Clust-learn

Release notes:

**Clustering**

- Optimal clustering computation with other metrics than `inertia` uses metrics scores for clusters >1
- `Silhouette` and `Calinski and Harabasz` scores are concave increasing (instead of convex decreasing)

**Classifier**

- `plot_shap_importances()` : pandas.DataFrame.append was deprecated in version 1.4.0. It has been replaced by pandas.concat

Release notes:

**Classifier**
- Variable relationships are computed before balancing classes

Release notes:

**Clustering**:

- Fix for cluster distribution using a categorical variable and weights when one category is not present for one cluster (method describe_clusters_cat())
- Allow category order in plot_cat_distribution_by_cluster()

**Classification**:

- Allow for class balancing
- min_samples_leaf for feature selection adjusted considering number of classes
- Allow for any class coding (before it was compulsory to use integers starting at 0)

New release includes:

- MCA with own Benzecri and Greenacre correction for inertia calculation because `prince` no longer applies it + other adjustments due to latest changes in `prince`
- Minimum required `pingouin` version upgrades because of incompatibilities with newer versions of pandas
- Improvements in visualizations
- Updated notebook guides

New release includes:

- Clustering with sample weights
- Clustering with any algorithm that meets scikit-learn standard
- A broader range of package versions in requirements

New release includes:

- Fix in dimensionality reduction when either numerical or categorical variable list is `None`