Re: [問題] How to decide the multi-class?
※ 引述《mengjia (...新人)》之銘言：
: ※ 引述《hyperstring (try and win)》之銘言：
: : What I mean is in "multi-class" case.
: : If a instance is : 3,5,7,11,25,35 1:10, 2:30....
: : So it finally belongs to "only one" class?
: : Case 1, ie. this instance just belongs to class 3.
: : Case 2, or this instance belongs to some of those classes,like 3, 11, 25
: : Is the homework one dealing with the case 1 or case 2?
: : I am confusing now.
: : Thinks for answering.
: I think that 3,5,7,11,25,35 is a class ,and
^^^^^^^^^^^^^^^^
This instance is belonging to these classes
: 1:10, 2:30.... is a news content .
^^^^^^^^^^^^^^^^
They are [index]:[feature's value]
: Right?
: I am not sure.
But my question is in multi-class K-NN, if we want to calculate the distance.
Then to vote every instance, in a simple K-NN question, we just decide every
instance belong which one class.(Ex: data1 is class 3, data2 is
class 2).
However, I want to know in multi-class K-NN problem, if the training data
class is 2,3,6,9,25,36. What is the definition of its algorithm?
Should we calculate the nearest one class, and put this data into this class?
(Ex. Data 1 belongs to class 25).
Or we calculate the nearest K classes, and put this data into those classes?
(Ex. Data 1 belongs to 3, 9, 25, these 3 classes).
--
※ 發信站: 批踢踢實業坊(ptt.cc)
◆ From: 140.112.41.95
※ 編輯: hyperstring 來自: 140.112.41.95 (09/26 18:54)
討論串 (同標題文章)
本文引述了以下文章的的內容：
完整討論串 (本文為第 5 之 6 篇)：