Pgmpy

Added
1. `init_cpds` argument to `ExpecattionMaximiation.get_parameters` to specify initialization values.
2. BeliefPropagation with message passing for Factor Graphs.
3. Marginal Inference for undirected graphs.

Fixed
1. Incompatibality with networkx==3.2.
2. `CausalInference.get_minimal_adjustment_set` to accept string variable names.
3. Bug in EM when latent varaibles are present.
4. `compat_fns.copy` to consider the case when int or float is passed.
5. Fixes issue with `BayesianNetwork.fit_update` when running with CUDA backend.

Changed
1. Documentation Updates
2. Optimizations for Hill Climb Search algorithm.
3. Tests shutdown parallel workers in teardown.
4. Removes the `complete_samples_only` argument from `BaseEstimator.state_counts`.
5. Default number of cores to use changed to 1 for parameter estimation methods.

Added
1. Added support for python 3.11.
2. Adds `DAG.to_graphviz` and `PDAG.to_graphviz` methods to convert model to graphviz objects.
3. Adds pytorch as an alternative backend.
4. Adds unicode support for BIFReader.

Fixed
1. Warnings use a logger instance.
2. Fixes documentation.
3. Fixes variables name arguments for `CausalInference.get_minimal_adjustment_set`

Changed
1. Adds argument to specify samples for ApproxInference.
2. Memory optimizations for computing structure scores.
3. Switches joblib backed to loky.
4. Runtime optimizations for sampling.
5. Runtime optimizations for Variable Elimination.
6. All config variables moved to `pgmpy.global_vars`.

Added
1. BIFReader made compatible with the output of PyAgrum
2. Support for all available CI tests in PC algorithm.
3. References for read/write file formats.

Removed
1. Removes `DAG.to_pdag` method.

Changed
1. Fixes for ApproxInference for DBNs.
2. Make `xml.etree` the default parser instead of using lxml.

Added
1. AIC score metric from score based structure learning.
2. Adds support for NET (HUGIN) file format.
3. Adds argument reindex to `state_counts` method.

Fixed
1. Bug in GibbsSampling when sampling from Bayesian Networks.
2. Fix seed for all simulation methods.
3. Memory leaks when using `lru_cache`.

Changed
1. Caching disabled for computing state name counts during structure learning.
2. Pre-computation for sampling methods are optimized.

Added
1. `BayesianNetwork.get_state_probability` method to compute the probability of a given evidence.
2. `BayesianEstimator.estimate_cpd` accepts weighted datasets.

Fixed
1. Fixes bug in `CausalInference.estimate_ate` with front-door criterion.
2. Fixes inference bugs when variable has a single state.

Added
1. `BayesianNetwork.get_random_cpds` method to randomly parameterize a network structure.
2. Faster Variable Elimination using tensor contraction.
3. `factors.factor_sum_product` method for faster sum-product operations using tensor contraction.

Fixed
1. Bug in `DynamicBayesianNetwork.initialize_initial_state`. 1564
2. Bug in `factors.factor_product`. 1565

Changed
1. Runtime improvements in `DiscreteFactor.marginalize` and `DiscreteFactor.copy` methods.