Tutorial 46: NeuroBench Benchmarking

Generate NeuroBench-compatible evaluation reports for standardized comparison with other neuromorphic frameworks.

1. Compute Metrics

import numpy as np
from sc_neurocore.benchmarks import compute_metrics

result = compute_metrics(
    predictions=np.array([0, 1, 2, 0, 1]),
    targets=np.array([0, 1, 2, 0, 0]),
    spike_counts=np.array([30, 25, 40, 20, 35]),
    weights=[np.random.randn(10, 8)],
    timesteps=16,
    task="mnist",
)
print(result.summary())

2. Export JSON

print(result.to_neurobench_json())

3. Available Tasks

from sc_neurocore.benchmarks import TASKS
for name, task in TASKS.items():
    print(f"{name}: {task.n_classes} classes, baseline {task.baseline_accuracy}")

Tasks: keyword spotting, DVS gesture, heartbeat anomaly, MNIST, SHD.

Further Reading

• NeuroBench
• API: Benchmarks