v1.5.1 01/2024

Bug fixes:

Update to JIDT version 1.6.1. Fix formatting issues and minor bugs.

v1.5 12/2023

New features:

• Implementation of purely Python-based (C)MI-estimators and MPI-support for serial (C)MI-estimators by @daehrlich

v1.4 04/2022

New features:

• Implementation of significant subgraph mining by @aarongutknecht as described in the biorXiV preprint

Bug fixes:

• Ensure that output in results classes contains only numpy and not JPype types. This eases further processing of outputs, especially loading and saving of results (if JPype types are saved, a JVM has to run when loading them again)

v1.3 02/2022

New features:

• Implementation of history-dependence estimator for neural spike data (by @DrMichaelLindner)

v1.2.2 05/2021

Fixes:

• Fix call to maximum stats in multivariate network inference (use correct conditioning set when performing statistics)

Minor fixes:

• Update PID references in README (#67)

v1.2 02/2021

New features:

• Checkpointing: allows to restart an analysis in case of a crash (by @makavelj)
• Random seed for Data class: sets seed to control generation of data permutations (by @makavelj)

Fixes:

• OpenCl estimators crash for certain problem instances (#30, #10)
• Minor fixes:
  • Compatibility with latest networkx version (#55)
  • Export of DiGraph objects for nodes without links (#50)

v1.1 05/2020

New features:

• Multivariate, differentiable Partial Information Decomposition (Makkeh et al., 2020, arXiv:2002.03356 [cs.IT])

Fixes:

• Fixing compatibility for JPype1 0.7 and improving Java memory usage
• Adds Kraskov algorithm 2 and improves memory error handling
• Minor fixes:
  • Fixes installation process, #37, #41
  • Fixes max stats for AIS and bivariate MI estimation
  • Fixes surrogate creation for maximum seq. statistics

Release peer reviewed and accepted for publication in JOSS

This release was peer-reviewed for publication in the Journal of Open Source Software.

Updated development release

Updates to documentation, gh-pages, and unit-/system-tests.

Improvements:

• Improve handling memory exhaustion for JidtDiscrete estimators. This also required including a new JIDT jar. Please note there are occasions where the OS cannot provide more memory (even though heap is large enough) where Java crashes and via jpype1 this seems to kill python. I will continue to investigate this but it may not be solveable.

Bug fixes:

• Fixes #15 by adding Kraskov algorithm 2 option for all JidtKraskov estimators. This required including a new JIDT jar. Unit test included for MI.

Updated development release

This is the second development release. Note that algorithms are still
in beta stage. Also, there may be changes to the API in future releases.

To get started with using IDTxl have a look at the wiki
pages describing the installation
process
and the example
script
for network inference. There are also examples in the docstrings of the
algorithm classes. Further documentation: http://pwollstadt.github.io/IDTxl/
and https://github.com/pwollstadt/IDTxl/wiki.

Stable algorithms (see the
demos for examples):

• multivariate_transfer_entropy.py
• bivariate_transfer_entropy.py
• multivariate_mutual_information.py
• bivariate_mutual_information.py
• active_information_storage.py
• partial_information_decomposition.py
• network_comparison.py (group-level statistics)
• visualise_graph.py
• core-estimators, see the wiki page for examples

Added features:

• (lagged) multivariate and bivariate MI estimation for network inference
• bivariate TE estimation for network inference
• Results() class: replaces results dictionary, adds functionality to generate
  adjacency matrices and access detailed results for individual targets
• generation of synthetic test data in the Data() class (coupled logistic maps and autoregressive processes)
• demo scripts for network inference algorithms and core estimators

Improvements:

• add jar-file supporting JAVA v6 (fixes #9)
• cleaned up console output
• update of the Tartu estimator

Bug fixes:

• OpenCL estimators now run on Nvidia and AMD cards (fixes #10)
• Labeling of nodes in source graph
• The minimum statistics did not use the correct conditioning set during the pruning step of the multivariate TE algorithm, causing a bias in the test
• Non-uniform embedding was not built correctly for bivariate measures

Known issues and missing features:
- OpenCL estimators fail on AMD cards in some cases due to driver settings
that introduce limitations on maximum variable size
- spectral multivariate transfer entropy estimation will be added in a future release
- the Kraskov2 algorithm will be available for estimation in a future release (issue #15)