Result 'infinity' for gaussian estimator

[F Sahl]

Dear Dr. Lizier,

I am working on the transfer entropy with the IDTxl Toolbox and I would like to calculate the connection between photoplethymsograms.

For the calculation the GaussianCMI estimator is used with the following

settings:

settings = {'cmi_estimator': 'JidtGaussianCMI',

'max_lag_sources': 5,

'min_lag_sources': 1,

'tau_target': 1,

'tau_sources': 1}

For one calculation 7 timeseries with each 600 samples are used.

The results yield now and then "infinity" for the

transfer entropy which in my understanding cannot be.

If I increase the maximum lag this result occurs more often.

Are you familiar with this result and do you know the cause of it?

Kind regards,
Fabienne Sahl

[Joseph Lizier]

Dear Fabienne,

The result here means that the source is completely predicting the target (with no remaining uncertainty), given the conditionals. For continuous-valued variables, which can be specified with infinite precision, this means an infinite amount of information is provided.

So it sounds as though there is a linear redundancy amongst the variables here - does that sound plausible?

It's also possible that, with only 600 samples available, we're moving to dimensionalities that are not supported by the amount of data, though I would not have expected that. If you get the error again, perhaps you can send the data to me (and details on target, and selected sources) and we can take a look.

--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/6ee3a5fe-4eb9-4d89-a36b-6e622d736df5n%40googlegroups.com.

[F Sahl]

Hello Joe, 

thank you for your quick response! 

I'll have another look at the data and think about it. Thanks for your thoughts.

If I am struggling again I'll come back to this.

Thanks again, 

Fabienne