coverage range=(0.2,0.8)
Closed this issue · 1 comments
“The coverage range defines the range of accepted range of target area ratios compared to all cells. If, after generation, the generated area does not fall within that range, a new target zone is generated.”I still don’t understand the coverage range issue mentioned last time. Can you explain it in detail?The coverage is the acceptable range of the ratio of the target area to all cells. Does it refer to the range of cell ratios occupied by the randomly generated target area? So what impact does this range have on CPP results? What is its significance? What does it determine?If it is modified to coverage range=(0.1, 0.5), what will be the difference from (0.2, 0.8)?
It only determines what kind of scenarios are generated. The algorithm is the following:
num_shapes = random_choice(range(params.num_obstacles[0], params.num_obstacles[1] + 1)
while True:
target = draw_random_shapes(num_shapes) # Boolean array of (m, m)
target = target & ~obstacles
if coverage_range[0] <= sum(target) / (m**2) <= coverage_range[1]:
break
return target
Therefore, it has an impact on the scenarios generated and will thus have an effect on the statistics of the agent performance. If you decrease it to 0.1-0.5, the problem likely gets easier and it should require on average fewer steps to cover everything.