Tutorial 4: Python, Conda, and Jupyter

Objective: Set up the reproducible class Python environment on Anvil and launch a Jupyter Notebook.

1. Why use Conda?¶

On an HPC (High Performance Computer), you cannot simply pip install packages globally because you do not have “Admin” (root) privileges. If everyone installed different versions of NumPy or PyTorch into the system folders, everything would break.

Conda solves this by creating isolated “bubbles” (environments).

Isolation: You can have one bubble for this class (with numpy 1.26) and another for your research (with numpy 2.0) without conflict.
Reproducibility: Everyone in the class uses the exact same bubble, so code that runs for me will run for you.

2. Setting Up the Class Environment¶

To ensure compatibility, we have created a shared central environment on Anvil. You do not need to install anything yourself. You simply need to tell Anvil where to find our class “bubble.”

Step 1: Open a Terminal¶

Log in to Anvil OnDemand.
Click Clusters > Anvil Shell Access (or open a Terminal within an existing Jupyter session).

Step 2: Run the Setup Commands¶

Copy and paste the following commands into your terminal. You only need to run these once to set up your account.

# 1. Tell Anvil where to look for our class modules
module use /anvil/projects/x-chm250117/etc/modules

# 2. Load the specific class environment module
module load conda-env/molsimclass-py3.12.8

# 3. Register the environment as a Jupyter Kernel
conda-env-mod kernel -p /anvil/projects/x-chm250117/apps/molsimclass

What did I just do?

module use: You added our class’s private software folder to your path.
module load: You actually loaded the software environment.
conda-env-mod kernel: You created a shortcut (a “Kernel”) so the Jupyter web interface knows this Python environment exists.

3. Launching Jupyter with the Class Kernel¶

Now that you have registered the environment, you can use it in your notebooks.

Go to Anvil OnDemand.
Click Interactive Apps > Jupyter Notebook.
Settings:

Account: compute (or your specific allocation).
Partition: shared.
Number of Hours: 1 or 2.

Click Launch.
Once the session is ready, click Connect to Jupyter.

Selecting the Kernel¶

When your notebook opens, it might default to the standard “Python 3”. To use the class tools:

Open your notebook.
In the top right corner, click on the kernel name (e.g., “Python 3”).
In the dropdown menu, select: molsimclass (or Python [conda env: molsimclass]).

Verification: Run this Python code in a cell. It should point to the project folder, not your home folder:

import sys
print(sys.executable)
# Expected output: /anvil/projects/x-chm250117/apps/molsimclass/...

4. Creating Your Own Environment (Optional)¶

If you want to create a custom environment for your own separate research projects, we recommend using the system’s conda-env-mod tool. Unlike standard conda commands, this tool automatically generates a module file (so you can easily load it later) and a Jupyter kernel (so it appears in your notebook dropdowns).

Step 1: Load the Conda module You must load the system’s base conda module to access the creation script.

module load conda

Step 2: Create the environment Use the conda-env-mod script to create your environment. We recommend adding the --jupyter flag immediately so it sets up the kernel for you.

# Create an environment named 'my_md_env' with Jupyter support
conda-env-mod create -n my_md_env --jupyter

Follow the on-screen prompts. You may need to type y to confirm the creation.

Step 3: Load your new environment Once created, the script will print specific instructions on how to load your new environment. You must load this module to use it. It generally follows this pattern:

# 1. Tell the system where to look for your custom modules
module use $HOME/privatemodules

# 2. Load your specific environment module
# (Note: The actual version suffix 'py3.X.X' will be displayed in the output of Step 2)
module load conda-env/my_md_env-py3.8.8

Step 4: Install Additional Packages Once the module is loaded (Step 3), you are “inside” your environment. You can now use standard conda or pip commands to install any software you need.

Install via Conda (Recommended for scientific libraries):

# Install standard libraries
conda install numpy matplotlib

# Install MDAnalysis from the conda-forge channel
conda install -c conda-forge mdanalysis

Install via Pip (If not available on Conda):

pip install package_name

Note on Jupyter: Because you used the --jupyter flag in Step 2, you do not need to manually install ipykernel. Your environment will automatically appear as “Python (My my_md_env Kernel)” in the Jupyter interface.

Detailed Documentation: For more advanced usage, troubleshooting, or cluster-specific policies, please refer to the official Anvil User Guide: Installing Python Packages.