whestbench.

How-to — Recipes for specific tasks

Sourced from whest-starterkit @ aaa3882.

How-to — Recipes for specific tasks

← Documentation

Task-oriented guides. Each one answers "I want to do X — give me the steps and the gotchas." Use these alongside the Tutorial (the trail) and the Reference (exact APIs).

Writing and iterating

DocWhen to read
write-an-estimator.mdImplementing your custom estimator from scratch — minimal structure, contract checklist, common first failure.
inspect-mlp-structure.mdTraversing the MLP object: fields, weights, shapes.
validate-run-package.mdThe standard local loop: validaterunpackage. Includes a "Useful flags" table covering --seed, --n-samples, --profile, --max-threads, etc.
use-evaluation-datasets.mdPre-creating an evaluation dataset for fast, reproducible iteration.

Optimizing

DocWhen to read
algorithm-ideas.mdSurvey of estimation strategies — Monte Carlo, mean propagation, covariance, hybrid routing, plus open directions (low-rank, layer-adaptive, spectral, importance sampling, higher moments).
manage-flop-budget.mdWhere your FLOPs go and how to fit a tighter budget. Includes a line-by-line walkthrough of examples/02_mean_propagation.py.
performance-tips.mdConcrete patterns — matmul placement, free ops, diagonal vs covariance, env-var knobs.

Debugging and shipping

DocWhen to read
debugging-checklist.mdTiered checklist for "estimator runs but something feels wrong" — Tier 0 pure-Python loop, Tier 1 sanity, Tier 2 correctness, Tier 3 optimization.
pre-submission-checklist.mdOne-screen gate before you click "submit" on AIcrowd.

➡️ Where to look next

  • Need the exact contract / report fields? → Reference.
  • Estimator throws an error you don't recognize? → Troubleshooting.
  • Climbing the formality ladder one stage at a time? → Tutorial.

On this page