I have a time series data over one month and it is normalized according to min-max normalization. I would like to see relationships within data. When using PC learning Genie tells me samples out of bounds. Does it mean that it is not possible to use PC learning results to see a true picture about relationships within data? Regardless of samples out of bounds are the equations generated by PC useful in some ways? Are there any ways to improve this situation?
Thanks in advance,
PC and samples out of bounds
-
- Site Admin
- Posts: 1417
- Joined: Mon Nov 26, 2007 5:51 pm
Re: PC and samples out of bounds
What's the exact message and when exactly are you getting it - during PC learning, during parameter learning or later?ind wrote:When using PC learning Genie tells me samples out of bounds.
Re: PC and samples out of bounds
I got this message after parameters have been learned. Does it assume normally distributed data?
-
- Site Admin
- Posts: 1417
- Joined: Mon Nov 26, 2007 5:51 pm
Re: PC and samples out of bounds
The message is caused by default bounds specified per node. These are not adjusted by PC or continuous parameter learning procedure you can invoke from the pattern editor window.ind wrote:I got this message after parameters have been learned.
You can either ignore the message, edit the bounds in node definitions and/or check the "Reject outlier samples" in network properties/sampling page.