jlizier
3118e57384
Making conditional MI calculators (continuous) handle empty conditionals, and properly normalise new data according to the old (except for linear Gaussian calculator, which explicitly bars normalisation now).
2016-10-18 13:21:29 +11:00
joseph.lizier
21d1e68ba7
Bug fix for Kraskov conditional MI algorithm 2 -- kdTrees for var1 and conditional, as well as var2 and conditional, were only being passed the radii for var1 (or var2) but not the radius of the conditional as well. This doesn't seem problematic at face value, however in traversing the tree the wrong radius was being used at levels corresponding to the conditional(s) in checking those nodes and determining whether their children needed to be traversed as well. Patched by implementing a new method in KdTree to take all radii in the situation where one variable has already been checked. Fixes issue 39.
2015-01-21 12:23:09 +00:00
joseph.lizier
733da8265d
Dynamic correlation exclusion, or Theiler window, added to all Kraskov estimators, i.e. Mutual Info, Conditional MI, Multi-info, and thereby the derived classes including transfer entropy. Fixes issue 35
2014-11-25 12:31:44 +00:00
joseph.lizier
82ff884b99
Explicitly setting return array values in Kraskov conditional MI calculator algorithm 2.
2014-11-14 10:27:48 +00:00
joseph.lizier
ba537e948a
Enhancements to fast nearest neighbour searching for conditional mutual information; utilises points found matching the conditional search criteria in the joint conditional with var1 and var 2 searches.
2014-11-04 00:57:26 +00:00
joseph.lizier
6ade6d4311
Adding fast nearest neighbour searcher for univariate data (uses only a sorted array rather than whole k-d tree). Adds an abstract nearest neighbour searcher so calling classes don't necessarily need to know whether they're dealing with uni or multi-variates. MI and conditional MI Kraskov altered to use this where possible.
2014-10-30 04:04:16 +00:00
joseph.lizier
9c05f4693a
Fixes issue 32 -- adds fast neighbour search to Kraskov conditional MI calculators (and by implication to TE calculators). Requires adding a neighbour search within different radii for each variable to the k-d tree class.
2014-10-15 10:36:58 +00:00
joseph.lizier
70bcc7c922
Mirrors multi-threading from Kraskov MI calculators to Kraskov conditional MI calculators, and so fixes issue 29. Also adds some minor modifications to multithreading code for Kraskov MI calculators.
2014-09-16 01:36:39 +00:00
joseph.lizier
1ecdd8829f
Extra imports to fix Javadoc references
2014-08-13 03:48:54 +00:00
joseph.lizier
786784e83c
Javadocs made release-ready in the continuous.kraskov package.
2014-08-13 02:36:09 +00:00
joseph.lizier
26c98df75c
Correcting Conditional MI Calculator Kraskov algorithm 2 in line with Wibral et al.'s equation correcting how this should be done.
2014-08-12 05:24:54 +00:00
joseph.lizier
6709163afd
Adding GPLv3 license statement to all code headers in the infodymamics.measures.continuous.kraskov, symbolic and discrete packages (continuing ...)
2014-08-06 05:43:16 +00:00
joseph.lizier
4bd22a0984
Reconditioned Conditional MI calculator Kraskov as a child class of Conditional MI calculator Common
2013-03-11 12:55:29 +00:00
joseph.lizier
cd67da53a4
Patching linear Gaussian estimators to give empirical statistical significance given a certain number of observations.
...
Changed implementation of our conditional MI calculators Kraskov via algorithm 2 to my current interpretation of how this should be done.
Added a utility for lagged covariance in MatrixUtils
2013-03-04 06:08:38 +00:00
joseph.lizier
9d36e5d9a6
Bug fix to debug prints in Kraskov Conditional MI calculators
2013-01-15 01:29:57 +00:00
joseph.lizier
6761de7e02
Added debug prints to Kraskov Conditional MI calculators
2013-01-15 01:14:13 +00:00
joseph.lizier
f466012556
Made Kraskov Conditional MI calculator implement the common Conditional MI interface
2013-01-14 12:51:34 +00:00
joseph.lizier
9b9abda733
Bug fix of computation of epsilon (box width) on conditional variable in Kraskov conditional MI calculator 2
2013-01-14 11:13:58 +00:00
joseph.lizier
617880f9f5
Making Kraskov MI and higher order calculators use MAX_NORM in the marginal spaces by default (previously this had to be supplied via a property - it was the standard choice made, but was not the default). This aligns with the default specified in the Kraskov paper.
2012-12-20 10:38:03 +00:00
joseph.lizier
56d868ca00
Uploading whole of Java Information Dynamics toolkit for the first time.
2012-05-08 00:18:37 +00:00