Multivariate Conditional TE

[Ronaldo Nunes]

Hi

I am trying to compute a TE matrix for LFP signals. I will consider the embedding dimension and embedding delay as 1. Is there any optimized way to choose the lag between source and target variable and conditionals and target.

It's been a problem because I have 19 LFP signals and the possibilities of lags, considering the source, conditionals, and target, are enormous.

I am using ConditionalTransferEntropyCalculatorKraskov.

Best,

Ronaldo Nunes 

[Joseph Lizier]

Hi Ronaldo,

Yes, there are a large number of possibilities there!

I'd suggest that:

1. For this application, you would be best to go directly to IDTxl, which wraps higher level algorithms around JIDT for these purposes. For a given target, it will gradually learn the multivariate set of sources, optimising the lag for each along the way. The final conditional TEs it returns for each source will be the conditional TE from that source conditioned on the parent set (without that particular source included).

2. Otherwise, without going and learning the set of parents for each source first, it's a little undefined what you want to / should do. You could compute conditional TE from every source to the target, conditioned on every other source. But as you say, how to set the lag for each conditional. You could set them in a pairwise fashion - finding the lag that maximises TE from the given conditional source to the target, and then using that lag for that conditional in the calculation for every other source to that target. That probably makes the most sense. But it still ignores the possibility that the lag you might consider could be changed once you start to consider it in conjunction with the other sources.

hope that helps,

--joe
+61 408 186 901 (Au mobile)

--
JIDT homepage: https://github.com/jlizier/jidt
---
You received this message because you are subscribed to the Google Groups "Java Information Dynamics Toolkit (JIDT) discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to jidt-discuss...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/jidt-discuss/60ba79ad-b36d-45e0-83ad-133528b92b71n%40googlegroups.com.

[Ronaldo Nunes]

Hi Joe

Thank you very much for the suggestions. 

Indeed, I considered IDTxl but I would like to verify the Conditional TE for all possible pairs of nodes, including those non-significant statistically. Unfortunately, as far as I know, IDTxl returns only the significant values for TE.

I think that the way to go is following your second suggestions.

Again, thank you 

Best,

Ronaldo Nunes

To view this discussion on the web visit https://groups.google.com/d/msgid/jidt-discuss/CAL81BvRttn1z%2B%3DfoX7dGzG%3DG3bE8iKQmGQ9vFB%3Dd6b57D1HZuw%40mail.gmail.com.