Tags

* aGrUM
  * Added `gum::Potential<GUM_SCALAR>::memoryFootprint` and `gum::IBayesNet::memoryFootprint` (thanks to Benjamin Datko).
  * Fixed a bug when subscripting `gum::DiscretizedVariable` with the (last) interval (thanks to Gaspard Ducamp).
  * Removed gcc warnings (psabi, false positive uninitialized).

* pyAgrum
  * Added `pyAgrum.Potential.memoryFootprint` and `pyAgrum.BayesNet.memoryFootprint` (thanks to Benjamin Datko).
  * Fixed a bug when subscripting `gum::DiscretizedVariable` with the (last) interval (thanks to Gaspard Ducamp).
  * Fixed a bug in `pyAgrum.skbn.BNDiscretizer` for `NoDiscretization` method with given `paramDiscretization`.
  * Moved function `pyAgrum.lib.export.toFast` to method `pyAgrum.{BayesNet|InfluenceDiagram|MRF}.toFast()`.
  * Moved `pyAgrum.skbn.BNDiscretizer` to `pyAgrum.lib.discretizer`.
  * Worked on documentation for CTBN
  * Fixed `fontcolor` configuration for causal model.

We continue to add (full python) experimental models. In 1.15, we introduce the CTBN model (Continuous Time Bayesian Network) featuring, as usual, modelization and representation, inference (exact and sampling) and a learned algorithm.

* aGrUM
  * Added `gum::NodeId gum::EssentialGraph::idFromName(const std::string& name)` and  `const std::string&  
    gum::EssentialGraph::nameFromId(gum::NodeId node)`. 

* pyAgrum
  * Added `pyAgrum.EssentialGraph.idFromName(str)->int` and `pyAgrum.EssentialGraph.nameFromId(int)->str` 
  * Improved documentation of `pyAgrum.lib.explain`
  * Better `pyAgrum.clg.CLG.toDot()` and `pyAgrum.clg.CLG._repr_html()`.
  * New model Continuous Time Bayesian Network `pyAgrum.ctbn`.
  * Formatted and adjustments in `pyAgrum.ctbn`.
  * Updated documentations for python experimental models notebooks.
  * Updated thumbnails for python experimental models notebooks.
  * Added serialization (pickle) for CLG and CTBN (consistent with other models in pyAgrum).
  * Improved `pyAgrum.lib.utils.{apply_}dot_layout`
  * Added `pyAgrum.lib.utils.async_html2image` for exported HTML as png or pdf (notably for `pyAgrum.lib.notebook.getSideBySide` and `pyAgrum.lib.notebook.getPotential`).

* aGrUM
  * Fixing compilation errors for `clang18` (thanks to @yurivict).

* pyAgrum
  * Fixing python version for `readthedocs`.

* aGrUM
  * `gum::Potential::marg{Xyz}{In|Out}` are renamed `gum::Potental::{xyz}{In|Out}` to cut down on some verbosity.
  (for instance : `p.margSumOut` becomes `p.sumOut`)
  * Better error messages for `gum::BNLearner<GUM_DATA>`

* pyAgrum
  * Dropping support of python 3.8 (following [NEP29](https://numpy.org/neps/nep-0029-deprecation_policy.html)).
  * Fixing some memory leaks (in special cases) found by Christophe Gonzales
  * new `pyAgrum.lib.explain.generalizedMarkovBlanket{Names}` and documentation.
  * `pyAgrum.Potential.marg{Xyz}{In|Out}` are renamed `pyAgrum.Potental.{xyz}{In|Out}` to cut down on some verbosity.
      (for instance : `p.margSumOut` becomes `p.sumOut`)
  * Better error messages for `pyAgrum.BNLearner`
  * Adding new python's experimental model : Conditional Linear Gaussian (`pyAgrum.clg.*`) : model, learning and inference.

* aGrUM
  * Better (compact) format for numerical labels.
  * Better uniformly distributed random discrete distribution.

* pyAgrum
  * Better (compact) format for numerical labels.
  * Better uniformly distributed random discrete distribution.
  * Continue to implement some automatic `pyAgrum.Potential` filling tools with `pyAgrum.Potential.fillFromExpression`.
  * Working on better plots for probability+utility
  * Speeding up `gum.BayesNet.addArcs()` with *jit* memory allocation
  * Slight adjustments to histogram display.
  * Fixing documentation of `gum.BNLearner.mutualInformation`.
  * Fixing (deprecated) arguments in `pyAgrum.skbn.BNClassifier`'s methods.

This release is mainly about documentation.

* pyAgrum
  * Fixing last references to LIP6's notebooks instead of rtd's notebooks (thanks to Benjamin Datko).
  * Adding 'Book of Why's notebooks to readthedoc's documentation.
  * Updating Notebooks and thumbnails.
  * Improving documentation coverage for pyAgrum class methods (from 86.19% to 87.05%)
  * `gum.BNLearner.learnEssentialGraph` for all learning algorithm.
  * Better wrapping of `pyAgrum.BNLearner.setPossibleEdges()`.

* Internal
  * New servers for (linux and windows) CI !
* aGrUM
  * Fast syntax for empirical `gum::DiscretizedVariable`: name suffixed by '+'.
  * Reorganizing a bit `gum::DiscreteVariable`'s hierarchy.
  * Working on quasi-continuous (numerical discrete variable with large domain size) :
    * Logical operators  (`&`,`|`, `~`) to combine evidence (monodimensionnal `gum::Potential` not forced to sum to 1).
    * New `gum::Potential` and `gum::BayesNet` methods to generate numerical evidence : `evEq`, `evIn`, `evLt`, `evGt`.
    * Fast syntax now accepts any type as default (and not only the size of the domain).

* pyAgrum
  * Fixing a bug in `pyAgrum.lib.bn2scores.
  * Fast syntax for empirical `pyAgrum.DiscretizedVariable`: name suffixed by '+'.
  * Reorganizing `pyAgrum.lib._colors.py` and `pyAgrum.lib.utils.py`.
  * Changing a default for `pyAgrum.Instantiation.todict`.
  * Improving (a lot) histograms for `pyAgrum.DiscretizedVariable`.
  * Working on quasi-continuous (numerical discrete variable with large domain size) :
    * Logical operators (`&`,`|`, `~`) to combine evidence (monodimensionnal `pyAgrum.Potential` not forced to sum to 1).
    * New `pyAgrum.Potential` and `pyAgrum.BayesNet` methods to generate numerical evidence : `evEq`, `evIn`, `evLt`, `evGt`.
    * Fast syntax now accepts any type as default (and not only the size of the domain).
    * `pyAgrum.Potential` accepted as evidence for all inference.
    * Long-awaited `pyAgrum.Potential.fillFrom{Function|Distribution}`.  
  * Adding `pyAgrum.config.{push|pop}`.
  * Removing obsolete code for `python2.7`.
  * Removing some oldest deprecationWarning.
  * Adding '__iter__' for graphs, graphical models, variable and potentials.

* aGrUM
  * Adding a new job during CI for sphinx documentation.
  * Fixing a typo in `gum::MeeRules::propagate(s)`.
* pyAgrum
  * Fixing a typo in `pyAgrum.MeekRules.propagate(s)`.
  * Adding new rules for 'python typification' of generated files.
  * Fixing and updating documentations and notebooks.

* aGrUM
  * Adding a new class `gum::MeekRules` to complete a learned PDAG to CPADG or BN.
  * Preventing 'infinities' in the definition of numerical discrete variable.
  * Adding a new constructor for `gum::PDAG` from`gum::MixedGraph`.
  * Better (stricter) `operator==` between `gum::DiscreteVariable`.
  * Adding a new static constructor `gum::{Undi|Di}Graph::completeGraph(int n)`.
  * Adding `closestIndex()` and `closestLabel()` methods for all numerical `gum::DiscreteVariable`.
  * Fixing an exception raised in inferences due to the removal of uninformative potentials containing only barren variables
  * new syntax for `fastVariable` : `[a:b:n]` = `gum.Discretiezd` with n intervalls from a to b.

* pyAgrum
  * Pickle for graphs and graphicals models !
  * Preventing 'infinities' in the definition of numerical discrete variable.
  * Adding a new class `pyAgrum.MeekRules` to complete a learned PDAG to CPADG or BN.
  * Better (stricter)  `__eq__` and `__ne_`_ between `pyAgrum.DiscreteVariable`.
  * Adding a new constructor for `pyAgrum.PDAG` from`pyAgrum.MixedGraph`.
  * Adding a new `expert` discretization method to `pyAgrum.skbn.BNDiscretizer` (thanks to Ahmed Mabrouk).
  * `pyAgrum.skbn.BNDiscretizer` can now take csv filename for input.
  * Adding `pyAgrum.fastGraph` to create `pyAgrum.{Di|Undi|Mixed}Graph`.
  * Dataframes can now come from `pandas` or `polars` (thanks to Mads Linksou).
  * Adding a `compact` mode for histogram for discretized variable controlled by `histogram_mode`
  `histogram_epsilon` in `gum.config` (thanks to Gaspard Ducamp).
  * Adding a new static constructor `pyAgrum.{Undi|Di}Graph::completeGraph(int n)`.
  * Adding subscripting to graphical models : `g[name]` gives now a direct access to the `pyAgrum.DiscreteVariable` by its
name.
  * New syntax for `pyAgrum.fast{Var|BN|ID|MRF}` : `[a:b:n]` = `pyAgrum.DiscretizedVariable` with n intervalls from a to b.
  * Adding `closestIndex()` and `closestLabel()` methods for all numerical `pyAgrum.DiscreteVariable`.
  * Adding `g.adjacencyMatrix()` for all graphs and graphical models (thanks to Kenneth Lee).
  * Adding new `bgcolor` parameter (None by default) for `showROC/PR` in `pyAgrum.lib.bn2roc` and `pyAgrum.skbn` (thanks to Clara Charon).
  * Adding optional `dialect` parameter to `pyAgrum.lib.bn2roc.computeScores`.
  * new syntax for `fastVariable` : `[a:b:n]` = `pyAgrum.Discretiezd` with n intervalls from a to b.

* Internal
  * Removing some warnings in CMakeLists.txt
  * Refactoring CI, cleaning wheelhouse

* aGrUM
  * Adding `gum::LazyPropagation::mpe{Log2Posterior}` to compute the Most Probable Explanation with some evidence.
  * Improving MIIC algorithm and application of Meek Rules in ordre to to better solve (partially) oriented cycle problems.
  * Fixing a bug in `gum::BNLearner` : EM parameter estimation bug due to a cache used in record counters.
  * Fixing a bug in `gum::GibbsSampling` : some evidence may not be taken into account (thanks to Micromix@discord).
  * `gum::PDAG` must accept partially directed cycle (thanks to Kenneth Lee).

* pyAgrum
  * Updating and adding 3.12 in metadatas.
  * Adding `pyAgrum.LazyPropagation.mpe{Log2Posterior}` to compute the Most Probable Explanation with some evidence.
  * Removing reference to `sys.log` and to `setuptools`.
  * Removing legacy setuptools related files.
  * Fixing a bug in `pyAgrum.BNLearner` : EM parameter estimation bug due to a cache used in record counters.
  * Fixing a bug in `pyAgrum.GibbsSampling` : some evidence may not be taken into account (thanks to Micromix@discord).
  * Fixing a bug in frontdoor detection in a causal model  with multiple connected components.
  * Fixing some regexp expressions.
  * `pyAgrum.PDAG` must accept partially directed cycle (thanks to Kenneth Lee).

* aGrUM
  * add contraints to `MIIC` BN learning algorithm.
  * `MIIC` is now the default for `gum::BNLearner`.
  * Fixing some minor bugs in `gum::BNLearner`.
  * Removing old `3off2` BN learning algorithm (use MIIC instead).
  * adding `pyAgrum::MIIC::learnPDAG` and give access to this method via `gum::BNLearner`.

* pyAgrum
  * package for python 3.12 !
  * New configuration for specification of graphviz engine and rankdir .
  * Constraint for `MIIC` algorithm in `pyAgrum.BNLearner`.
  * `MIIC` is now the default for `pyAgrum.BNLearner`.
  * Fixing minor bugs in `pyAgrum.BNLearner`.
  * new method `pyAgrum.BNLearner.learnPDAG`.
  * Back to classical representation of do() operator in `pyAgrum.causal`.
  * Examples from the Book Of Why (Pearl) are now in the sphinx documentation.
  * The name of a loaded BN should not include the complete path but only the name of the file (thanks to Jules Cassan).

* aGrUM
    * Learning algorithm `gum::learning::MIIC` can use the weighted databases.
    * Internal improvements for `act` tool, `cmake` and compilers (`clang`).
  
* pyAgrum
    * New visualisation for `gum::DiscretizedVariable` + new config to select this visualisation.
    * `pyAgrum.BNLearner` can use now the weighted databases for all learning algorithms.
    * Documentation improvements.
    * `pyAgrum.lib.bn2roc`
        * adding new functions `get{ROC|PR}points()`.
        * accepting `pandas.DataFrame` as data source (`datasrc`).
        * adding Fbeta (beta!=1) scores to bn2roc.
        * adding F-Beta threshold on ROC and PR curves.
        * `bn2roc` functions now force many parameters to be keyword-arguments in order to prevent the risk of mixing arguments.
        * adding new functions `anim{ROC|PR}`.
    * `pyAgrum.skbn.Discretizer` can propose a set of labels (that includes the labels from the database) when `"NoDiscretization"` is selected. (see tutorial `52-Classifier_Discretizer`).

* aGrUM
    * Removing GCC warnings about overloaded virtual methods.
    * Fixing a bug in jointree inference in rare cases (when an irrelevant soft evidence is added).
  
* pyAgrum
    * Fixing a bug in jointree inference in rare cases (when an irrelevant soft evidence is added).
    * Removing deprecated methods from pyAgrum<1.1.0.
    * Adding new syntax for set (conditional) marginal with dict : `p[{"A":0,"C":1}]={"no":0.3,"yes":0.7}`.
    * small updates for pyAgrum's documentation.

* aGrUM
  * Fixing a bug (still) in `gum::PDAG::cSeparation` (thanks to Keneth Lee).
  * Fixing a bug in `gum::BIFXMLIdWriter::save`.
  * typos in `gum::PDAG::toDot()`.

* pyAgrum
  * Fixing a bug in `pyAgrum/{load|save}ID`.
  * Wrapping `gum::PDAG::moralizedAncestralGraph()`.
  * Adding a warning when importing `pyAgrum.lib.notebook` if graphviz is not installed.

* aGrUM
  * Bugfixes in `gum::PDAG::cSeparation` (thanks to Keneth Lee).
  * Using 'table' instead of 'default' for marginal cpt in `gum::BIFWriter`.
  * Adding `std::cout<<(gum::Timer& t)`

* pyAgrum
  * Updating build image for rtd.
  * Improving documentations.
  * Documentation for pyAgrum.InformationTheory.
  * Uniformizing argume `cmap` to `cmapNode` in `pyAgrum.lib.notebook`.
  * Adding beeswarmplot for shapley values figures in `pyAgrum.lib.explain`.

* aGrUM
  * Adding new class `gum::InfomationTheory` that regroups every information theory concepts (entropies, conditional entropies and mutual information, ...) with only one inference, which allows also conditioned information theory concepts.  
  * Adding `gum::Potential::expectedValue()`
  * Adding the alternative *fast* syntax for NumericalDiscreteVariable : `{1.5:3.5:3}` meaning from 1.5 to 3.5 in 3 steps.
  * Updating `GraphicalModel::exists(const std::string&)` : return false instead of throwing an exception.
  * Adding `gum::graphicalModels::variables()` and a new alias for `gum::VariableSet`.
  * Fixing a bug in `gum::ShaferShenoyInference` for joint posterior when some nodes in the target received hard evidence.
  * Improving error message w.r.t `gum::*::jointPosterior`.
  * Adding mingw as a target for CI.
  * Fixing dangling pointers in testsuite, only detected by MSVC.
  * Optimized `gum::PDAG::cSeparation`
  * Renaming `gum::Set::isProper{Sub|Super}Set` to `gum::Set::isStrict{Sub|Super}Set`.

* pyAgrum
  * swig>4.1 is now needed.
  * Enabling installation with poetry (thanks to Lorenzo Conti)
  * Adding new class `pyAgrum.InfomationTheory` that regroups every information theory concepts (entropies, conditional entropies and mutual information, ...) with only one inference, which allows also conditioned information theory concepts.
  * Adding `pyAgrum.Potential.expectedValue()`.
  * Restructuring a bit the sphinx documentation.
  * Down-exporting the read-only API for every types of discrete Variable to `pyAgrum.DiscreteVariable`. (see `VariablesTestSuite.testExportDerivedReadOnlyAPIforDiscreteVariable`)
  * Updating notebooks to the new down-exported read-only API of `pyAgrum.DiscreteVariable`.
  * Adding the alternative *fast* syntax for NumericalDiscreteVariable : `{1.5:3.5:3}` meaning from 1.5 to 3.5 in 3 steps.
  * Updating docs for new *fast* syntax.
  * Updating `GraphicalModel::exists(const std::string&)` : return false instead of throw an exception.
  * Initiating a new notebook about the interaction with ipywidget
  * Optimized and correctly wrapped `gum::PDAG::cSeparation` (thanks to Kenneth Lee).
  * New functions `pyAgrum.lib.explain.nestedMarkovBlankets` and `pyAgrum.lib.explain.nestedMarkovBlanketsNames`.

* aGrUM
  * Adding `gum::{AllDiscreteVariable}.toFast()` to write the variable in *fast* syntax.
  * Fixing a bug in BN's random generators : `domain_size` incremented by 1 (thanks to Kenneth Lee).
  * Fixing a bug in *fast* syntax : `'A[n]' with n<2`.

* pyAgrum
  * Adding `pyAgrum.{AllDiscreteVariable}.toFast()` to write the variable in *fast* syntax.
  * Fixing a bug in BN's random generators (e.g. `pyAgrum.randomBN`) : `domain_size` incremented by 1 (thanks to Kenneth Lee).
  * Fixing a bug in *fast* syntax : `'A[n]' with n<2`.
  * Modifying (lightly) the structure for documentation.

* aGrUM
  * Add `gum::Potential::sign` and `gum::Potential::new_sign` 
  * Bug fix in causalImpact found by Musfiqur Rahman
  * Generalization and rationalization of access of aGrUM's generator of pseudo-random values
  * Adding `gum::Timer::toString()`.
  * Improving BN random generator.
  * get rid of `sprintf` as much as possible.
  * Fixing random seed.
  * Fixing `gum::Potential::random()` : not a distribution.
  * Updating *Fast* syntax : enumerating consecutive integers (e.g `{1|2|3|4}`) creates a `gum::RangeVariable[1,4]` rather than an `gum::IntegerVariable{1|2|3|4}`.
* pyAgrum
  * Add `pyAgrum.Potential.sign` and `pyAgrum.Potential.new_sign` 
  * Bug fix in causalImpact found by Musfiqur Rahman
  * Adding `pyAgrum.Timer.__str()__`.
  * Improving BN random generator.
  * Fixing random seed.
  * Fixing `pyAgrum.Potential.random()` : not a distribution.
  * Adding access to `gum.fastVariable()`.
  * Updating *Fast* syntax : enumerating consecutive integers (e.g `{1|2|3|4}`) creates a `pyAgrum.RangeVariable[1,4]` rather than an `pyAgrum.IntegerVariable{1|2|3|4}`.
  * add a new module to export in different formats (for now just GML).

Quick fix for the configuration of pyAgrum documentation generation by readthedoc.

The main points of this tag is the renaming of *Markov network* (`MarkovNet`, `MN`) for the better known *Markov Random Field* (`MarkovRandomField`, `MRF`) and a new reader/writer of XDSL format (Genie/Smile) for Bayesian networks. Other improvements and corrections have naturally also been made.

* aGrUM
  * Renaming `gum::MarkovNet` to `gum::MarkovRandomField`. Renaming `gum::*MN*` to `gum::*MRF*` when necessary.
  * Fixing glitches and bugs induced or revealed by `gum::MarkovNet`->`gum::MarkovRandomField`.
  * new `XDSL` Reader/writer for Bayesian network.
  * Renaming `gum::Learning::BNLearner::learnMixedStructure` to `gum::Learning::BNLearner::learnPDAG`
  * working on documentation : better rendering for doxygen pages.
  * Renaming `gum::dSeparation` to `gum::dSeparationAlgorithm`.

* pyAgrum
  * Renaming `pyAgrum.MarkovNet` to `pyAgrum.MarkovRandomField`. Renaming `pyAgrum.*MN*` to `pyAgrum.*MRF*` when necessary.
  * new `XDSL` Reader/writer for Bayesian network.
  * Renaming `pyAgrum.BNLearner.learnMixedStructure()` to `pyAgrum.BNLearner.learnPDAG()`.
  * For figure containing nodes drawn by matplotlib (e.g. inference), use the same font for all nodes (default from matplotlib) (thanks to Jonathon Blackford).
  * Working on documentation : better rendering for readthedocs pages, improved structuration, new thumbnail image for some notebooks/tutorials.
  * Significant improvement of the documentation coverage. 
  * Improving `gum.DiscreteVariable.to[typeOfVariable]`, renaming as `gum.DiscreteVariable.as[typeOfVariable]` and adding documentations.