You will need this ...
https://fairy-stockfish.github.io/nnue/
If you are using linux, make the engine executable:
chmod +x ./fairy-stockfish-largeboard_x86-64-modern
python3 safe_fen.py
Run in a script both engines at depth 8 for each fen eval. Then match the fens that work best in both variants.
Checked 960/960 positions
Done! Balanced positions for both variants saved to balanced_both.csv
FEN,CrazyhouseEval,AtomicEval
rnnbbkrq/pppppppp/8/8/8/8/PPPPPPPP/RNNBBKRQ w KQkq - 0 1,0.15,0.07
My first attempt gave only one result. Crazyhouse and Atomic are very different variants:
Crazyhouse allows drops ... positions that are normally balanced in classical chess can become lopsided quickly.
Atomic has explosive captures ... a position that is fine in normal or Crazyhouse can be completely unbalanced in Atomic.
So when you intersect “balanced in Crazyhouse and balanced in Atomic,” the set becomes extremely small.
I had 113 fens that were balanced just for Crazy house and only 2 fens for atomic.
Processed 960/960 positions
Finished! Balanced positions saved to balanced_atomic.csv
FEN,Eval
rnnbbkrq/pppppppp/8/8/8/8/PPPPPPPP/RNNBBKRQ w KQkq - 0 1,0.07
brnbqkrn/pppppppp/8/8/8/8/PPPPPPPP/BRNBQKRN w KQkq - 0 1,0.26
If you want to see the 113 for crazyhouse, I have them ... @moveNumber37
.
atomic-2cf13ff256cc.nnue
balanced_atomic.csv
balanced_both.csv
balanced_crazyhouse.csv
best_fen2.py
crazyhouse-8ebf84784ad2.nnue
fairy-stockfish-largeboard_x86-64-modern
safe_fen2.py
1 directory, 8 files
Instead of EVAL_MARGIN = 0.3, I could try 0.4 or 0.5.
This would give us more candidates, which could still be reasonably “balanced” in both variants.
You will need this ...
https://fairy-stockfish.github.io/nnue/
If you are using linux, make the engine executable:
chmod +x ./fairy-stockfish-largeboard_x86-64-modern
python3 safe_fen.py
Run in a script both engines at depth 8 for each fen eval. Then match the fens that work best in both variants.
Checked 960/960 positions
Done! Balanced positions for both variants saved to balanced_both.csv
FEN,CrazyhouseEval,AtomicEval
rnnbbkrq/pppppppp/8/8/8/8/PPPPPPPP/RNNBBKRQ w KQkq - 0 1,0.15,0.07
My first attempt gave only one result. Crazyhouse and Atomic are very different variants:
Crazyhouse allows drops ... positions that are normally balanced in classical chess can become lopsided quickly.
Atomic has explosive captures ... a position that is fine in normal or Crazyhouse can be completely unbalanced in Atomic.
So when you intersect “balanced in Crazyhouse and balanced in Atomic,” the set becomes extremely small.
I had 113 fens that were balanced just for Crazy house and only 2 fens for atomic.
Processed 960/960 positions
Finished! Balanced positions saved to balanced_atomic.csv
FEN,Eval
rnnbbkrq/pppppppp/8/8/8/8/PPPPPPPP/RNNBBKRQ w KQkq - 0 1,0.07
brnbqkrn/pppppppp/8/8/8/8/PPPPPPPP/BRNBQKRN w KQkq - 0 1,0.26
If you want to see the 113 for crazyhouse, I have them ... @moveNumber37
.
atomic-2cf13ff256cc.nnue
balanced_atomic.csv
balanced_both.csv
balanced_crazyhouse.csv
best_fen2.py
crazyhouse-8ebf84784ad2.nnue
fairy-stockfish-largeboard_x86-64-modern
safe_fen2.py
1 directory, 8 files
Instead of EVAL_MARGIN = 0.3, I could try 0.4 or 0.5.
This would give us more candidates, which could still be reasonably “balanced” in both variants.