Week 21 — Coding Projects

Core

Empirically study convergence across many parallel trials.

• NumPy: Run repeated Monte Carlo experiments. Plot sample mean against N. Show confidence-like spread over many trials.
• Metal: Parallel generation of many trial estimates, aggregate on CPU for plotting. · Reading: MBT — batched compute workloads and reductions.
• Vulkan: Parallel multi-trial estimator with large-batch compute dispatch. · Reading: Vulkan Book — large-batch compute dispatch and result aggregation.
• CUDA: Embarrassingly parallel simulation of many trials. · Reading: CUDA Book — embarrassingly parallel simulation structure.
• Stretch: Compare bounded vs. heavy-tailed examples.
• Verify: Sample means stabilize with larger N · Variability across trials shrinks · Pathological distributions converge more awkwardly.