mesa-frames 0.1.0-alpha

[adamamer20]

As we near the end of Google Summer of Code, here's a prioritized list of tasks that need to be completed before the deadline:

1. GridPolars Implementation:

   • Initially, I implemented grid capacity using a list of pl.ranges with the idea of leveraging lazy evaluation for better performance. However, this approach is only advantageous when capacity is consistent across all cells. Reimplementing grid behavior with Polars expressions would require significant effort without a clear performance gain. Therefore, for now, I'd stick with NumPy and update the mixin to PolarsMixin as needed.

2. Example Models:

   • We need to provide at least two complete example models, such as the Sugarscape and Schelling models, to demonstrate how to build models using mesa-frames.

3. Documentation:

   • This includes an introductory tutorial, performance tips, FAQ, and the deployment of API documentation.

4. Automatic Build and Publishing:

   • Set up an automatic build and publishing pipeline for the package.

5. GPU Integration (Time Permitting):

   • This should be relatively straightforward, as it primarily involves switching from pandas to cudf and from numpy to cupy.

6. General Review and Bug Discovery:

   • Conduct a thorough review and address any found bugs.

I'd focus on these priorities and leave any additional features or improvements for after the GSoC period.

[rht]

Can't wait for the Sugarscape one. This is the code for FLAME 2 GPU Sugarscape. It is going to be much simpler in mesa-frames. We could write a paper for this later.

[adamamer20]

Do you really think it could be published as a paper? That would be awesome!
Also, do you think the sugarscape example should implement instantaneous growback (like FLAME2) or constant growback (like mesa-examples)?

[rht]

Yes, SamPy was very specific on epidemiology. There was https://github.com/VU-IVM/honeybees, but they used NumPy/CuPy. So mesa-frames is still the first to use DF for ABM.
I think it's fine to use instantaneous growback so that it is an apple vs apple benchmark with FLAME 2.