0.10.4

• Changed to Cython implementation for the loss functions and impurity measures of CSLogitClassifier, CSBoostClassifier, CSTreeClassifier, and CSForestClassifier. This improves the training time and memory efficiency of these models significantly. Training time speedups observed were up to 300x for CSTreeClassifier and CSForestClassifier, 30x for CSLogitClassifier, and 1.5x for CSBoostClassifier depending on the dataset size and parameters.
• Changed arguments to CSTreeClassifier, CSForestClassifier, and CSBaggingClassifier to be in line with scikit-learn's decision tree and ensemble models.
• CSForestClassifier, and CSBaggingClassifier no longer support stacking combination method. Use StackingClassifier instead for stacking.
• Extracted the construction of the cost matrix into a separate class CostMatrix away from Metric to allow reusing the cost matrix in custom metrics.
• ProfLogitClassifier no longer uses the EMPC metric by default. Users now need to explicitely pass a loss to the model.
• CSLogitClassifier no longer accepts any callable as loss function. Users now need to pass a Metric instance for a custom loss function.
• savings_score and expected_savings_score now accept two more baseline options 'one' and 'zero' to always predict the positive and negative class, respectively.
• Metrics with with the Savings strategy now also accepts baseline options like savings_score and expected_savings_score.
• Models which use a Metric instance as their loss function with the Cost or Savings strategy as their loss function now are pickleable. The MaxProfit strategy will be updated to be pickleable in a future release.
• Models which use a Metric instance as their loss function can now request arguments necessary for the metric to be passed during the fit method through Metadata Routing.
• Fix CSLogitClassifier not properly calculating gradient penalty.
• Fix default values not being properly when using aliases in CostMatrix.
• Fix Metric throwing errors when certain terms cancelled out.