compute

@calab/compute

Worker pool, warm-start caching, kernel math, downsampling, and synthetic data generation for the CaLab monorepo.

Depends on @calab/core.

@calab/core
  ↑
@calab/compute
  ↑
apps/catune

Exports

Export	Source	Description
createWorkerPool, WorkerPool	worker-pool.ts	Generic worker pool with priority queue, cooperative cancellation via MessageChannel yields, and intermediate result callbacks
computePaddedWindow, computeSafeMargin	warm-start-cache.ts	Windowed computation helpers — compute padded solve region with safety margins for overlap-and-discard
shouldWarmStart, WarmStartCache	warm-start-cache.ts	3-tier warm-start cache (lambda-only change, kernel change with momentum reset, cold start)
computeKernel, computeKernelAnnotations	kernel-math.ts	Double-exponential kernel computation and annotation helpers for chart display
downsampleMinMax	downsample.ts	Min-max downsampling for efficient chart rendering of large traces
makeTimeAxis	time-axis.ts	Time axis generation from sample count and sampling rate
DEMO_PRESETS, DEFAULT_PRESET_ID, getPresetById, getPresetLabels	demo-presets.ts	6 built-in synthetic demo presets with known ground truth
generateSyntheticTrace, generateSyntheticDataset	mock-traces.ts	Synthetic calcium trace generation for demo data

Design Notes

• Worker factory injection — createWorkerPool(() => new Worker(...)) keeps the new Worker(new URL(..., import.meta.url)) pattern in the app so Vite can detect and bundle the worker.
• Raw postMessage (not Comlink) so the event loop can process cancel messages between solver batches.
• MessageChannel yields (<1 ms) instead of setTimeout(0) (~4 ms) for cooperative multitasking.