Open wingos80 opened 2 months ago
Added functionality with commit 4d91fd7
Number of cells in far cells assumed to be: $MINES - N{flags} - \Sigma x{informed}$. Did not prioritize selecting far cells closest to the revealed cells, this might improve solver performance (TODO).
Some improvements in win rates:
Still need to add truly random selection of far cells, currently only using np.argmin() which selects the first lowest index. Also need to experiment with only computing far cells estimates if informed cells are all uncertain (e.g. x values between 0.2 and 0.8).
Add logic to add distant tiles to the unknown vector if neighbouring tiles of revealed tiles are uncertain, e.g. $0.2 < x < 0.7$. Really useful when the cells neighbouring revealed cells cannot be deterministically solved, i.e. it's impossible to know exactly where/how many bombs are in those cells.
Could do something like:
Equivalently, adding the far cells to the solution vector, setting the value of these cells to be $n{bombs}/n{FarCells}$, then selecting the cell with the lowest value. Computing $n{bombs}$ could be done in different ways, one of the way is to set $n{bombs}$ equal to total number of bombs undetected, or to set $n_{bombs}$ to be total number of bombs undetected - Sum($x$) ... etc.