Assignment1 large variance on image data
6 views
Skip to first unread message
Linus Färnstrand
Jan 26, 2012, 4:23:43 AM1/26/12
to ml_chalmers
Hi.
I am having problems running the Multivariant Gaussian Distribution on
the 256 dimensional data with the range 0-255. Since the variance
becomes big everything just ends up as Inf and the returned
probability becomes 0 for all images. Might I have the wrong variance
(even quite small variances makes it not computable) or can I scale
the data on 3a&b also?
I am having problems running the Multivariant Gaussian Distribution on
the 256 dimensional data with the range 0-255. Since the variance
becomes big everything just ends up as Inf and the returned
probability becomes 0 for all images. Might I have the wrong variance
(even quite small variances makes it not computable) or can I scale
the data on 3a&b also?
Vinay Jethava
Jan 26, 2012, 4:59:01 AM1/26/12
to ml_ch...@googlegroups.com
On question 3 - the first part works without any scaling needed. For the other two features - it is ok to scale the dataset to [0, 1] regime. In any case, you are welcome to scale it since it would not change the accuracy results. Bonus question: Why is this the case? Does the same apply in the general multi-variate Gaussian case?
Reply all
Reply to author
Forward
This conversation is locked
You cannot reply and perform actions on locked conversations.
0 new messages