Long standing bug in mm values of 3x3 patterns I checked back it is in since very early in your branch

[dsmic]

Since circular patterns this is not used, but it may explain why our 3x3 patterns were bad.

Especially I wondered, why they were unsucessfull in playouts (instead of the mogo patterns)

Please CHECK if you agree this is a bug!

around line 2138

level=this->matchFeatureClass(Features::PATTERN3X3,board,cfglastdist,cfgsecondlastdist,move);
  if (patterngammas->hasGamma(level) && !move.isPass() && !move.isResign())
  {
    if (pretty)
      ss<<" pattern3x3:0x"<<std::hex<<std::setw(4)<<std::setfill('0')<<level;
    else
//probably buggy?!:      ss<<" "<<(int)patternids->getGamma(level);
      ss<<" "<<(base+level-1);
  }
  base+=patternids->getCount();

[dsmic]

Think I did not notice the double usage of gamma as count in this case

[Francois van Niekerk]

Edited issue to add code tags for readability.

[Francois van Niekerk]

Yes, patternids != patterngammas.

For a moment I thought you might have been correct, but it seemed a bit unlikely because the bug would also translate to the progressive widening, where it clearly works to some degree. Glad to see I don't have to redo my tests ;)

[dsmic]

Sorry for the noise,

I am busy with replacing the MOGO type patterns in playouts with something learned from games. Just to let you know, that we do not make the work double:)

[Francois van Niekerk]

Sounds interesting. I am not working on playouts at the moment.