/**
* This function will generate getRandomTurnCount() number of non cancelling,
* random turns. If a puzzle wants to provide custom scrambles
* (for example: Pochmann style megaminx or MRSS), it should override this method.
*
* NOTE: It is assumed that this method is thread safe! That means that if you're
* overriding this method and you don't know what you're doing,
* use the synchronized keyword when implementing this method:<br>
* <code>protected synchronized String generateScramble(Random r);</code>
* @param r An instance of Random
* @return A PuzzleStateAndGenerator that contains a scramble string, and the
* state achieved by applying that scramble.
*/
public virtual PuzzleStateAndGenerator GenerateRandomMoves(Random r)
{
var ab = new AlgorithmBuilder(MergingMode.NoMerging, GetSolvedState());
while (ab.GetTotalCost() < GetRandomMoveCount())
{
var successors = ab.GetState().GetScrambleSuccessors();
try
{
string move;
do
{
move = Functions.Choose(r, successors.Keys);
// If this move happens to be redundant, there is no
// reason to select this move again in vain.
successors.Remove(move);
} while (ab.IsRedundant(move));
ab.AppendMove(move);
}
catch (InvalidMoveException e)
{
Assert(false, e.Message, e);
return(null);
}
}
return(ab.GetStateAndGenerator());
}
public static PuzzleStateAndGenerator ApplyOrientation(CubePuzzle puzzle, CubeMove[] randomOrientation,
PuzzleStateAndGenerator psag, bool discardRedundantMoves)
{
if (randomOrientation.Length == 0)
{
return(psag);
}
// Append reorientation to scramble.
try
{
var ab = new AlgorithmBuilder(MergingMode.NoMerging, puzzle.GetSolvedState());
ab.AppendAlgorithm(psag.Generator);
foreach (var cm in randomOrientation)
{
ab.AppendMove(cm.ToString());
}
psag = ab.GetStateAndGenerator();
return(psag);
}
catch (InvalidMoveException e)
{
Assert(false, e.Message, e);
return(null);
}
}
public override PuzzleStateAndGenerator GenerateRandomMoves(Random r)
{
var scramble = _threePhaseSearcher.RandomState(r);
var ab = new AlgorithmBuilder(MergingMode.CanonicalizeMoves, GetSolvedState());
try
{
ab.AppendAlgorithm(scramble);
}
catch (InvalidMoveException e)
{
Assert(false, e.Message, new InvalidScrambleException(scramble, e));
}
return(ab.GetStateAndGenerator());
}
public PuzzleStateAndGenerator GenerateRandomMoves(Random r, string firstAxisRestriction,
string lastAxisRestriction)
{
var randomState = Tools.RandomCube(r);
var scramble =
_twoPhaseSearcher.Solution(randomState, ThreeByThreeMaxScrambleLength, ThreeByThreeTimeout,
ThreeByThreeTimemin, Search.INVERSE_SOLUTION, firstAxisRestriction, lastAxisRestriction).Trim();
var ab = new AlgorithmBuilder(MergingMode.CanonicalizeMoves, GetSolvedState());
try
{
ab.AppendAlgorithm(scramble);
}
catch (InvalidMoveException e)
{
Assert(false, e.Message, new InvalidScrambleException(scramble, e));
}
return(ab.GetStateAndGenerator());
}
public static PuzzleStateAndGenerator ApplyOrientation(CubePuzzle puzzle, CubeMove[] randomOrientation,
PuzzleStateAndGenerator psag, bool discardRedundantMoves)
{
if (randomOrientation.Length == 0)
{
return(psag);
}
// Append reorientation to scramble.
try
{
var ab = new AlgorithmBuilder(MergingMode.NoMerging, puzzle.GetSolvedState());
ab.AppendAlgorithm(psag.Generator);
// Check if our reorientation is going to cancel with the last
// turn of our scramble. If it does, then we just discard
// that last turn of our scramble. This ensures we have a scramble
// with no redundant turns, and I can't see how it could hurt the
// quality of our scrambles to do this.
var firstReorientMove = randomOrientation[0].ToString();
while (ab.IsRedundant(firstReorientMove))
{
//azzert(discardRedundantMoves);
var im = ab.FindBestIndexForMove(firstReorientMove, MergingMode.CanonicalizeMoves);
ab.PopMove(im.Index);
}
foreach (var cm in randomOrientation)
{
ab.AppendMove(cm.ToString());
}
psag = ab.GetStateAndGenerator();
return(psag);
}
catch (InvalidMoveException e)
{
Assert(false, e.Message, e);
return(null);
}
}
public override PuzzleStateAndGenerator GenerateRandomMoves(Random r)
{
// For fewest moves, we want to minimize the probability that the
// scramble has useful "stuff" in it. The problem with conventional
// Kociemba 2 phase solutions is that there's a pretty obvious
// orientation step, which competitors might intentionally (or even
// accidentally!) use in their solution. To lower the probability of this happening,
// we intentionally generate dumbed down solutions.
// We eventually decided to go with "Tom2", which is described by Tom
// Rokicki (https://groups.google.com/d/msg/wca-admin/vVnuhk92hqg/P5oaJJQjDQAJ):
// START TOM MESSAGE
// If we're going this direction, and the exact length isn't critical, why not combine a bunch of the
// ideas we've seen so far into something this simple:
//
// 1. Fix a set of prefix/suffix moves, say, U F R / U F R.
// 2. For a given position p, find *any* solution where that solution prefixed and suffixed with
// the appropriate moves is canonical.
//
// That's it. So we generate a random position, then find any two-phase solution g such
// that U F R g U F R is a canonical sequence, and that's our FMC scramble.
//
// The prefix/suffix will be easily recognizable and memorable and ideally finger-trick
// friendly (if it matters).
//
// Someone wanting to practice speed FMC (hehe) can make up their own scrambles just
// by manually doing the prefix/suffix thing on any existing scrambler.
//
// It does not perturb the uniformity of the random position selection.
//
// It's simple enough that it is less likely to suffer from subtle defects due to the
// perturbation of the two-phase search (unlikely those these may be).
//
// And even if the two-phase solution is short, adding U F R to the front and back makes it
// no longer be unusually short (with high probability).
// END TOM MESSAGE
// Michael Young suggested using R' U' F as our padding (https://groups.google.com/d/msg/wca-admin/vVnuhk92hqg/EzQfG_vPBgAJ):
// START MICHAEL MESSSAGE
// I think that something more like R' U' F (some sequence that
// involves a quarter-turn on all three "axes") is better because it
// guarantees at least one bad edge in every orientation; with EO-first
// options becoming more popular, "guaranteeing" that solving EO
// optimally can't be derived from the scramble is a nice thing to
// have, I think. (Someone who was both unscrupulous and lucky could
// see that R' F R doesn't affect U2/D2 EO, and therefore find out how
// to solve EO by looking at the solution ignoring the R' F R. That
// being said, it still does change things, and I like how
// finger-tricky/reversible the current prefix is.) Just my two cents,
// 'tho.
// END MICHAEL MESSSAGE
var scramblePrefix = SplitAlgorithm("R' U' F");
var scrambleSuffix = SplitAlgorithm("R' U' F");
// super.generateRandomMoves(...) will pick a random state S and find a solution:
// solution = sol_0, sol_1, ..., sol_n-1, sol_n
// We then invert that solution to create a scramble:
// scramble = sol_n' + sol_(n-1)' + ... + sol_1' + sol_0'
// We then prefix the scramble with scramblePrefix and suffix it with
// scrambleSuffix to create paddedScramble:
// paddedScramble = scramblePrefix + scramble + scrambleSuffix
// paddedScramble = scramblePrefix + (sol_n' + sol_(n-1)' + ... + sol_1' + sol_0') + scrambleSuffix
//
// We don't want any moves to cancel here, so we need to make sure that
// sol_n' doesn't cancel with the first move of scramblePrefix:
var solutionLastAxisRestriction = scramblePrefix[scramblePrefix.Length - 1].Substring(0, 1);
// and we need to make sure that sol_0' doesn't cancel with the first move of
// scrambleSuffix:
var solutionFirstAxisRestriction = scrambleSuffix[0].Substring(0, 1);
var psag = GenerateRandomMoves(r, solutionFirstAxisRestriction, solutionLastAxisRestriction);
var ab = new AlgorithmBuilder(MergingMode.NoMerging, GetSolvedState());
try
{
ab.AppendAlgorithms(scramblePrefix);
ab.AppendAlgorithm(psag.Generator);
ab.AppendAlgorithms(scrambleSuffix);
}
catch (InvalidMoveException e)
{
Assert(false, e.Message, e);
return(null);
}
return(ab.GetStateAndGenerator());
}