Each counting "bit" gets a TTL. Curious. I'd think decay would be more appropriate—closer to a model of perceptual impact—than complete expiration. How would we do that? Each bit a process. You know where this is going…

Interesting. It could be said that the entire vector, assuming a hit, could be no newer than the oldest of its points. Also if you used some sort of decaying model you could make a guess at how false a positive might be based on a sort of communal agreement — how many share the same age

Reading Ilya Grigorik's article led me to this post here while searching for time based Bloom filter implementations, which I did not find. I came up with a solution which I call the "continuous" Bloom filter, which I just described in a post on my blog: http://colinsurprenant.com/blog/2012/05/12/contin…

Each counting "bit" gets a TTL. Curious. I'd think decay would be more appropriate—closer to a model of perceptual impact—than complete expiration. How would we do that? Each bit a process. You know where this is going…

Interesting. It could be said that the entire vector, assuming a hit, could be no newer than the oldest of its points. Also if you used some sort of decaying model you could make a guess at how false a positive might be based on a sort of communal agreement — how many share the same age

Reading Ilya Grigorik's article led me to this post here while searching for time based Bloom filter implementations, which I did not find. I came up with a solution which I call the "continuous" Bloom filter, which I just described in a post on my blog: http://colinsurprenant.com/blog/2012/05/12/contin…

Hope it may be useful.

Colin