A perfect player describes nobody#
SLOP2875 · Week 3
Swipe, or press →
Last week the player became a function#
This week is about how good that function is allowed to be.
A perfect player measures a game nobody will play#
The first simulated player#
Re-decided its direction every frame. Every 16 ms.
Fifteen runs, ordinary upgrade choices.
It won 12.
On those numbers#
The game was too easy.
It wanted harder bosses.
Then the player was made to wait#
Re-decides about every 150 ms, not every 16.
Holds its last direction in between.
Aims up to nine degrees off.
Same fifteen seeds#
| Reaction time | Wins out of 15 | Median survival |
|---|---|---|
| none | 12 | 120.0 s |
| 100 ms | 3 | 103.4 s |
| 150 ms | 1 | 81.5 s |
| 250 ms | 0 | 60.7 s |
Nothing about the game changed between those rows#
Every decision made against the first row#
would have been made for a player who does not exist.
Reading, weeks 3 to 5#
Isaksen, Gopstein and Nealen, Exploring Game Space Using Survival Analysis.
Flappy Bird, with a player built from precision, reaction time and actions per second.
The player model this week. The survival analysis next week.
Thursday#
Two simulated players. One of them with a reaction time.
The win rate of each over the same fifteen seeds.
Week 3, in full#
The page for this week has the numbers and where they came from.