The in-processing fit of the logistic regression takes about a minute and a half on Healthcare Readmission #1238

Where: faircode/strategies.py fit_in_processing, faircode/benchmark.py.

Current gap: With the 101,766-row Healthcare Readmission audit the S3 ExponentiatedGradient fit with the logistic regression took 98 s in a single run, which is why tests/test_benchmark_convergence.py skips that combination.

Suggested approach: Profile the fit (iterations, solver tolerance, repeated refits) and find what can be cut without changing results, or report the time per combination in the benchmark log.

Done when: the slow test can run that combination, or the time is recorded with the run.