Skip to content

Tigramite 4.1
Compare
Choose a tag to compare
@jakobrunge jakobrunge released this 01 Dec 20:51
· 627 commits to master since this release
4.1
eee4809

Tigramite 4.1 is a further improvement, but the main changes are under the hood. The PCMCI method is now published in Science Advances: https://advances.sciencemag.org/content/5/11/eaau4996

Upgrade Steps

  • Pull new version from master branch
  • Install via python setup.py install
  • (potentially install R-dependencies before)
  • Take note of the further installation instructions
  • Take a look at the new tutorials to see all functionality

Breaking Changes

  • Parameters related to the data itself are not set in the DataFrame object anymore. You will need to rewrite the initialization of PCMCI and some of its attribute functions
  • Removed GPACE class, which is replaced by GPDC
  • In run_pcmci tau_min = 0 is new default. This just means that there are further values in the returned val_matrix and p_matrix that were set to 0 (or 1) before
  • CMIknn now has transform = 'ranks' as default since it works better
  • Again, take a look at the new tutorials to see all functionality

New Features

  • Improved independence tests with run_test_raw function to use directly on data
  • Improved RCOT conditional independence test (thanks to E. Strobl)
  • Dedicated functions to estimate FullCI and BivCI (see Science Advances paper)
  • Several more under the hood, take a look at the new tutorials to see all functionality

Bug Fixes

  • small bugs (e.g., related to var_names)
  • Adapted to API changes of other packages (e.g., networkx)

Improvements (thanks to Ewen Gillies! And the issues posted by various users)

  • Added a comprehensive test suite
  • Refactored and highly modular code that can be utilized also elsehere, especially independence_tests module
  • Added n_jobs to CMIknn to be used in cKDtree
  • Improved doc of CMIsymb
  • Improved RCOT API
  • include_lagzero_parents parameter in return_significant_parents()
  • time_bin_with_mask now has parameter mask instead of sample_selector
  • cut_off is explicitely handled in run_test() in independence_tests.py
  • take a look at the new tutorials to see all functionality
Assets 2
Loading