jumbled boards and big word graphs#95
Merged
Merged
Conversation
The "X of N leaves valued so far" progress line used lat_len (the lattice's full combinatorial size, 5379616 for English) as N. Most of that lattice is unreachable: any subrack whose per-tile count exceeds that tile's total bag frequency (e.g. 7 Q's) has no valid draw and can never be valued. The true ceiling only counts subracks that pass the same per-tile check the globally_possible closure already applies per-rack elsewhere in this file (4114349 for English, verified by an independent DP this session, dp[7]=3199724 matching the known full-rack-drawable count). A real 4x256 run had already hit exactly 4114349 valued after generation 1 -- it had converged, but the old denominator made it look 24% incomplete. Fix computes the achievable count once at census startup (unrank every lattice index, keep those within base_freqs) and uses it in place of lat_len in the progress eprintln!. Does not touch any leave value or the accumulator arrays, which stay lat_len-sized. Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>
The single-position CLI only handled classic (spell-a-word) configs loaded from a 22-bit .kwg. Two axes were missing: - jumbled configs (any anagram of the letters counts as a word) need the config selected with a jumbled- prefix (super- for the 21x21 board, combining as jumbled-super-), and their word graph is an alphagram dawg (.kad), not a gaddawg. The word-prune step, which rebuilds only the words still playable on the board into a small graph for the solver, hardcoded a gaddawg -- so on a jumbled config it produced a graph the movegen could not read. It now branches on game_rules(): a gaddawg for classic, a dawg over make_alphagrams for jumbled, matching how the rest of movegen dispatches. Factored into a build_pruned_kwg helper shared with the batch path (classic, so byte-identical there). - word lists too large for a 22-bit node index (e.g. Dutch) are built as a 24-bit .kbwg. A -big suffix on the config name (jumbled-dutch-big) loads the source graph as Node24 instead of Node22; the pruned graph stays Node22 since the surviving word set is always tiny. Tests cover the config-name resolution (jumbled/super/big) and that build_pruned_kwg emits the right shape per game_rules (a word round- trips under its own config's lookup but not the other's). Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>
Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
this pr broadens the single-position endgame cli beyond classic spell-a-word configs to two previously unsupported axes: jumbled boards, where any anagram of the placed letters counts as a word, and word lists too large to index with 22 bits. jumbled configs are chosen with a jumbled- prefix (super- for the 21x21 board, combining as jumbled-super-) and use an alphagram dawg (.kad) rather than a gaddawg; the word-prune step that rebuilds only the still-playable words into a small graph for the solver used to hardcode a gaddawg, so on a jumbled config it produced a graph the movegen could not read. it now branches on game_rules() and builds the matching graph shape, factored into a build_pruned_kwg helper shared with the batch path. a -big suffix on the config name loads an oversized source graph as a 24-bit .kbwg while keeping the tiny pruned graph at 22 bits. the pr also carries an unrelated census fix so its progress line reports a reachable denominator, plus the endgame doc update.
Root cause
the census progress line "X of N leaves valued so far" used lat_len, the lattice's full combinatorial size (5379616 for English), as N. most of that lattice is unreachable: any subrack whose per-tile count exceeds that tile's bag frequency (e.g. 7 Q's) has no valid draw and can never be valued. a real 4x256 run had already converged at 4114349 valued after generation 1, but the inflated denominator made it look 27% incomplete.
Changes
Correctness
the census fix touches only the progress eprintln! denominator, not any leave value or accumulator array. the endgame refactor keeps the classic batch path byte-identical (build_pruned_kwg builds a gaddawg there as before). the achievable ceiling was verified by an independent DP this session, with dp[7]=3199724 matching the known full-rack-drawable count. tests cover config-name resolution (jumbled/super/big) and that build_pruned_kwg emits the right shape per game_rules -- a word round-trips under its own config's lookup but not the other's.