Split Plotting Functionality into Optional Package Extra

[roberthjaplaio]

Description:

I'm using pymoo in a FastAPI application where visualization and plotting are not required. However, pymoo currently has a direct dependency on matplotlib, which significantly increases the application's size and install time, especially in serverless or containerized environments.

I would like to request splitting the plotting-related functionality into an optional package extra, such as pymoo[visualization]. This would allow users who need plotting to install it explicitly, while those who don't (like in my FastAPI application) can avoid the unnecessary matplotlib dependency.

Benefits:

• Reduced Application Size: Minimizes the footprint of pymoo in applications where plotting is not needed.
• Faster Installation: Improves installation speed by avoiding the matplotlib dependency.
• Cleaner Dependencies: Provides a more granular control over dependencies.
• Improved Compatibility: Can reduce dependency conflicts.

Proposed Changes:

1. Create a dedicated pymoo/visualization module/subpackage: Move all plotting-related code into this separate module/subpackage.
2. Define a visualization extra: In setup.py, define a visualization extra that includes matplotlib as a dependency.
3. Ensure no core modules import from pymoo/visualization: Verify that the core optimization modules do not import anything from the plotting module.
4. Documentation Update: Update the documentation to reflect the optional plotting extra and guide users on how to install it.

Example Usage (with the proposed change):

• Install with plotting:

  pip install pymoo[plotting]

• Install without plotting:

  pip install pymoo

Context:

My specific use case involves embedding pymoo into a FastAPI backend. In this context, generating plots is completely irrelevant, and the inclusion of matplotlib is a significant overhead.

I believe this change would make pymoo more versatile and efficient for a broader range of applications.

Would the pymoo maintainers consider implementing this feature?

Thank you for your time and consideration.

[blankjul]

Interesting point I have not thought about before.

Plotting is a little more interwoven in the framework, for instance there is PSOAnimation which helps to visualize PSO or the RunningMetric that visualizes convergence.

Just curious, when you use lambda functions are you trying to minimize the size of your docker image? Any reason why you could not simply install pymoo with --no-dependencies and control the dependencies yourself (and only install the ones you need)?

[roberthjaplaio]

We're actually not using lambda functions, but just an API running in a container orchestrator. We generally prefer leaner dependencies for several reasons, particularly when embedding libraries into a service:

• Lower memory footprint for containers.
• Reduced security surface area.
• Simplified dependency management and reduced risk of version conflicts.
• A generally cleaner and more focused codebase for developers.

Regarding a no-dependencies install. We are using uv and our pyproject.toml and therefore this isn't a straightforward option with our current setup (uv doesn't currently support this flag so it would need workarounds). This would also introduce a significant maintenance burden. We'd lose the safety net of automatic dependency management during pymoo updates, increasing the risk of runtime errors if we miss a new dependency required by a code path we use. This level of manual dependency tracking isn't something we'd ideally want to take on.

Ultimately, for backend integrations of pymoo within software solutions, especially APIs, the need for server-side plotting seems very niche. Visualization is typically handled on the frontend, often with JavaScript libraries. Plotting feels much more aligned with exploratory data analysis, research, and non-production use cases.

[blankjul]

I overall like the idea. Let me have a look at it.

matplotlib is imported in 12 files so it is a little bit of work. What do you think about the solution that all other files import an
__init__.py file which imports matplotlib or throws an exception to ask to install the "full" version of pymoo?

General, I would prefer if the (all analagous with uv)
pip install pymoo

command stays as is and to create an option for you ala

pip install pymoo[core]

for your use-case so to install only the most important dependencies. I did a little bit of research but think using the extra keyword can only add dependencies and not remove (or define the entire list). Do you know if this is possible or their is a workaround?

[blankjul]

can you check out the main branch with the commit above?

please reopen if you think that does not solve it!