Jupyter Notebook Not Recognizing "import torch"

Question

To install PyTorch on Ubuntu, as instructed by the official website, I did pip3 install torch torchvision, and I am able to run PyTorch using the python3.5 command.

However, when I run Jupyter Notebook (I'm just running Jupyter Notebook in the terminal and using Chrome to access my notebooks), it doesn't recognize the package, throwing ModuleNotFoundError: No module named 'torch' at me.

The other odd thing is that PyTorch seems to have only been installed on Python 3.5 and not on Python 3.6 because:

➜  ~ python3.5 -c "import torch; print(torch.__version__)"     
0.4.1
➜  ~ python3.6 -c "import torch; print(torch.__version__)"     
Traceback (most recent call last):
  File "<string>", line 1, in <module>
ModuleNotFoundError: No module named 'torch'
➜  ~

Therefore I'm guessing that Jupyter Notebook is not using Python 3.5.

This is the result when I type python in my terminal and press TAB:

➜  ~ python
python             python2.7          python3.5          python3.5m         python3.6          python3.6m         python3-config     python3m-config  
python2            python3            python3.5-config   python3.5m-config  python3.6-config   python3.6m-config  python3m

Also, when I run which python, I get /home/mhy/anaconda3/bin/python since I installed Anaconda recently. I then activated my env and installed PyTorch using the following commands:

source activate my_env
conda install torch torchvision

But I wasnt able to run PyTorch on Anaconda either.

Answer 1

First, check whether you run your Jupyter notebook on the different environment than the one where PyTorch is installed:

(my_env) instance-1:~$ source activate my_env
(my_env) instance-1:~$ python
>>> import  sys
>>> print(sys.executable)
/home/instance-1/anaconda3/envs/my_env/bin/python

and then run the notebook in the my_env environment: (my_env) instance-1:~$ jupyter notebook And in the notebook execute print(sys.executable), then if you get something like ‘/home/instance-1/anaconda3/bin/python’, you have created the notebook in the environment Jupyter uses as a default.

I have been struggling with the same problem, once I discovered that in the Jupyter notebook console you have options to create notebooks in different environments. So just go to the 'New' section and select something like Python(myenv) to create a notebook in the desired environment(my_env in this case).

Another possible solution might be to install Jupyter into your environment: activate the desired environment and run conda install -c anaconda jupyter