# Source

# ITE / CHANGELOG.txt

# CHANGELOG.txt

+-Initialization and estimation of high-level information theoretical estimators (H/I/D/A/C/K): 'eval' changed to 'function handles' to speed up computations; see '<X>_initialization.m' and '<X>_estimation.m', where <X>='H'/'I'/'D'/'A'/'C'/'K'.

+-Cost object initialization now allows setting field values (alpha, number of kNN-s, ...) through its argument. This makes ITE more user-friendly. It also enables overriding default values and is useful in meta estimators for parameter inheritence. See 'post_initialization.m'; '<X>_initialization.m' and '<X><cost_name>_initialization.m', where <X>='H'/'I'/'D'/'A'/'C'/'K'. For the usage in meta estimators, see

+'HTsallis_HRenyi_initialization.m', 'IRenyi_DRenyi_initialization.m', 'ITsallis_DTsallis_initialization.m', 'IRenyi_HRenyi_initialization.m', 'DJensenRenyi_HRenyi_initialization.m', 'DJensenTsallis_HTsallis_initialization.m', 'DsymBregman_DBregman_initialization.m', 'KJS_DJS_initialization.m'. For further details, see doc: Section 3.3.2.

+-Quick tests introduced: consistency of the estimators, positive semi-definiteness of Gram matrices determined by distribution kernels, image registration; see directory 'quick_tests', doc: Section 5, Section F.

+-Typo discovered and corrected: 'HShannon_DKL_N_estimation.m': 'Y_normal = ... + m' -> 'Y_normal = ... + repmat(m,1,num_of_samples)' (typo for d>1).

+-Comment on the relation between mutual information and total correlation (or multi-information): added [doc + code (see 'IShannon_HShannon_initialization.m')].

+-To ease the usage of ITE 'ITE_add_to_path.m' and 'ITE_remove_from_path.m': added. These functions make it possible to add/remove the ITE code directory to/from the Matlab/Octave PATH.

+-Explicit instructions for the installation of the toolbox added to 'ITE_install.m' when no compilation is carried out.

+-In pathological cases (quite small dimension and large number of samples) the estimated Renyi additive constant (see 'estimate_HRenyi_constant.m') can be Inf. This case is now handled by introducing 'replace_Infs_with_max.m', see 'compute_length_HRenyi_kNN_1tok.m', 'compute_length_HRenyi_kNN_S.m'.

+- 'MatlabBGL' and the associated GSF based Renyi entropy estimator ('HRenyi_GSF_initialization.m', HRenyi_GSF_estimation.m', 'compute_length_HRenyi_GSF.m'): deleted (self-containedness improved + space considerations); 'compute_MST.m','HRenyi_MST_initialization.m', 'HRenyi_MST_estimation.m': modified accordingly.

+-The embedded 'sqdistance.m' (squared Euclidean distance computation) function could give some small, but negative values in the diagonal of 'sqdistance(Y)'; corrected.

+-ITE_install.m: made to be more user-friendly; detection of the already (i) deleted 'ann_wrapperO'/'ann_wrapperM' directory, (ii) downloaded ARfit package: added.

-Probability product kernel estimation based on k-nearest neighbors: added; see 'KPP_kNN_k_initialization.m' and 'KPP_kNN_k_estimation.m'.

-Jensen-Shannon kernel estimation: added; see 'KJS_DJS_initialization.m' and 'KJS_DJS_estimation.m'.