FastStatisticalModels4Python

Current MacBook Air evidence set

This is the curated local MacBook Air result set for the PyCon talk. Use this directory instead of older quick-validation outputs.

Scope

The MacBook Air tier is for correctness, debugging, reference checks, and lightweight evidence. It should not be presented as final large-scale speed evidence or compared directly with Linux server CPU/A100 results.

Files

Core full-grid validation:

• kmeans_correctness.csv: full required k-means grid, 1,440 scenarios covered.
• permutation_equivalence.csv: full required permutation equivalence grid.
• permutation_calibration.csv: original full-grid null calibration rows.
• permutation_power.csv: original full-grid power rows.

Extra evidence added after visual review:

• kmeans_shape_stress.csv: targeted K=20/50, d=100, N=50,000, and selected N=100,000 k-means shape stress.
• permutation_calibration_extended.csv: 100 extra null calibration replicates.
• permutation_power_extended.csv: 168 rows over denser delta and signal-fraction settings.
• permutation_runtime_scaling_extended.csv: p/R runtime scaling with explicit memory-risk skips.
• extra_evidence_summary.json: row and status counts for the extra evidence.

Figure metadata:

• figure_manifest.csv: canonical list of primary figures and their source CSVs.
• figures/: deck-ready figures generated by experiments.visualization.plot_macbook_air_evidence.

Current status

• k-means full grid: 3,840 pass rows and 480 expected skipped_memory_risk rows for unsafe reference broadcast cases.
• k-means extra shape stress: 1,164 pass rows.
• permutation full equivalence: 450 pass rows and 45 expected skipped_memory_risk rows.
• permutation extra evidence: 100 calibration pass rows, 168 power pass rows, 108 runtime pass rows, and 27 expected runtime memory-risk skips.
• No fail rows are present in the curated latest CSV files.

Primary figures

• figures/kmeans_recovery_scenario_facets.png
• figures/kmeans_shape_stress_runtime.png
• figures/kmeans_runtime_recovery_tradeoff.png
• figures/kmeans_reference_equivalence.png
• figures/permutation_calibration_extended.png
• figures/permutation_power_extended.png
• figures/permutation_runtime_scaling_extended.png
• figures/permutation_equivalence_detail.png

Reproduction

Use the project default local environment:

python -m experiments.run_macbook_evidence_extra \
  --output-dir experiments/results/macbook_air_long/latest \
  --checkpoint-every 20 --max-iter 15

python -m experiments.visualization.plot_macbook_air_evidence \
  --results-dir experiments/results/macbook_air_long/latest

The extra evidence runner appends rows and skips existing run_ids, so it can be resumed after interruption.

This site is open source. Improve this page.