unsolved
Member
OO_4x4x4 Version 1.2.8 is available
Omnia Obtorquebantur Version 1.2.8 With Leaf Node + Massive Tree Branch Pruning Without Parallel Search
http://lightningcloudcomputing.com/OO_4x4x4.zip
I placed the new version online. There is a known bug where it will miss an optimal line since it over-pruned one critical position too early. Other than that, it is very fast AFTER DEPTH 6 (explained below).
Features:
This new version, while still with an embedded bug, will prove very valuable should enough of the cubing population test it and submit the solves it does find. After examining enough of them, I will be able to track it down and make the algorithm water-tight. Some of you will notice that during the pass of either depth-5 or depth-6, the search seems to slow down. It is because it is performing a massive RAM-based sort of data that it is collecting as it gathers Branch Pruning Data. What it is doing is a little lengthy to explain, but the short version is the program is trying to determine what branches of play, not what moves, are getting nowhere fast. If subsequent branches (varying moves in length) show up in later searches, a quick poll against the TFS databases will give a pass/fail regarding taking that branch to the next level of search. It's pretty complicated, but once I get this nailed down, OO_4x4x4 will be able to search with blinding speed.
Depth 7- 8- and 9- will fly by at much faster than expected speeds. Don't forget, if you are probing the 5-TFS database, you are really searching up to ply 14 .
The single dedge flip, by luck or just plain matching all the specs of the algorithm, is able to be solved in just under 15 minutes on a "normal" 3.1 GHz Westmere. Some of you will have even faster systems, so please test it out.
Features Still To Come:
1. Fix that bug!
2. Fixed-Core Multiprocessing for 2-, 4-, 8-, 10-, 12-, and 16-core implementations.
3. Massively Parallel Bitboard version for those with >= 20 cores
4. Solving the 7-TFS database on my 192 GB machine at work
As always, any feedback is appreciated.
Pretty cool with all those 15 move dedge flips, but as for the display wouldn't having two characters instead of four between make the squares look more, well square? And also just so you know the de facto standard is having green on front and white on top.
Omnia Obtorquebantur Version 1.2.8 With Leaf Node + Massive Tree Branch Pruning Without Parallel Search
http://lightningcloudcomputing.com/OO_4x4x4.zip
I placed the new version online. There is a known bug where it will miss an optimal line since it over-pruned one critical position too early. Other than that, it is very fast AFTER DEPTH 6 (explained below).
Features:
- New Pruning Data Collector + Branch Pruning Algorithm
- Bi-Directional Searching from the scramble to the TFS databases, and the databases back towards the current position being solved
- Leaf Nodes are pruned to prevent duplicate positions, and Branches are pruned (entire lines of play, reducing the tree tremendously) based on a new experimental algorithm
- New graphics, as requested, shows Top over Front, and Left-Front-Right-Back, then Bottom under Front
This new version, while still with an embedded bug, will prove very valuable should enough of the cubing population test it and submit the solves it does find. After examining enough of them, I will be able to track it down and make the algorithm water-tight. Some of you will notice that during the pass of either depth-5 or depth-6, the search seems to slow down. It is because it is performing a massive RAM-based sort of data that it is collecting as it gathers Branch Pruning Data. What it is doing is a little lengthy to explain, but the short version is the program is trying to determine what branches of play, not what moves, are getting nowhere fast. If subsequent branches (varying moves in length) show up in later searches, a quick poll against the TFS databases will give a pass/fail regarding taking that branch to the next level of search. It's pretty complicated, but once I get this nailed down, OO_4x4x4 will be able to search with blinding speed.
Depth 7- 8- and 9- will fly by at much faster than expected speeds. Don't forget, if you are probing the 5-TFS database, you are really searching up to ply 14 .
The single dedge flip, by luck or just plain matching all the specs of the algorithm, is able to be solved in just under 15 minutes on a "normal" 3.1 GHz Westmere. Some of you will have even faster systems, so please test it out.
Features Still To Come:
1. Fix that bug!
2. Fixed-Core Multiprocessing for 2-, 4-, 8-, 10-, 12-, and 16-core implementations.
3. Massively Parallel Bitboard version for those with >= 20 cores
4. Solving the 7-TFS database on my 192 GB machine at work
As always, any feedback is appreciated.
Last edited: