home/onecore/anaconda3/envs/DEEPLABCUTe/plugins/platforms/libqminimal.so: But cannot seem to find a solution... |
|
Hello, just checking in on this. I don't have the time to figure out the installation of the conda env failure, but the pip installation is still helpful. Perhaps we could just push this out with a note suggestion to only use pip? |
|
I'm not quite sure I am following; can you clarify what is diff to ubuntu 20.04 install guide? also DLC can be used with latest CUDA, its not pinned to 11? |
|
|
||
| `nvcc --version` | ||
|
|
||
| If error messages, read them carefully as they often tell you how to fix it, or what to google :D |
There was a problem hiding this comment.
| If error messages, read them carefully as they often tell you how to fix it, or what to google :D | |
| If you see error messages, read them carefully as they often tell you how to fix it, or what to google :D |
|
|
||
| Install CUDA: | ||
|
|
||
| ```python |
There was a problem hiding this comment.
| ```python | |
| ```bash |
| `gcc --version` | ||
|
|
||
| output: | ||
| ```python |
There was a problem hiding this comment.
| ```python | |
| ```bash |
| output: | ||
| ```python | ||
| gcc --version | ||
| gcc (Ubuntu 9.3.0-17ubuntu1~20.04) 9.3.0 |
There was a problem hiding this comment.
This should be 22.04 as well, right?
| and run: | ||
|
|
||
| `bash Anaconda3-2021.05-Linux-x86_64.sh` |
There was a problem hiding this comment.
Here we could be a bit more explicit that the filename should be updated depending on the file downloaded.
We could also get users to install miniconda instead of Anaconda, which would be a bit more lightweight. Users should be able to do this entirely from the command line (haven't tested this installation exactly, as I do something similar but when installing miniconda in docker containers):
Note that depending on their system architecture, different versions should be installed (e.g., could be -aarch64.sh instead of -x86.sh)
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86.sh -O miniconda.sh -q
bash miniconda.sh -b
| If no git yet: | ||
|
|
||
| `sudo apt install git-all` |
There was a problem hiding this comment.
git is installed above, we should remove this to shorten the description a bit.
| DeepLabCut can be installed via: | ||
| * Docker: We strongly recommend for Ubuntu users to use Docker (https://hub.docker.com/r/deeplabcut/deeplabcut) - it's a much more reproducible environment. | ||
| * A conda file provided via the DeepLabCut website. This may be a bit more advanced in the development | ||
| * A simple installation through pip: either full GUI (recommeneded) or without GUI (called: DLCLite) |
There was a problem hiding this comment.
Once conda, cuda and all other "system" packages are installed, I'm not sure we should continue with the description on how to run DLC (installing via Conda or PIP, with or without GUI etc.).
I would prefer linking to the existing documentation, instead of duplicating it. The reasoning being that if anything changes in DLCs installation process, we'll need to come modify these files as well as the original ones (which is unlikely to happen), hence we would have deprecated installation docs here.
If we do decide to keep this detailed installation information, I would try to make it a bit more condensed, as noted in the comments below.
| Create an enviroment: | ||
|
|
||
| `conda create --name DLCenvGUI python=3.8` | ||
|
|
||
| Activate the enviroment: | ||
|
|
||
| `conda activate DLCenvGUI` | ||
|
|
||
| Install DLC: | ||
|
|
||
| `pip install 'deeplabcut[gui,tf]'` | ||
|
|
||
| Start the GUI with: | ||
|
|
||
| `python -m deeplabcut` | ||
|
|
||
| **[Do DeepLabCut stuff.](https://deeplabcut.github.io/DeepLabCut/docs/standardDeepLabCut_UserGuide.html)** |
There was a problem hiding this comment.
I believe this would be nicer in a bit more concise way (otherwise it spreads out a lot when reading the documentation):
conda create --name DLCenvGUI python=3.8 # Create an environment
conda activate DLCenvGUI # Activate it
pip install "deeplabcut[gui, tf]" # install DLC
python -m deeplabcut # Start the GUI
|
I think there are quite a few changes we should make before merging this. If we want to go forward with including this installation documentation I should be able to test the installation process, but not on a "fresh" install of Ubuntu 22.04, so I would not be able to test all of the package installations. |
Co-authored-by: n-poulsen <45132115+n-poulsen@users.noreply.github.com>
Co-authored-by: n-poulsen <45132115+n-poulsen@users.noreply.github.com>
|
alright, since we are moving to pytorch now eminently, I vote to close, but I agree we should clean up docs, and also see #2421 which I think is a great base to start edits from. |
|
Oh, interesting. Pytorch sounds great, as does 2421. Yes, please close, but do note that should you ever need someone to test future builds on fresh systems, do reach out to me. That is my life (I set up so many computers based on user needs). So fresh/blank slate systems are my problem to address. Plus/side: I see real fruitfulness in getting full OS images on the Cloud (Google right now), such that they could be shared. It seems (?) like this could be even easier than Docker to spin up than , as they seem to well handle hardware versioning. And...well...GPU POWER. |
E tensorflow/stream_executor/cuda/cuda_blas.cc:2981] Unable to register cuBLAS factory:qt.core.plugin.loader: In /home/oc/anaconda3/envs/DLCenvConda666/plugins/platforms/libqeglfs.so: Plugin uses incompatible Qt library (5.15.0) [release]