Fix bug when no uncovered neighbors are avaiable and let rewarded hints match expected value #484
Conversation
|
|
||
| val randomNum = (1..100).random() | ||
|
|
||
| return if (randomNum <= (x - x.toInt())*100) x.toInt() + 1 else x.toInt() |
There was a problem hiding this comment.
What if we always round up instead of add randomness?
There was a problem hiding this comment.
I would say this would lead to essentially the same problem, but with an additional hint unless q*M is an integer.
The szenario is that people subsequently increase mines to increase difficulty. But difficulty increases disproportionally stronger when adding mines because the number of rewarded hints does not increase with the number of mines.
Any reason why you would like to avoid randomness? Note that the randomness only affects the +1 part.
There was a problem hiding this comment.
An alternative idea would be to store the number of hints internally as a double, and only display it as an int. Then calcRewardHints() could simply return rewardedHints directly without toInt().
This PR brings a bugfix and a feature.
The bug is that currently, when using a hint but no mine with uncovered neighbor is available, no mine is revealed (e.g. when the complete coastline is flagged). Fix is to check this and reveal a random mine otherwise.
The feature is that currently, the number of rewarded hints
his always the rounded down fractionq(currenty 5%) of the total number of minesM, thush = floor(q*M). This means thath/M <= qand thereby induces a negative bias, especially when using small fields. The fix is to let the expected value ofhequalq*Mby randomly chosing betweenfloor(q*M)andceil(q*M)with the appropriate probability (all this sounds more complicated than it is).Cheers!