My Math Forum How to quantify differences/ similarities between groups of like objects?
 User Name Remember Me? Password

 Linear Algebra Linear Algebra Math Forum

 February 21st, 2016, 10:58 PM #1 Newbie   Joined: Feb 2016 From: denver Posts: 1 Thanks: 0 How to quantify differences/ similarities between groups of like objects? First let me apologize if this is not an appropriate forum for this question. Researching all the available SE sites led me here. Also, my background is not in mathematics, so layman's terms are appreciated The problem: Given a set of items within a category (e.g. types of fruit), and Given a table providing similarity values for each item as compared to each other item (how like an apple is a pear?), and Given a control set, X (a fruit basket containing, e.g. 60 apples and 40 oranges), and Given an arbitrary group of comparison sets (a fruit basket A containing 30 apples, 10 pears, and 60 strawberries) Construct a method for determining each comparison set's similarity to the control. It may be assumed that the total number of pieces of fruit in each set, including the control, will be exactly 100. Continuing the fruit example for five types of fruit (and considering visual likeness alone, i.e. how each type of fruit looks like another), here is the similarity table: Apple Orange Pear Banana Strawberry Apple 1.0 0.8 0.6 0.2 0.1 Orange 1.0 0.5 0.2 0.05 Pear 1.0 0.25 0.15 Banana 1.0 0.1 Strawberry 1.0 Control set X: {60A, 40B} Comparison sets: A: {100A} B: {40A, 40P, 20S} C: {80A, 20O} D: {60A, 30B, 10P} As a first approach I thought a simple solution would be to take each type of fruit within a set, compare it with an identical type of fruit in the control, and take the difference for each type. To make this somewhat easier, I fill each set with imaginary fruit where necessary: A => {100A, 0B, 0P, 0O, 0S} then simply take the absolute difference between each type of fruit and the control: dA => {100 - 60A, 40 - 0B, 0P, 0O, 0S} => {40A, 40B, 0P, 0O, 0S} and so on for B, C, D. This works ok for identical types (apples to apples) but ignores the fact that some types of fruit are visually quite similar (apples to oranges). So my next approach was to say "how alike is each type of fruit?". For set B: X: {60A, 40B, 0O, 0P, 0S} B: {40A, 0B, 0O, 20P, 20S} We can compare apples precisely since they appear in both sets, but for the other fruit, we can say "how alike are these?" and "what fruit is most like this one?". But here is where it gets a little sticky. Since there are pears in set B but not in X, and bananas in X but not in B, we could compare pears against bananas: Pears to bananas: (40B - 20P) * 0.25 = 5 but it seems just as valid to compare everything against apples (or whatever the predominant fruit is in the control). I think I can also break the methodology down into two pieces: 1:how alike are the individual types of fruit, and 2: how alike are the distributions of each type. I ran into problems with that approach however where certain distributions of un-like fruit ranked higher than less-similar distributions of more-like fruit - which on the face of it seems wrong. So my question is: is what I'm trying to do just complete madness? Or is this one of those "well understood, just not by me" problems? And if the latter, please help me to understand. general-topology scoring-algorithm

 Thread Tools Display Modes Linear Mode

 Similar Threads Thread Thread Starter Forum Replies Last Post joexian Number Theory 0 January 15th, 2013 08:50 PM winnie Differential Equations 1 December 8th, 2012 09:23 AM bentley4 Calculus 1 May 19th, 2011 11:23 AM Marz Algebra 3 March 13th, 2009 07:54 AM holiday Advanced Statistics 9 March 29th, 2008 10:40 AM

 Contact - Home - Forums - Cryptocurrency Forum - Top

Copyright © 2018 My Math Forum. All rights reserved.