Fix network visualization bug: Replace array indexing with dictionary lookup by codebyNJ · Pull Request #3045 · mesa/mesa

codebyNJ · 2025-12-30T16:06:51Z

Fixes #3023. @Sahil-Chhoker This is a new PR I have raised a cleaner version.

Problem: Code used node IDs as array indices instead of dictionary keys, causing IndexError that was silently suppressed.

Solution: Replace array indexing with dictionary lookup with proper error handling.

Before: Nodes [0,1,5,10,15] → 0/5 agents visible (all lost)
After: Nodes [0,1,5,10,15] → 5/5 agents visible

Tests: 6 comprehensive unit tests added covering:

Non-contiguous node IDs
Missing node positions with warnings
Large and negative node IDs
Edge cases (empty, single node, duplicates)

Files changed:

mesa/visualization/space_renderer.py - Fix implementation
tests/visualization/test_space_renderer.py - Unit tests

To Reproduce

import networkx as nx
from mesa import Agent, Model
from mesa.space import NetworkGrid
from mesa.visualization import SpaceRenderer
import numpy as np
import contextlib

# Create network with non-contiguous node IDs
G = nx.Graph()
G.add_nodes_from([0, 1, 5, 10, 15])
G.add_edges_from([(0, 1), (1, 5), (5, 10), (10, 15)])

model = Model()
space = NetworkGrid(G)
model.grid = space

for node_id in [0, 1, 5, 10, 15]:
    agent = Agent(model)
    space.place_agent(agent, node_id)

renderer = SpaceRenderer(model)

# Demonstrate the actual bug - what old code did
print("BEFORE FIX (old code with array indexing):")

pos_array = np.array([[0.1, 0.2], [0.3, 0.4], [0.5, 0.6], [0.7, 0.8], [0.9, 1.0]])
node_ids = np.array([[0], [1], [5], [10], [15]], dtype=float)

mapped_positions_old = []
for node_id in node_ids[:, 0].astype(int):
    with contextlib.suppress(IndexError):  # This is what the old code did!
        pos = pos_array[node_id]
        mapped_positions_old.append(pos)

print(f"Agents mapped: {len(mapped_positions_old)}/5")
print(f"Agents silently dropped: {5 - len(mapped_positions_old)}/5")
print(f"Which nodes were dropped: {[int(n) for n in node_ids[:, 0] if int(n) >= len(pos_array)]}")

# Demonstrate the fix - what new code does
print("\nAFTER FIX (new code with dictionary lookup):")

pos_dict = {
    0: np.array([0.1, 0.2]),
    1: np.array([0.3, 0.4]),
    5: np.array([0.5, 0.6]),
    10: np.array([0.7, 0.8]),
    15: np.array([0.9, 1.0])
}

mapped_positions_new = []
for node_id in node_ids[:, 0].astype(int):
    if node_id in pos_dict:  # Dictionary lookup - works for any node ID
        mapped_positions_new.append(pos_dict[node_id])

print(f"Agents mapped: {len(mapped_positions_new)}/5")
print(f"All agents visible: {len(mapped_positions_new) == 5}")

Explaination

Networks with non-contiguous node IDs (like [0, 1, 5, 10, 15] ) caused the old code to silently fail when using array indexing. By replacing array indexing with dictionary lookup, ensures all agents are properly visualized regardless of node ID values.

github-actions · 2025-12-30T16:14:39Z

Performance benchmarks:

Model	Size	Init time [95% CI]	Run time [95% CI]
BoltzmannWealth	small	🔵 +0.9% [-0.1%, +1.9%]	🔵 -0.0% [-0.1%, +0.1%]
BoltzmannWealth	large	🔵 +1.3% [-0.8%, +4.5%]	🔵 -3.3% [-7.1%, +0.5%]
Schelling	small	🔵 +0.1% [-0.9%, +1.0%]	🔵 -0.1% [-0.6%, +0.3%]
Schelling	large	🔵 +1.7% [+0.6%, +3.5%]	🔵 -0.1% [-0.9%, +0.7%]
WolfSheep	small	🔵 -0.4% [-1.3%, +0.3%]	🔵 -0.6% [-0.8%, -0.4%]
WolfSheep	large	🔵 -1.7% [-5.4%, +0.9%]	🔵 -0.2% [-1.0%, +0.8%]
BoidFlockers	small	🔵 +0.8% [+0.5%, +1.1%]	🔵 +1.8% [+1.6%, +2.0%]
BoidFlockers	large	🔵 +0.0% [-0.4%, +0.5%]	🔵 +0.7% [+0.3%, +1.0%]

Sahil-Chhoker · 2026-01-09T11:54:05Z

@codebyNJ Thanks for the PR. After reviewing the problem again, it appears we have two distinct issues: the Node ID mapping error (which you identified) and the layout synchronization race condition (identified by @codebreaker32).

While this PR fixes the mapping logic, we also need the graceful failure handling from @codebreaker32's approach to prevent crashes during frame delays. I’d like you two to collaborate on a hybrid solution in this PR. Ideally, we should combine the accuracy of dictionary lookups with NaN masking for missing nodes. Please update the tests to cover both scenarios efficiently.

for more information, see https://pre-commit.ci

codebyNJ · 2026-01-09T16:12:14Z

@Sahil-Chhoker

Hybrid fix combining approaches from #3045 and #3065

This PR addresses two distinct issues:

Non-contiguous node ID mapping - Using dictionary lookup instead of array indexing
Race condition resilience - Using NaN masking for missing nodes (from @codebreaker32's approach in #3065)
Missing agents are now hidden (NaN) rather than placed at origin, which is safer and prevents visual artifacts.

The below is the benchmarks after the hybrid fix

tpike3 · 2026-01-10T12:57:17Z

@Sahil-Chhoker I think the hybrid suggestion and collaboration is genius!

I will throw out a thought that anyone can feel free to dismiss, but in the hybrid vein, instead of a for loop (Line 171), what if you vectorized the operation (the mask approach of #3065)? I think this should further improve performance.

codebyNJ · 2026-01-10T14:49:50Z

@tpike3 I tried out your idea locally and also benchmarked by implementing vectorization using np.unique() with boolean masking.

If you want me to make the change I am ready to make it.

Vectorization Results

The vectorized approach (using np.unique() with boolean masking) is consistently faster:

Scenario	Loop	Vectorized	Improvement
Small networks (10 nodes)	112 µs	59 µs	-48%
Non-contiguous nodes	175 µs	62 µs	-65%
Large networks	787 µs	748 µs	-5%
Race conditions	153 µs	31 µs	-80%

Sahil-Chhoker · 2026-01-15T03:56:53Z

@codebyNJ thanks for benchmarking both approaches, can you implement the vectorized approach if it doesn't introduce the discrete node_id problem again.

codebyNJ · 2026-01-15T11:15:11Z

@Sahil-Chhoker I have updated with vectorization technique. It does not introduce discrete node_id issue again.

Sahil-Chhoker

Thanks @codebyNJ for this PR, this is approved from my side but it can take a look from @quaquel.

Sahil-Chhoker · 2026-01-16T12:01:08Z

mesa/visualization/space_renderer.py

+            # Warn if significant nodes missing (likely layout issue, not race condition)
+            if missing_nodes and len(missing_nodes) > len(pos_dict) / 10:
+                sample = missing_nodes[: min(5, len(missing_nodes))]
+                warnings.warn(
+                    f"Many nodes {sample}{'...' if len(missing_nodes) > 5 else ''} not found "
+                    f"in position layout ({len(missing_nodes)}/{len(node_ids)} agents). "
+                    f"This may indicate the network layout needs to be updated or regenerated.",
+                    UserWarning,
+                    stacklevel=2,
+                )


Did you run into such a problem while testing?

I didn't encounter this during testing of Issue #3023 specifically, but I added it as defensive handling since we're now using dictionary lookup. Without the warning, missing nodes would silently become NaN (invisible) with no indication to the user.

If you think this warning is unnecessary, I can remove it and just let missing nodes silently become NaN. Let me know your preference!

… lookup (mesa#3045)

quaquel added the bug Release notes label label Dec 30, 2025

EwoutH requested a review from Sahil-Chhoker January 9, 2026 10:17

Sahil-Chhoker mentioned this pull request Jan 9, 2026

Fix race around condition in space_renderer #3065

Closed

codebyNJ force-pushed the fix-network-visualization-bug branch from 79e3d2b to 2cee4db Compare January 9, 2026 15:24

Fix network visualization bugs: non-contiguous nodes and race conditions

23adae6

codebyNJ force-pushed the fix-network-visualization-bug branch from d88afd3 to 23adae6 Compare January 9, 2026 16:04

[pre-commit.ci] auto fixes from pre-commit.com hooks

b0ac0d3

for more information, see https://pre-commit.ci

quaquel mentioned this pull request Jan 12, 2026

Make cell.empty read only #3072

Open

codebyNJ added 2 commits January 15, 2026 16:29

perf: Vectorize network node position mapping using np.unique

82e0226

refactor: Clean up comments in network node mapping

5242c07

Merge branch 'main' into fix-network-visualization-bug

69eb5c3

Sahil-Chhoker approved these changes Jan 16, 2026

View reviewed changes

quaquel merged commit fb75535 into mesa:main Jan 17, 2026
14 checks passed

quaquel pushed a commit to quaquel/mesa that referenced this pull request Jan 18, 2026

Fix network visualization bug: Replace array indexing with dictionary…

503aac4

… lookup (mesa#3045)

codebreaker32 mentioned this pull request Jan 25, 2026

Race condition in SpaceRenderer._map_coordinates for NetworkGrid #3064

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Fix network visualization bug: Replace array indexing with dictionary lookup#3045

Fix network visualization bug: Replace array indexing with dictionary lookup#3045
quaquel merged 5 commits intomesa:mainfrom
codebyNJ:fix-network-visualization-bug

codebyNJ commented Dec 30, 2025 •

edited

Loading

Uh oh!

github-actions bot commented Dec 30, 2025

Uh oh!

Sahil-Chhoker commented Jan 9, 2026

Uh oh!

codebyNJ commented Jan 9, 2026 •

edited

Loading

Uh oh!

tpike3 commented Jan 10, 2026

Uh oh!

codebyNJ commented Jan 10, 2026

Uh oh!

Sahil-Chhoker commented Jan 15, 2026

Uh oh!

codebyNJ commented Jan 15, 2026

Uh oh!

Sahil-Chhoker left a comment

Uh oh!

Sahil-Chhoker Jan 16, 2026

Uh oh!

codebyNJ Jan 17, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Uh oh!

Conversation

codebyNJ commented Dec 30, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions bot commented Dec 30, 2025

Uh oh!

Sahil-Chhoker commented Jan 9, 2026

Uh oh!

codebyNJ commented Jan 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

tpike3 commented Jan 10, 2026

Uh oh!

codebyNJ commented Jan 10, 2026

Vectorization Results

Uh oh!

Sahil-Chhoker commented Jan 15, 2026

Uh oh!

codebyNJ commented Jan 15, 2026

Uh oh!

Sahil-Chhoker left a comment

Choose a reason for hiding this comment

Uh oh!

Sahil-Chhoker Jan 16, 2026

Choose a reason for hiding this comment

Uh oh!

codebyNJ Jan 17, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

codebyNJ commented Dec 30, 2025 •

edited

Loading

codebyNJ commented Jan 9, 2026 •

edited

Loading