在Numpy数组上执行盒模糊的奇怪结果

Question

我正在使用Numpy在数组上快速执行操作。具体地说，我希望将数组值中的每个位置设置为其所有近邻的平均值，从而有效地使值“扩散”和分散。它的预期用途实际上是模拟视频游戏中气味的行为；数组中的值表示特定对象的气味强度，并且在该对象的位置处，强度设置为每回合100。在

我知道有一些包是为了这个目的而存在的，但是我真的不想引入任何比我必须的更大的依赖性；Numpy有点像在推它！在

不幸的是，我的代码似乎不起作用。这些值完美地“传播”；不幸的是，它们实际上并没有“分散”，而是在不断累积，直到溢出为止。在

代码如下：

#Step 1: Set every object's scent value to its scentstrength at its position, simulating objects emitting scent
#(this also initializes the arrays if they haven't been yet as a side effect since scentmap is a defaultdict)
#Walls don't have smell, which is good 'cos there's a lot of em
scentsimulatedobjects=(theobj for theobj in self.objs if not isinstance(theobj,gameobjs.Wall))
for theobj in scentsimulatedobjects:
    thearray=scentmap[theobj]
    thearray[tuple(theobj.pos)]=theobj.scentstrength

#Step 2: Make the scents spread. This is done the same way as a simple "blur" effect in image processing;
#we just make every square's scent value the average of all of its neighbors. This is done by convolving the
#array with a blur convolution filter. I'm totally pretending I actually understand this stuff, I really don't.
spreadfactor=0.95
blur=numpy.array([spreadfactor, 1, spreadfactor])

for theobj, thearray in scentmap.items():
    for idx, therow in enumerate(thearray):
        thearray[idx] = numpy.convolve(therow, blur, "same")

    for idx in range(self.size):
        thearray[:, idx] = numpy.convolve(thearray[:, idx], blur, "same")

#Step 3: Scale the entire matrix by a decay factor, representing scents losing strength over time once their
#source #leaves
decayfactor=0.5
for theobj, thearray in self.scentmap.items():
    thearray*=decayfactor
    for x, therow in enumerate(thearray):
        for y, scentstrength in enumerate(therow):
            if scentstrength < 1:
                thearray[x,y]=0

我怎么解决这个问题？在

这是一个从上下文中分离出来的算法的pastebin，因此您可以在不下载整个项目的情况下对其进行破解，这是其中的一部分： http://pastebin.com/gr2E6VWP