forum.alglib.net
http://forum.alglib.net/

Questions about the w value in fisherldan function
http://forum.alglib.net/viewtopic.php?f=2&t=3801		 Page 1 of 1
Author:  FlyingBlue [ Wed May 03, 2017 10:05 am ]
Post subject:  Questions about the w value in fisherldan function
Hello everyone.
I have some questions on the fisherldan function.

1) What are the differents between the Fisherlda function and the Fisherldan function?
2) I use two points A(3, 1.732) and B(4, 0) to test the function and the result W shows two rows of vectors and it looks like a matrix (see the attachment TestResult1). It kind of shows the gradient of the line but where is the intercept?
3) I move these two points 1 unit to the right as A(4, 1.732) and B(5, 0) and the result is the same as the above (see the attachment TestResult2). Why this happans and what should I do next?

Thanks in advance

Attachments:
TestResult2.png
TestResult2.png [ 12.18 KiB | Viewed 5469 times ]
TestResult1.png
TestResult1.png [ 11.5 KiB | Viewed 5469 times ]
Author:  Sergey.Bochkanov [ Wed May 03, 2017 10:35 am ]
Post subject:  Re: Questions about the w value in fisherldan function
Hi!

Firsher LDA is not used to perform linear regression. It is used to reduce dimensionality of your data in a way which helps to linearly separate classes. First column of array W contains direction which gives best separability, second column contains direction which gives worst separability.

If you move all points in same direction, W should not change. If you rotate them - W will rotate too.
Author:  FlyingBlue [ Thu May 04, 2017 3:07 am ]
Post subject:  Re: Questions about the w value in fisherldan function
Sergey.Bochkanov wrote:
Hi!

Firsher LDA is not used to perform linear regression. It is used to reduce dimensionality of your data in a way which helps to linearly separate classes. First column of array W contains direction which gives best separability, second column contains direction which gives worst separability.

If you move all points in same direction, W should not change. If you rotate them - W will rotate too.


Thank you very much! That helps a lot!

Could you please tell me the differences between the fisherlda function and the fisherldan function? According to my understanding, the fisherlda is used for one or two dimentions data and fisherldan is used for three or more, is that correct?
Author:  Sergey.Bochkanov [ Thu May 04, 2017 10:11 am ]
Post subject:  Re: Questions about the w value in fisherldan function
fisherlda is used to get just one "most important" direction, typically in a problem with two classes (not dimensions! you can have more than two dimensions).

fisherldan returns full basis, you can pick arbitrary number of vectors from this basis.
Page 1 of 1 All times are UTC
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group
http://www.phpbb.com/