Setting up the environment
Installing conda and creating a new environment.
If you already have anaconda/conda/miniconda/conda-forge, proceed to step 7.
Step 1. Download conda installer.
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -O ~/miniconda.sh
NOTE 1a: if this fails with “wget: command not found” install wget via “sudo apt install wget” (Ubuntu) or “brew install wget” (MacOS). Then try to do step 1 again.
NOTE 1b: alternatively, you can just follow the link above using your browser and download the file this way.
Step 2. Install miniconda.
bash ~/miniconda.sh
Step 3. Read and accept all licences. Step 4. When asked where to install conda, just press enter to install it in the default directory. Step 5. IMPORTANT: Then it will ask you “Do you wish the installer to initialize Miniconda3”. Type “yes” and press enter. Step 6. IMPORTANT: Restart your terminal session – open and close your terminal window.
NOTE: If you don’t want conda to start base envorinment in every terminal you open, you can execute conda config --set auto_activate_base false.
Step 7. Create a new environment nlp_class with python 3.7.
conda create --name nlp_class python=3.7
NOTE: If you have conda not found error, it might mean you skipped step 5 and now need to add conda to your PATH manually. It it easy to do, here’s a solution.
Step 8. Activate this environment.
conda activate nlp_class
Step 9. Install required libraries.
(Updated Jan 25, added ipywidgets that is required for datasets widgets to work inside jupyter lab.)
python -m pip install jupyterlab torch transformers datasets scikit-learn ipywidgets
Step 10. Install a new kernel to a jupyter lab.
python -m ipykernel install --name nlp_class --user
(Updated Jan 27, added --user option to the commend which resolves /usr/local/share permission denied issue some students had)
NOTE: If step 10 fails, ask Vlad for help.
Step 11. Launch jupyter lab.
jupyter lab
This should open a browser tab with Jupyter Lab running.
Step 12. On the right, you will see a couple of available notebook enviornments under “Notebook” title. Select “nlp_class” and create a new notebook.
Step 13. Make sure all imports work. To do this execute the following code.
import torch
import sklearn
import transformers
import datasets
If this fails, ask Vlad for help.
Now you should be good to go. If you are wondering why we need conda and environments, you can read this blogpost.