Fix: Property layer visualization for HexGrid

[Sahil-Chhoker]

Summary

Property layers were not properly visualized for HexGrids, and this PR aims to resolve that issue. It utilizes the same coordinate system as the draw_hex_grid function and generates a PolyCollection that overlays color values on the grid.

Bug / Issue

fixes #2433

Before

After

Additional Notes:

I am not entirely sure if this was the best approach to address the issue. If there is a more effective or elegant solution, I would greatly appreciate your recommendations.

[github-actions]

Performance benchmarks:

Model	Size	Init time [95% CI]	Run time [95% CI]
BoltzmannWealth	small	🔵 -2.3% [-3.4%, -1.2%]	🔵 -1.8% [-2.3%, -1.3%]
BoltzmannWealth	large	🔵 -0.0% [-0.5%, +0.5%]	🔵 -0.3% [-0.8%, +0.4%]
Schelling	small	🔵 -0.6% [-1.0%, -0.2%]	🔵 -0.3% [-0.4%, -0.2%]
Schelling	large	🔵 -0.0% [-0.6%, +0.4%]	🔵 +2.9% [+1.6%, +4.1%]
WolfSheep	small	🔵 +0.0% [-0.4%, +0.5%]	🔵 +0.6% [+0.5%, +0.8%]
WolfSheep	large	🔵 +1.0% [+0.5%, +1.5%]	🔵 +2.3% [+0.5%, +4.4%]
BoidFlockers	small	🔵 -1.1% [-1.9%, -0.3%]	🔵 -0.1% [-0.4%, +0.1%]
BoidFlockers	large	🔵 +0.7% [+0.2%, +1.2%]	🔵 -0.1% [-0.5%, +0.2%]

[quaquel]

Thanks for this. The example looks great. I am taking a look at the code now.

[quaquel]

Just a few questions and ideas. Looks very good already.

[Sahil-Chhoker]

Are you sure it's fine as it is? I had a few questions about it:

1. Regarding the repetitive part, I think I’ve got an answer for now. I’ll get back to you as early as tomorrow.
2. I'm hardcoding the value of zorder to -1. Is this the correct approach?

[quaquel]

I gave it a first quick pass and focussed on the high-level stuff. The zorder one is a good question (and actually interacts with #2642). Ideally (but beyond this pr), zorder should become controllable for property layers to explicitly control layering.

[Sahil-Chhoker]

Thank you for the review! I have added the suggested changes. Please let me know if you have any additional recommendations to further improve the code quality.

[quaquel]

looks good, just a few remaining clarifications

[quaquel]

I think this is ready to be merged. @Sahil-Chhoker do you have any outstanding doubts? If not, I'll merge this later today.

[Sahil-Chhoker]

Thanks @quaquel, I think I'm fine.

[Sahil-Chhoker]

@quaquel, I ran into a bug while testing this feature again with the lru_cache functionality.
The function call hexagons = _get_hexmesh(width, height) won't display the grid. I don't know exactly what might be causing this, but it can be fixed by either of these two options:

1. hexagons = _get_hexmesh(width, height, size=1.0)
2. Removing lru_cache.

I am very sorry for pushing code without properly testing it.

[quaquel]

can you share a minimum example with the bug? I might be able to help diagnose it.

[Sahil-Chhoker]

Sorry for the late response, here is my example I was testing the code with:

from mesa.experimental.cell_space.grid import HexGrid
from mesa import Model
import math
from mesa.experimental.cell_space import CellAgent
from mesa.visualization import SolaraViz, make_space_component
from mesa.experimental.cell_space.property_layer import PropertyLayer
import numpy as np

class myAgent(CellAgent):
    def __init__(self, model, cell=None):
        super().__init__(model)
        self.cell = cell
    
 
class myModel(Model): 
    def __init__(self, seed=None, width=10,  height=15 , n_agents=4):
        super().__init__(seed=seed) 
        self.width = width
        self.height = height 
        self.num_agents = n_agents
        self.grid = HexGrid( 
            (self.width, self.height), capacity=math.inf, torus=False, random=self.random
        )
 
        self.test_layer = PropertyLayer("test", (self.width, self.height), default_value=1, dtype=int)
        self.test_layer.data = np.random.randint(2, 10, (self.width, self.height))
        self.grid.add_property_layer(self.test_layer)
        myAgent(self, self.grid.all_cells.cells[17])


def portrayal(agent):
    if agent is None:
        return None
    
    return {
        "color": "tab:orange",
        "size": 70,
    }


def post_process(ax):
    ax.set_aspect("equal")
    ax.set_xticks([])
    ax.set_yticks([])
    ax.get_figure().set_size_inches(8, 8) 

property_layer_portrayal = { 
        "test": { 
            "colormap": "Blues",
            "alpha": 0.7,
            "vmin": 0, 
            "vmax": 10,
        }
    }


model_params = { 
    "seed": {
        "type": "InputText",
        "value": 42,
        "label": "Random Seed",
    },
    "n_agents": {
        "type": "SliderInt",
        "value": 4,
        "min": 1,
        "max": 10,
        "step": 1,
        "label": "Number of Agents",
    },
    "width": 6,
    "height": 5,
}

model = myModel()

space_component = make_space_component(
    agent_portrayal=portrayal, draw_grid=True, post_process=post_process, propertylayer_portrayal=property_layer_portrayal
)

page = SolaraViz(
    model,
    components=[space_component],
    model_params=model_params,
    name="Test Model",
)
page

[Sahil-Chhoker]

@quaquel, While testing the code mentioned above, I discovered another bug: whenever the model's step count is increased, the hex grid simply disappears. After diagnosing the issue, I found that lru_cache was once again the culprit. I'm not sure why it’s causing so many problems, but I would recommend removing it from the code entirely.

[quaquel]

What happens if you use a normal cache instant of lru_cache?

[Sahil-Chhoker]

What happens if you use a normal cache instant of lru_cache?

cache behaves like lru_cache, but it stores results infinitely, unlike lru_cache, which deletes the least recently used values based on its max_size. Ultimately, what I mean is that the results were the same and buggy.

[quaquel]

I think I know what's going on. _get_hexmesh is a generator rather than a function. Why did you make it a generator?

[Sahil-Chhoker]

I thought it was cleaner than storing results in a list and returning them. Function does work as intended. Should I change it into a function instead?

[quaquel]

It depends on what we prefer. In case of a normal function. we can memoize the return allowing us to effectively generate the hexmesh vertices only once.

[Sahil-Chhoker]

Now that I have this knowledge that generators don't go well with caching (thanks to you), I think we should change this to a function.

[quaquel]

Can you put in a new PR which does this? Otherwise, I'll pick up tomorrow.