diff options
| author | Sebastiano Tronto <sebastiano@tronto.net> | 2024-07-15 11:15:35 +0200 |
|---|---|---|
| committer | Sebastiano Tronto <sebastiano@tronto.net> | 2024-07-15 11:15:35 +0200 |
| commit | 3d060c348fdfff074a9b902d56f539664789d831 (patch) | |
| tree | 12705ad2896be47961fab55735ec68e13abad4bc /TODO.txt | |
| parent | 67c35b638c04ab7a860aac0c166936e9481633c9 (diff) | |
| download | nissy-core-3d060c348fdfff074a9b902d56f539664789d831.tar.gz nissy-core-3d060c348fdfff074a9b902d56f539664789d831.zip | |
Solved bug in cocsep generation
Diffstat (limited to 'TODO.txt')
| -rw-r--r-- | TODO.txt | 30 |
1 files changed, 9 insertions, 21 deletions
| @@ -1,30 +1,14 @@ | |||
| 1 | Bug in cocsepdata | 1 | Bug in esep table generation |
| 2 | - Add tests for ttrep | ||
| 3 | - check that ttrep indeed brings to representative | ||
| 4 | - Fix? | ||
| 5 | - Once fixed, fix other tests | ||
| 6 | - maybe add longtest, e.g. as a tool? | ||
| 7 | - Clean up mixed bfs fromdone / fromnew, use as new for h0k4 | ||
| 8 | - Re-do stats | 2 | - Re-do stats |
| 9 | 3 | - Add long-running test for h0k4 (maybe as a tool?) | |
| 10 | Check stats for all tables using H48stats solver | ||
| 11 | - try DFS for h0 solver | 4 | - try DFS for h0 solver |
| 12 | - compare results, the bfs method could be wrong | 5 | - use dfs for computing big table, save distance %3 until the last two steps, |
| 13 | - if faster: remove bfs | 6 | then clean the table and double loop over moves to fill the value |
| 14 | - if slower: why do I get different results with the new bfs? | 7 | - dfs for tables with h=1 to 10? |
| 15 | 8 | ||
| 16 | Bug in esep table generation | ||
| 17 | - Fails for UFRUFU, try command | 9 | - Fails for UFRUFU, try command |
| 18 | ./run solve -solver H48 -options "2;20" -n 1 -M 10 -cube \ | 10 | ./run solve -solver H48 -options "2;20" -n 1 -M 10 -cube \ |
| 19 | "$(./run frommoves -moves "UFRUFU")" | 11 | "$(./run frommoves -moves "UFRUFU")" |
| 20 | - Fundamental error in how tables are generated, each coordinate has too | ||
| 21 | many representative. I need to use the big table with the full coordinate | ||
| 22 | first (h=11, ~241 billion positions, 60Gb with k=2). From this the smaller | ||
| 23 | tables can be easily deduced. | ||
| 24 | - Investigate the possibility of computing smaller tables directly in some | ||
| 25 | other way, even if slow. | ||
| 26 | - use dfs for computing big table, save distance %3 until the last two steps, | ||
| 27 | then clean the table and double loop over moves to fill the value | ||
| 28 | 12 | ||
| 29 | Solver | 13 | Solver |
| 30 | - cleanup h48 solver | 14 | - cleanup h48 solver |
| @@ -118,6 +102,10 @@ switch. Here NISS may be useful. | |||
| 118 | * see if vcube's method to flip all corners is better | 102 | * see if vcube's method to flip all corners is better |
| 119 | * find a better way for computing the inverse? | 103 | * find a better way for computing the inverse? |
| 120 | * Transform with big table: make static cube actually static (how?) | 104 | * Transform with big table: make static cube actually static (how?) |
| 105 | * Use selfsim: in generating some tables, it is in thery possible to only check | ||
| 106 | the few transformations that give self-similarity instead of all 48. | ||
| 107 | The performance drop is almost insignificant, but I would like to figure out | ||
| 108 | the mistake I made previously. | ||
| 121 | 109 | ||
| 122 | ## Improvements and other things | 110 | ## Improvements and other things |
| 123 | 111 | ||