Separate model selection

[simon-hirsch]

This is quite well now.

[Copilot]

Pull Request Overview

This PR separates model selection functionality by introducing a new log‐likelihood based information criterion function and refactoring the online GAMLSS estimator to support different model selection approaches. Key changes include:

• Adding a new function, information_criterion_log_likelihood, to compute criteria based on log‐likelihood.
• Refactoring model selection and beta update functions in online_gamlss.py with new parameters and renaming of iteration variables.
• Updating convergence condition checks and internal data structures to support the new model selection logic.

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
src/rolch/information_criteria.py	Adds a log-likelihood based IC function to support additional criteria.
src/rolch/estimators/online_gamlss.py	Refactors model selection methods and updates convergence conditions.

Comments suppressed due to low confidence (1)

src/rolch/estimators/online_gamlss.py:823

• The updated convergence condition in the inner iteration now only checks 'it_inner > 1', which is not equivalent to the former cumulative condition (iteration_inner + iteration_outer >= 2). Please review this logic change to ensure it meets the intended convergence criteria.

if (abs(olddv - dv) <= self.abs_tol_inner) & (it_inner > 1):

[Copilot]

Pull Request Overview

This PR refactors the information criterion functionality by replacing the old function‐based implementation with a new InformationCriterion class and updates related references across the codebase. Key changes include:

• Converting the information criterion calculation from a function to a class with methods from_rss and from_ll.
• Updating dependent modules (online_linear_model, init.py, and docs) to use the new class interface.
• Removing the previous select_best_model_by_information_criterion function.

Reviewed Changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated no comments.

File	Description
src/rolch/information_criteria.py	Replaces function with class implementation and updates docs.
src/rolch/estimators/online_linear_model.py	Adapts model selection to use the new InformationCriterion class.
src/rolch/init.py	Removes legacy function references and updates exports.
docs/model_selection.md	Updates documentation reference to the new class.

Comments suppressed due to low confidence (2)

src/rolch/information_criteria.py:6

• Typographical error: 'Calcuate' should be corrected to 'Calculate'.

    """Calcuate the information criteria.

src/rolch/information_criteria.py:14

• The documentation table includes the 'max' criterion, but the init method does not accept 'max'. Please update the documentation or adjust the allowed criteria to resolve this inconsistency.

| "max" | Select the largest model           |                          |

[BerriJ]

I did the following changes to this PR:

• Removed: information_criterion, select_best_model_by_information_criterion and information_criterion_log_likelihood
• Added a new class: InformationCriterion

Now we use InformationCriterion whenever we need to calculate an ic there are two methods from_rss and from_ll to calculate the ic from the residual sum of squares or from the log likelihood, respectively.
Also I removed select_best_model_by_information_criterion as it was basically a wrapper around best_ic = np.argmin(ic). Instead, we now use best_ic = np.argmin(ic) directly in gamlss and linearmodel.

[simon-hirsch]

Now we use InformationCriterion whenever we need to calculate an ic there are two methods from_rss and from_ll to calculate the ic from the residual sum of squares or from the log likelihood, respectively.
Also I removed select_best_model_by_information_criterion as it was basically a wrapper around best_ic = np.argmin(ic). Instead, we now use best_ic = np.argmin(ic) directly in gamlss and linearmodel.

What about EBIC and EFIC, see here: https://ieeexplore.ieee.org/abstract/document/10146465

[Copilot]

Pull Request Overview

This pull request refactors the model selection functionality by replacing the standalone functions with a dedicated InformationCriterion class.

• Introduces the InformationCriterion class with methods for calculating criteria from RSS and log-likelihood values.
• Updates references in estimator modules and documentation to use the new class.

Reviewed Changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 1 comment.

File	Description
src/rolch/information_criteria.py	Refactored information criterion functions into a class with methods from_rss and from_ll.
src/rolch/estimators/online_linear_model.py	Updated model selection to use the InformationCriterion class instead of the removed functions.
src/rolch/init.py	Removed deprecated function exports and exposes the new InformationCriterion class.
docs/model_selection.md	Updated API documentation to reflect the change to the InformationCriterion class.

[BerriJ]

Lookgs good.