Revamp survival analysis interface

[aGuyLearning]

Fixes #840

This pull request includes refactoring and enhancements to the survival analysis tools, particularly focusing on the Cox Proportional Hazards (CoxPH), Weibull Accelerated Failure Time (Weibull AFT), and Log-Logistic Accelerated Failure Time (Log-Logistic AFT) models. The changes improve the flexibility and usability of these models by adding new parameters and restructuring the code.

Refactoring and Enhancements:

• Code Refactoring:

  • Split the _regression_model function into two separate functions: _regression_model_data_frame_preparation and _regression_model_populate_adata.

• Cox Proportional Hazards Model:

  • Added new parameters to the cox_ph function to provide more control over the model fitting process, such as inplace, key_added_prefix, alpha, label, baseline_estimation_method, penalizer, l1_ratio, strata, n_baseline_knots, knots, breakpoints, weights_col, cluster_col, robust, formula, batch_mode, show_progress, initial_point, and fit_options.

• Weibull Accelerated Failure Time Model:

  • Enhanced the weibull_aft function with additional parameters similar to those added to the cox_ph function.

• Log-Logistic Accelerated Failure Time Model:

  • Updated the log_logistic_aft function to include new parameters.

• Testing Adjustments:

  • Modified the _sa_func_test method in tests/tools/test_sa.py to accommodate the updated function signatures, ensuring that tests pass with the new parameters.

[aGuyLearning]

This is not a finished PR, but a starting point for a discussion. The Weibull_aft and log_logistic_aft model summaries include multi index rows. The question is, if we can reduce them and how that works with the limitations of the adata.var.

We were hoping to include all the summary data into the adata, so as to have a similar style as scanpy.

[aGuyLearning]

• Add model result dataframe to .uns
  • Add key name of .uns to be added to the function arguments ( default: function name )
  • PLOT: Search for function name on default
  • PLOT: Take information from .uns object

Do this for:

• cox_ph
• weibull_aft
• log_logistic_aft

[eroell]

Add model result dataframe to .uns

just a quick write-down of our offline discussion:

indeed, I think this is the best way to go from here if we do want to offer nice plots as request in the linked issue.

It is required that we store the survival analysis results in the adata object to produce corresponding plots with the functional API like ep.pl.<fancy_plot>(adata, ...).

However it can't be stored in adata.var, as in addition to the covariates, the models typically include an intercept term, making the model fit results to be of length len(adata.var.columns) + 1.

storing the summary results of the fitter as a very basic DataFrame in adata.uns is the most straightforward way in this case imo

[eroell]

Allows to pass keywords as requested in #744. Does not finish off this issue though, as ideally also show in a notebook tutorial later what effect regularization has.

[eroell]

From my side good. @Zethson, request your review here as this a quite specific choice we're making here.
See comment my comment a quick write-down of our offline discussion above about the rationale behind this choice.
If you have objections, we're interested to hear them

[Zethson]

Great! Just minor points.