Installation

The PCMDI Metrics Package (PMP) is available for Linux-64 and macOS-64. Modern PMP releases generally support both Intel and Apple Silicon systems natively. Windows support has not been thoroughly tested.

You can find the package details on conda-forge.

PMP can be installed either with a conda-based package manager such as mamba or conda, or by using Docker.

Conda or Mamba Installation

Prerequisites

For a native PMP installation, you must have a conda distribution installed. We recommend one of the following:

Install the Anaconda package (we recommend installing this for each user)
Alternatives include Miniconda or Miniforge/Mambaforge
If using Anaconda or Miniconda, we recommend also installing mamba for better performance

Ensure your conda distribution is in your PATH. For example, if Anaconda is installed in ${HOME}/anaconda:

# For bash/zsh
export PATH="${HOME}/anaconda/bin:${PATH}"

# For tcsh
setenv PATH "${HOME}/anaconda/bin:${PATH}"

Standard Installation

We recommend creating a fresh virtual environment for PMP to avoid dependency conflicts.

Using Mamba (Recommended)

mamba create -n pmp_env -c conda-forge pcmdi_metrics
conda activate pmp_env

Using Conda

conda create -n pmp_env -c conda-forge pcmdi_metrics
conda activate pmp_env

To speed up installation with conda, you can explicitly specify the Python and PMP versions.

conda create -n pmp_env -c conda-forge python=3.10 pcmdi_metrics=4.2.1
conda activate pmp_env

Installation in Existing Environments

mamba install -c conda-forge pcmdi_metrics
# OR
conda install -c conda-forge pcmdi_metrics

To learn more about conda environments see: http://conda.pydata.org/docs/using/envs.html

Docker Installation

Docker packages PMP together with its runtime environment in a container image. This is useful when you want a reproducible installation without managing the Python and system dependencies directly on your host machine.

Prerequisites

Docker installation requires Docker Desktop or another Docker-compatible runtime.

Why use Docker?

Reproducible PMP environment across different systems
Reduced dependency conflicts on the host machine
Easier deployment on Linux servers, cloud systems, and shared platforms

In the examples below, pmp:latest is just a local image tag name.

Build Locally

Build the PMP Docker image locally from the repository root:

docker build -t pmp:latest .

Run the default container command to confirm PMP is installed:

docker run --rm pmp:latest

Start an interactive shell in the container:

docker run --rm -it pmp:latest bash

Mount a host directory so the container can access your input data or write output files:

docker run --rm -it -v /path/to/local/data:/data pmp:latest bash

Inside the container, PMP is installed in the preconfigured conda environment and the command-line tools are available on PATH.

Use a Published Image

If PMP is published on Docker Hub, users will be able to pull a prebuilt image instead of building locally.

Current Docker Hub image location:

docker.io/lee1043/pcmdi_metrics

Example pull command:

docker pull lee1043/pcmdi_metrics:latest

Note

The preferred long-term image location would be docker.io/pcmdi/pcmdi_metrics. For now, the published image is available from docker.io/lee1043/pcmdi_metrics.

Advanced & Legacy Installation

Support for Apple Silicon (M1/M2/M3)

Modern versions of PMP (v3.10+) generally support ARM64 architecture natively. However, if you are installing PMP v3.8 or lower, you may encounter issues due to legacy dependencies (like older CDAT components).

If native installation fails, you can force an Intel-based (x86) environment:

# Create the environment as an x86 architecture
CONDA_SUBDIR=osx-64 conda create -n pmp_legacy -c conda-forge python=3.8

# Lock the architecture for this environment
conda activate pmp_legacy
conda config --env --set subdir osx-64

# Install the legacy PMP version
mamba install -c conda-forge pcmdi_metrics=3.8.0

Troubleshooting

Environment Cleanup

Warning

Ensure no environment variables from old UV-CDAT installations (e.g., PATH, PYTHONPATH, LD_LIBRARY_PATH) are set, as these can cause runtime conflicts.

Firewalls and SSL

If your institution has strict SSL certificate inspection, you can bypass verification (at your own risk):

conda config --set ssl_verify False