This patch update addresses a few of the Information theoretic SPIs, in addition to convergent cross-mapping, which were outputting inconsistent results across multiple trials on the same benchmarking dataset. This stochasticity has been addressed for all information-theoretic SPIs in the latest version of the pyspi dependency JIDT (v1.6.1, see here).
As a result, the following changes have been made:
JIDT dependency has been upgraded from 1.5 to 1.6.1 to reflect the inclusion of a NOISE_SEED property for all JIDT calculators.
Random seed has been added for all JIDT-based SPIs.
Random seed has been added for convergent cross-mapping ccm SPI.
Results tables for the cml7 benchmarking dataset have been re-generated using the new fixed seed.
The following SPIs (which were producing varying outputs) should now produce consistent results:
This patch update addresses a few of the Information theoretic SPIs, in addition to convergent cross-mapping, which were outputting inconsistent results across multiple trials on the same benchmarking dataset. This stochasticity has been addressed for all information-theoretic SPIs in the latest version of the pyspi dependency
JIDT(v1.6.1, see here).As a result, the following changes have been made:
JIDTdependency has been upgraded from1.5to1.6.1to reflect the inclusion of aNOISE_SEEDproperty for allJIDTcalculators.JIDT-based SPIs.ccmSPI.cml7benchmarking dataset have been re-generated using the new fixed seed.The following SPIs (which were producing varying outputs) should now produce consistent results:
ccm(and all related estimators)cce_kozachenkoce_kozachenkodi_kozachenkoje_kozachenkosi_kozachenko_k-1