Fix merge_ebms missing hyperparameters (#576)

[ugbotueferhire]

---

Description

This PR resolves Issue #576 where calling merge_ebms() produced broken EBM objects.
Because merge_ebms instantiates the merged EBM via __new__(), the __init__() method is bypassed. Consequently, the resulting EBM model lacked all fundamental hyperparameters (e.g., learning_rate, max_bins, outer_bags). This severely violated the scikit-learn estimator contract, causing operations like repr(merged_ebm), merged_ebm.get_params(), and sklearn.base.clone(merged_ebm) to raise an AttributeError.
Changes Made:

1. Dynamic Parameter Initialization: Implemented _initialize_merged_model_params(merged_model, source_models) to dynamically copy hyperparameters from the first source model onto the newly initialized merged model using get_params(deep=False).
2. Neutralize Callbacks: The callback parameter is a stateful training parameter that retains unwanted references post-merge. This is explicitly wiped (set to None) during initialization.
3. Comprehensive Regression Suite: Added 7 robust regression tests to secure both Classifiers and Regressors against future edge-cases.

Code Snippet

The core surgical fix leverages get_params() to respect the scikit-learn constructor variables dynamically:

def _initialize_merged_model_params(merged_model, source_models):
    first_source_model = source_models[0]
    hyperparameters = first_source_model.get_params(deep=False)
    if "callback" in hyperparameters:
        hyperparameters["callback"] = None
    for parameter_name, parameter_value in hyperparameters.items():
        setattr(merged_model, parameter_name, parameter_value)

Testing Strategy

To prove the fix and guarantee zero regressions, a rigorous suite was added to test_merge_ebms.py:

• test_merge_ebms_repr: Exactly reproduces and asserts the AttributeError: 'ExplainableBoostingClassifier' object has no attribute 'cyclic_progress' crash reported in merge_ebm produces broken classifiers #576.
• test_merge_ebms_get_params / test_merge_ebms_sklearn_clone: Asserts strict downstream scikit-learn compliance.
• test_merge_ebms_has_all_hyperparameters: Introspects the explicit __init__ signature to mathematically guarantee no hyperparameters are left undefined.
• test_merge_ebms_predictions_unchanged: Asserts that updating scikit-learn metadata metadata doesn't shift original score distributions natively.

Verification (Test Execution)

✓ ruff passed (0 linting errors)
✓ 100% test coverage tracked on `_merge_ebms.py` added lines
============================= test session starts =============================
platform win32 -- Python 3.13.0, pytest-9.0.2
rootdir: /interpret/python/interpret-core
collected 9 items
python/interpret-core/tests/glassbox/ebm/test_merge_ebms.py 
  ✓ [ 11%] test_merge_ebms
  ✓ [ 22%] test_merge_ebms_multiclass
  ✓ [ 33%] test_merge_ebms_repr 
  ✓ [ 44%] test_merge_ebms_get_params
  ✓ [ 55%] test_merge_ebms_sklearn_clone 
  ✓ [ 66%] test_merge_ebms_has_all_hyperparameters 
  ✓ [ 77%] test_merge_ebms_callback_is_none
  ✓ [ 88%] test_merge_ebms_regressor_repr 
  ✓ [100%] test_merge_ebms_predictions_unchanged
=========================== 9 passed in 40.59s ===========================

[ugbotueferhire]

Hi @paulbkoch,
I've submitted this PR to fix Issue #576. It resolves the bug where merge_ebms() produced models without hyperparameters, causing downstream scikit-learn failures like clone().
I corrected the missing initialization and added 7 comprehensive regression tests. All CI tests and formatting checks are completely green! I'd really appreciate it if you could take a quick look when you have a moment.

[paulbkoch]

Hi @ugbotueferhire -- The code change you made looks good, although this PR is on the develop branch, which is very old. Can you make this change on the main branch.

[ugbotueferhire]

Hi @ugbotueferhire -- The code change you made looks good, although this PR is on the develop branch, which is very old. Can you make this change on the main branch.

alr thank you for the feedback

[ugbotueferhire]

@paulbkoch I've updated to main branch

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 66.39%. Comparing base (412ca5c) to head (d8bc434).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #661      +/-   ##
==========================================
+ Coverage   66.32%   66.39%   +0.06%     
==========================================
  Files          75       75              
  Lines       11581    11589       +8     
==========================================
+ Hits         7681     7694      +13     
+ Misses       3900     3895       -5     

Flag	Coverage Δ
bdist_linux_311_python	66.12% <100.00%> (+0.06%)	⬆️
bdist_linux_312_python	66.12% <100.00%> (+0.08%)	⬆️
bdist_linux_313_python	66.11% <100.00%> (+0.07%)	⬆️
bdist_linux_314_python	66.06% <100.00%> (+0.08%)	⬆️
bdist_linuxarm_311_python	66.14% <100.00%> (+0.08%)	⬆️
bdist_linuxarm_312_python	66.15% <100.00%> (+0.09%)	⬆️
bdist_linuxarm_313_python	66.15% <100.00%> (+0.10%)	⬆️
bdist_linuxarm_314_python	66.06% <100.00%> (+0.07%)	⬆️
bdist_mac_311_python	66.30% <100.00%> (+0.08%)	⬆️
bdist_mac_312_python	66.28% <100.00%> (+0.06%)	⬆️
bdist_mac_313_python	66.28% <100.00%> (+0.08%)	⬆️
bdist_mac_314_python	66.22% <100.00%> (+0.10%)	⬆️
bdist_win_311_python	66.32% <100.00%> (+0.08%)	⬆️
bdist_win_312_python	66.30% <100.00%> (+0.08%)	⬆️
bdist_win_313_python	66.30% <100.00%> (+0.06%)	⬆️
bdist_win_314_python	66.22% <100.00%> (+0.08%)	⬆️
sdist_linux_311_python	66.08% <100.00%> (+0.10%)	⬆️
sdist_linux_312_python	66.07% <100.00%> (+0.08%)	⬆️
sdist_linux_313_python	66.08% <100.00%> (+0.12%)	⬆️
sdist_linux_314_python	65.97% <100.00%> (+0.04%)	⬆️
sdist_linuxarm_311_python	66.07% <100.00%> (+0.09%)	⬆️
sdist_linuxarm_312_python	66.09% <100.00%> (+0.08%)	⬆️
sdist_linuxarm_313_python	66.07% <100.00%> (+0.07%)	⬆️
sdist_linuxarm_314_python	66.01% <100.00%> (+0.09%)	⬆️
sdist_mac_311_python	66.21% <100.00%> (+0.10%)	⬆️
sdist_mac_312_python	66.21% <100.00%> (+0.10%)	⬆️
sdist_mac_313_python	66.21% <100.00%> (+0.10%)	⬆️
sdist_mac_314_python	66.12% <100.00%> (+0.06%)	⬆️
sdist_win_311_python	66.30% <100.00%> (+0.08%)	⬆️
sdist_win_312_python	66.30% <100.00%> (+0.10%)	⬆️
sdist_win_313_python	66.32% <100.00%> (+0.08%)	⬆️
sdist_win_314_python	66.22% <100.00%> (+0.08%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[paulbkoch]

LGTM. The docs failure appears unrelated. Thanks @ugbotueferhire!