I do agree. It's a good solution too. I'll keep it in mind. This problem was very narrow, Given two numbers return a value between -1 and 1 representing similarity. That's all. Thanks to all respondents Ben, Jim, John, Aaron, etc. (sorry if I left anyone out). Crowdsourcing works. Collaboration works too. Michael. Date: Sat, 22 Feb 2014 12:05:51 +0800 Subject: Re: [agi] Numeric Similarity From: [email protected] To: [email protected]
It's still important that the little cogs in the big machine are not broken, though ;) Anyway we have similar issues in OpenCog and deal with them via context-specific quantile normalization as I noted above. The heuristic I suggested to you earlier works but you need to specify a reference context which is associated with some distribution of numbers. I don't see how to avoid this relativity to a reference context without sinking into meaninglessness... ben On Sat, Feb 22, 2014 at 12:02 PM, Piaget Modeler <[email protected]> wrote: You are correct context is very important, and it is meaningless to say 1000 ~ 100 = 0.0011098 Numeric similarity is a small cog in a larger machine. We can talk about it later if you want. Cheers, ~PM Date: Sat, 22 Feb 2014 11:48:14 +0800 Subject: Re: [agi] Numeric Similarity From: [email protected] To: [email protected] But that is almost meaningless For instance, it says similarity(1000000, 1001000) = 1/ 1001 which is very small yet, in many contexts (e.g. city populations), those numbers would be considered very similar You need to consider the context in order to have a useful, general numeric similarity measure Unless you already have some implicit context you are not mentioning.. ben On Sat, Feb 22, 2014 at 11:44 AM, Piaget Modeler <[email protected]> wrote: Two measures: The first measure 1 - | a - b | where a = 1000, and b = 100 .: -900. I said that I wasn't quite happy with this measure but the value is still less than 1. So I was willing to work with it. The second measure 1 / ( 1 + | a - b |) where a = 1000, and b = 100 .: 0.0011098 The second measure is better because the value falls within the range -1 to +1. So I'm happier with Aaron Hosford's recommendation. Michael. Date: Sat, 22 Feb 2014 10:59:23 +0800 Subject: Re: [agi] Numeric Similarity From: [email protected] To: [email protected] I don't understand... if a=1000 and b = 100, then a-b=900 so 1 - | a - b | = 1-900 = -899 That doesn't make sense as a similarity score... Your posited measure seems only applicable when |a-b|<=1, which indeed would be the case if a and b lie in [-1.1] -- Ben On Sat, Feb 22, 2014 at 10:56 AM, Piaget Modeler <[email protected]> wrote: A and B are any two numbers that can be represented computationally. We are looking for the similarity of any two numbers as part of a larger similarity based matching scheme. The function Aaron provided will allow the similarity to be computed in the range 0 .. +1 which is fine. As longas it's between -1 and +1 it will do. Cheers, Michael. Date: Sat, 22 Feb 2014 09:54:20 +0800 Subject: Re: [agi] Numeric Similarity From: [email protected] To: [email protected] I assume your numbers a and b are between -1 and 1? you didn't mention that in your original email ! ... On Fri, Feb 21, 2014 at 10:02 PM, Piaget Modeler <[email protected]> wrote: Thanks to all respondents. In the end I found a classic numeric similarity metric: 1 - | a - b | It's not ideal since numeric scores can dominate other attribute scores. Ergo, I have to devise a good weighting scheme. Nothing's perfect I suppose. Cheers, ~ PM AGI | Archives | Modify Your Subscription -- Ben Goertzel, PhD http://goertzel.org "In an insane world, the sane man must appear to be insane". -- Capt. James T. Kirk AGI | Archives | Modify Your Subscription AGI | Archives | Modify Your Subscription -- Ben Goertzel, PhD http://goertzel.org "In an insane world, the sane man must appear to be insane". -- Capt. James T. Kirk AGI | Archives | Modify Your Subscription AGI | Archives | Modify Your Subscription -- Ben Goertzel, PhD http://goertzel.org "In an insane world, the sane man must appear to be insane". -- Capt. James T. Kirk AGI | Archives | Modify Your Subscription AGI | Archives | Modify Your Subscription -- Ben Goertzel, PhD http://goertzel.org "In an insane world, the sane man must appear to be insane". -- Capt. James T. Kirk AGI | Archives | Modify Your Subscription ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
