poker-flopsolve 0.1.0 → 0.3.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/README.md CHANGED
@@ -241,6 +241,18 @@ Measured on an Apple M2, single threaded:
241
241
  `defineSpot` reports the projected cost before anything is allocated, and refuses a spot
242
242
  over `maxBytes` with concrete suggestions. `maxCombos` caps range width.
243
243
 
244
+ ### WebAssembly engine
245
+
246
+ The traversal runs in WebAssembly (`assembly/engine.ts`): 32-bit vectors and storage,
247
+ SIMD on every contiguous row, card sums and showdown sweeps in 64-bit. Its storage lives
248
+ in its own memory and the rest of the solver reads it through views, so every query,
249
+ warm start and analysis works unchanged. With threads (Node), all of them share one
250
+ WebAssembly memory and split the runouts as below. It runs wherever it can — 32-bit
251
+ storage, the default, and WebAssembly SIMD, which current Node and browsers have —
252
+ and `run.engine` says which ran. `engine: 'js'` forces the JavaScript traversal;
253
+ `precision: 'f64'` or `'i16'` use it too. Against the JavaScript traversal it is about
254
+ 2.4× faster per iteration, and about 3× in Chromium.
255
+
244
256
  ### SIMD kernels
245
257
 
246
258
  Regret matching and the average-strategy pass — the per-hand normalisation at every
@@ -320,8 +332,10 @@ deterministic at any thread count.
320
332
 
321
333
  ### Algorithm and benchmark
322
334
 
323
- The default update rule is Discounted CFR (`algorithm: 'dcfr'`, α 1.5, β 0, γ 2). CFR+
324
- is still available as `algorithm: 'cfr+'`. `npm run bench` times a fixed river, turn, flop
335
+ The default update rule is Discounted CFR (`algorithm: 'dcfr'`, α 1.5, β 0, γ 3, with
336
+ positive regret counted from the second iteration so the uniform first iteration is
337
+ forgotten). CFR+ is still available as `algorithm: 'cfr+'`. The table below predates the
338
+ 0.2.0 schedule, which needs 10–15% fewer iterations on flop and turn spots. `npm run bench` times a fixed river, turn, flop
325
339
  and two-tone flop. Measured in a 4-core cloud container, seconds of solving to reach each
326
340
  target (exploitability checks excluded):
327
341
 
@@ -336,6 +350,26 @@ The solver is compute-bound rather than memory-bound — a 150 MB flop streams i
336
350
  accumulators at about 0.25 GB/s — so the cost of an iteration tracks the work in it, not
337
351
  its bytes.
338
352
 
353
+ ### Compared with postflop-solver
354
+
355
+ `npm run bench:compare` runs the same spots through this solver and
356
+ [b-inary/postflop-solver](https://github.com/b-inary/postflop-solver) (Rust, built
357
+ natively for the machine) at the same 32-bit precision, thread count and checking
358
+ schedule. It needs a Rust toolchain; postflop-solver is fetched at build time and never
359
+ bundled. Measured in a 4-core cloud container, time to 0.3% of pot:
360
+
361
+ | Spot | Threads | poker-flopsolve | postflop-solver | Ratio |
362
+ |---|---|---|---|---|
363
+ | Turn | 1 | 1.4 s | 0.9 s | 1.7× |
364
+ | Flop | 1 | 29.4 s | 18.5 s | 1.6× |
365
+ | Monotone flop | 1 | 9.5 s | 5.2 s | 1.8× |
366
+ | Turn | 4 | 0.5 s | 0.3 s | 2.0× |
367
+ | Flop | 4 | 9.1 s | 5.1 s | 1.8× |
368
+ | Monotone flop | 4 | 2.9 s | 1.4 s | 2.1× |
369
+
370
+ 0.2.0, still in JavaScript, was 4–6× behind; the WebAssembly engine below is most of
371
+ the difference.
372
+
339
373
  ### Accuracy targets
340
374
 
341
375
  `solve(spot, { targetPercent })` stops as soon as exploitability reaches that percentage of
@@ -433,10 +467,9 @@ type-checked, tested and built; if the `version` in `package.json` is not on npm
433
467
  is published, with provenance. So a release is: bump `version`, merge to `main`. A push
434
468
  that leaves the version alone publishes nothing.
435
469
 
436
- It needs one repository secret, `NPM_TOKEN`: an npm granular access token with read and
437
- write access to `poker-flopsolve` (for the very first publish, before the package exists,
438
- one with permission to publish new packages). Without it the workflow still tests and
439
- builds, and says it skipped publishing.
470
+ It publishes by npm trusted publishing: npm trusts this repository's `publish.yml`
471
+ (set up under the package's settings on npmjs.com), the job proves who it is with GitHub's
472
+ OpenID Connect token, and no npm token is stored anywhere.
440
473
 
441
474
  ## License
442
475