Working with Conda

Conda installs binary packages such as PyTorch, TensorFlow and NumPy with their compiled libraries, which is why many data science projects use it. To run notebooks in a conda environment, you create the environment, install ipykernel into it and register it as a Jupyter kernel.

Create a New Environment

conda create --name torchEnv

This creates an isolated environment named torchEnv. To pin a Python version, name it when you create the environment:

conda create --name torchEnv python=3.11

Activate the Environment

conda activate torchEnv

From now on, packages you install and code you run in this shell use torchEnv.

Install Required Packages

The ipykernel package is what lets the environment run as a Jupyter kernel:

conda install -c conda-forge ipykernel

You can install the other libraries your project needs here too, such as numpy, torch or pandas.

Register the Kernel

With ipykernel installed, register the environment so Jupyter and Zasper can offer it:

python -m ipykernel install --user --name=torchEnv

torchEnv now appears in Zasper’s kernel picker, and you can choose it for a notebook from there.

Registering is what makes the environment available wherever you open Zasper. Without it, Zasper still finds a conda environment that has ipykernel when you start Zasper from a shell with that environment activated, and your base Anaconda, Miniconda or Miniforge install when its Python has ipykernel. Those appear in the Launcher as Python 3.x (torchEnv) or Python 3.x (miniconda3) without a kernelspec. See Installing Jupyter Kernels.

A kernel installed inside an environment, such as conda install -c conda-forge r-irkernel, is found without registering it with --user and without activating the environment, because Zasper reads every environment conda knows about. It runs with that environment’s own R or other command.