I am on a Windows 10 64 bit system.
Pytorch for cuda has been working successfully for some time.
Today I tried to upgrade to the latest version of Pytorch (1.13) using
conda install pytorch torchvision torchaudio pytorch-cuda=11.7 -c pytorch -c nvidia
Now I cannot import torch. I get the error:
OSError: [WinError 126] The specified module could not be found. Error loading "C:\Users\alan\anaconda3\lib\site-packages\torch\lib\caffe2_nvrtc.dll" or one of its dependencies.
I have tried both in a Jupyter notebook and in the Spyder IDE.
I have tried completely removing Anaconda and reinstalling afresh and then reinstalling Pytorch with no success.
I do not believe I have any other versions of python installed.
The offending dll (caffe2_nvrtc.dll) does seem to be in the file location specified.
I have found various similar problems reported but they all date back to 2020 or earlier and none of them seemed to have a satisfactory solution.
Can anyone point me in the correct direction
I still do not understand why using conda did not work but I tried again using pip
and that did work
I experienced the same problem as yours today.
It turns out that when I use the anaconda prompt then the problem disappears.
Anaconda prompt could do it
Then I speculated that the only difference between these two scenarios is that when I use the anaconda prompt, I use the base anaconda environment, probably somehow the conda environment in spyder is not activated.
So the solution is to open the spyder in anaconda prompt like this:
Then it works.
This picture will tell you the cause for the problem.
problem...
After installing miniconda with python 3.7 I do the following:
conda create --name opencv_test python=3.7
conda activate opencv_test
conda install -c anaconda opencv
Then to test the environement I do:
python
>>> import cv2
>>> exit()
Everything goes according to plan
Then if I launch PyCharm or vs code, set my new environment as the interpreter and create a simple file that contains only
import cv2
On launch I get:
ImportError: DLL load failed: The specified module could not be found.
What I learned from the internets:
This, this and this, all those posts refer to a problem that seems similar, the difference being that the issue for me only happens when using IDEs
What I tried before posting here:
Recreating the environment multiple times switching channels.
As it seemed to be a path issue, I added the paths which contained opencv dll to the environment paths in PyCharm settings one at a time
What I just tried after spending a considerable amount of time creating this post:
conda update conda
This weirdly downgrades python to 3.6
conda create --name opencv_test_2 python=3.6
conda activate opencv_test_2
conda install -c anaconda opencv
This seems to fix all the issues but:
I don't know what just happened and why the issue only appeared on IDEs
I'd like to use python 3.7
So in short, why when using python 3.7 and PyCharm I seem to be unable to import opencv even if it works on command line?
Thanks in advance.
I've gotten myself into some kind of horrible virtualenv mess. Help?!
I manage environments with conda. Until recently, I only had a python2 jupyter notebook kernel, but I decided to drag myself kicking and screaming into the 21st century and installed a python3 kernel; I forget how I did it.
My main (anaconda) python defaults to 2.7.
So here I am, merrily trying to use beautiful soup from inside my shiny new python3 kernel, and I don't seem to be able to do anything to get at whatever environment it's finding packages in. Viz (all from notebook):
from bs4 import BeautifulSoup
-> ImportError: No module named 'bs4'
Ok, fine, I'll install it using shell magic. Right? Right?
! pip install bs4
--> Collecting bs4
Downloading bs4-0.0.1.tar.gz
Requirement already satisfied (use --upgrade to upgrade): beautifulsoup4 in /Users/[MY-USER]/anaconda/lib/python2.7/site-packages (from bs4)
[...]
Successfully built bs4
Installing collected packages: bs4
Successfully installed bs4-0.0.1
from bs4 import BeautifulSoup
-> ImportError: No module named 'bs4'
Oh no. Does it think I'm in a 2.7 env even though I'm running a python3 kernel? That won't do.
! conda info --envs
--> # conda environments:
#
flaskenv /Users/[MY-USER]/anaconda/envs/flaskenv
mesa /Users/[MY-USER]/anaconda/envs/mesa
py35 /Users/[MY-USER]/anaconda/envs/py35
root * /Users/[MY-USER]/anaconda
Ok, I can fix that. One of those is a 3.5 env.
! source activate py35
--> prepending /Users/[MY-USER]/anaconda/envs/py35/bin to PATH
! conda install beautifulsoup4
--> Fetching package metadata .......
Solving package specifications: ..........
# All requested packages already installed.
# packages in environment at /Users/[MY-USER]/anaconda:
#
beautifulsoup4 4.4.1 py27_0
concerning...
! pip install bs4
--> Requirement already satisfied (use --upgrade to upgrade): bs4 in /Users/[MY-USER]/anaconda/lib/python2.7/site-packages
more concerning...
from bs4 import BeautifulSoup
-> ImportError: No module named 'bs4'
ARRGH!!! headdesk Am I going to have to kill the kernel in order to fix this (and re-run a bit of work)? Is killing the kernel even going to work? How do I get my jupyter kernel to know what environment it's supposed to be running under?
thanks!
This is a tricky part of ipython / Jupyter. The set of kernels available are independent of what your virtualenv is when you start jupyter Notebook. The trick is setting up the the ipykernel package in the environment you want to identify itself uniquely to jupyter. From docs on multiple ipykernels,
source activate ENVNAME
pip install ipykernel
python -m ipykernel install --user --name ENVNAME --display-name "Python (whatever you want to call it)"
If you only want to have a single Python 3 kernel, from the conda environment, just use python -m ipykernel install --user and it will reset the default python to the one in the virtualenv.
And yes, you will need to restart the kernel and re-run the prior steps.
See Also Using both Python 2.x and Python 3.x in IPython Notebook
#tschundler's solution works perfectly if your environment has already been created.
If you want to change the default kernel at the creation of your virtual environment and avoid any manual configuration, you just need to add jupyter at the end of the conda command:
conda create --name ENVNAME python=PYTHONVERSION jupyter
The correct kernel will then be used when you use ipython or jupyter notebook.
In my case somehow jupyter wasn't able to 'pick' the virtual environment's python. So I had to edit ~/.local/share/jupyter/kernels/{my_env_name}/kernel.json
and add path to the interpreter
in the argv key.
There is also an easy way here
workon my-virtualenv-name # activate your virtualenv, if you haven't already
pip install tornado==4.5.3
pip install ipykernel==4.8.2
You should now be able to see your kernel in the IPython notebook menu: Kernel -> Change kernel and be able to switch to it (you may need to refresh the page before it appears in the list). IPython will remember which kernel to use for that notebook from then on.
This worked for me. source
pip install --user ipykernel
python -m ipykernel install --user --name=myenv
Output
Installed kernelspec myenv in /home/user/.local/share/jupyter/kernels/myenv
and go to above directory
open kernel.json
{
"argv": [
"/home/user/anaconda3/envs/myenv/bin/python", # path to your virtual environment python
"-m",
"ipykernel_launcher",
"-f",
"{connection_file}"
],
"display_name": "myenv",
"language": "python"
}
I know this is an old question, but nobody has mentioned nb_conda_kernels, which was made to solve this problem. You must have it installed in the environment from which you launched the running notebook, and any other environment which has kernel/s installed (ipykernel, r-irkernel, etc ) will be available as an option (Start Preferred Kernel):
Sharing my experience here, in case someone made the same mistake:
TL;DR
In VSCode I created the virtual environment from an Ubuntu terminal but forgot to activate the WSL extension too. Once I activated the WSL extension, I was able to change the Jupyterkernel to my virtual environment.
Full explanation
Using VSCode I work with the WSL extension (see here). I created my virtual environment from an Ubuntu terminal [Created a new terminal and selected 'Ubuntu (WSL)'] and then went to my Jupyter notebook to change my kernel, but it did not appear in the list.
I tried the solution proposed above of adding the path to kernel.json in my virtual environment, adapting to my case (I don't use conda) but I was surprised to see that the folder bin in my environment did not have a file named python.
The solution was to activate the WSL extension: lower right corner > click on the arrows (Open a remote connection) > reopen folder in WSL.
After this, I could see in the explorer two new entries in the bin folder for Python and Python3 (the arrow on the right side seem to indicate it's a symbolic link).
Virtual environment now shows the python symbolic links
Then I went to my Jupyter notebook and changed the kernel to my virtual environment.
In the end, I did not have to change the path in kernel.json.
P.S: The above is my understanding of my problem and the solution given my non-existing knowledge about Linux, so any comments to complement/improve this answer are very welcome!
I just tried to use pd.HDFStore in IPython Notebook with a Python 3 kernel (Anaconda 2&3 on Ubuntu 14.04)
import pandas as pd
store = pd.HDFStore('/home/Jian/Downloads/test.h5')
but it throws the following error
ImportError: HDFStore requires PyTables, "libhdf5.so.9: cannot open shared object file: No such file or directory" problem importing
I initially thought it's because pytables is somehow missing, but when I check $source activate py34 and $conda list, pytables 3.2.0 is already installed under anaconda python3 environment.
Also, if I switch to Python 2, for example, $source activate py27 and start ipython notebook, it works properly and no import error is thrown.
I guess that I must miss something for configuring pytables under anaconda python 3 env, but I cannot figure it out. Any help is highly appreciated.
Update:
I just tried on a fresh install of Anaconda3-2.3.0-Linux-x86_64 from official website and it ends up with the same error. When I try $locate libhdf5.so.9 in command line, nothing shows up.
This is a known issue that we are working on. When it is fixed, conda update --all will update the libraries and fix the issue.
I'm trying to run my code and I don't know what specific package I need in order to get my import statement to work. Below is my header and I keep getting an error saying ImportError no module named statistics. I have looked at a bunch of different pages to see where I can download a solution, but I am trapped. I know my code works because I ran it on my schools lab. If anyone can help, that' be great!
Just note I am a beginner and am using Linux on my virtual machine with Python 2.7
import sys
import requests
import matplotlib.pyplot as plt
import statistics as stat
Statistics "A Python 2.* port of 3.4 Statistics Module" (PyPI).
If you use 2.7.9, you will have pip installed, and pip install statistics within the 2.7 directory should install the module for 2.7 (I am not a pip or virtual machine expert, so might be slightly off.)
It comes pre-installed in python --Version 3. To import in python version 2 in Ubuntu, open Terminal and type
sudo pip install statistics
Enter your password and it will get installed.
Ps: you need to have pip already installed.
You need to be using python 3.4 to import statistics. Upgrade to the current version and you should have no problems.