Inconsistent results of induction for survival

[Margaid]

When executing rule induction for survival analysis on the complete dataset, I've encountered inconsistent results between the Python wrapper and the Java RuleKit. I observed variations specifically in the parameters of IBS (Integrated Brier Score), the count of rules, and the number of conditions per rule. The datasets used for this comparison include BMT-ch and mgus, both sourced from the RuleKit/data/contrast-sets/survival repository.

Here is a snippet of my code:

def get_ruleset_stats(model) -> pd.DataFrame:
    tmp = model.parameters.__dict__
    del tmp['_java_object']
    return pd.DataFrame.from_records([{**tmp, **model.stats.__dict__}])

def survival_python_wrapper(x, y):
    clf = SurvivalRules(
        survival_time_attr='survival_time'
    )
    clf.fit(x, y)
    start_time=time.time()
    prediction = clf.predict(x)
    end_time=time.time()

    model_stats = get_ruleset_stats(clf.model)
    display(model_stats)

    ibs=clf.score(x,y)
    prediction_time=end_time-start_time    

    results={'IBS': np.round(ibs,4), 'number of rules': model_stats['rules_count'][0],
            'number of conditions': model_stats['conditions_per_rule'][0]*model_stats['rules_count'][0],
            'model building time': model_stats['time_total_s'][0],
            'prediction time': prediction_time }
    display(results)

[cezary986]

Fixed in version 2.1.18.0