Convergence problem in SparseLinearDML

Hi,
when training a SparseLinearDML instance, I get the log below, and then the last line repeating for a couple hundred times (overnight), finally resulting in a divide-by-zero and nans.

Identical setup using LinearDML works fine.

Any suggestions on fixing this?

BTW linear_first_stages must be False for this run as else it runs out of RAM (tries to allocate a TiB or something like that)
Thanks a lot!

fitting backdoor.econml.dml.SparseLinearDML {'init_params': {'model_t': DummyClassifier(), 'model_y': <auto_causality.utils.AutoMLWrapper object at 0x000001AB5DE50370>, 'discrete_treatment': True, 'linear_first_stages': False}, 'fit_params': {}}
calling AutoML fit method with  {'time_budget': 300, 'verbose': 1, 'task': 'regression', 'n_jobs': 11, 'pred_time_limit': 1e-05}
calling AutoML fit method with  {'time_budget': 300, 'verbose': 1, 'task': 'regression', 'n_jobs': 11, 'pred_time_limit': 1e-05}
Objective did not converge. You might want to increase the number of iterations. Duality gap: 0.0, tolerance: 0.0
Objective did not converge. You might want to increase the number of iterations. Duality gap: 0.0, tolerance: 0.0
Objective did not converge. You might want to increase the number of iterations. Duality gap: 0.0, tolerance: 0.0
(a couple hundred more like this)
Objective did not converge. You might want to increase the number of iterations. Duality gap: 0.0, tolerance: 0.0
Objective did not converge. You might want to increase the number of iterations. Duality gap: 0.0, tolerance: 0.0
divide by zero encountered in true_divide
invalid value encountered in matmul
The final model has a nonzero intercept for at least one outcome; it will be subtracted, but consider fitting a model without an intercept if possible.
Increasing the treatment variable(s) [RECEIVED_CAMPAIGN] from [0] to [1] causes an increase of nan in the expected value of the outcome [VOL_GBP_AFTER_90D], over the data distribution/population represented by the dataset.
manual mean impact: nan nan std: nan
Traceback (most recent call last):
  File "C:\Program Files\JetBrains\PyCharm 2021.1.2\plugins\python\helpers\pydev\pydevd.py", line 1483, in _exec
    pydev_imports.execfile(file, globals, locals)  # execute the script
  File "C:\Program Files\JetBrains\PyCharm 2021.1.2\plugins\python\helpers\pydev\_pydev_imps\_pydev_execfile.py", line 18, in execfile
    exec(compile(contents+"\n", file, 'exec'), glob, loc)
  File "C:/Users/egor.kraev/Transferwise/causality/notebooks/CRM_impact.py", line 243, in <module>
    intrp.interpret(est_obj, X_train)
  File "C:\Users\egor.kraev\.conda\envs\generic3.9\lib\site-packages\econml\cate_interpreter\_interpreters.py", line 193, in interpret
    self.tree_model_.fit(X, y_pred.reshape((y_pred.shape[0], -1)))
  File "C:\Users\egor.kraev\.conda\envs\generic3.9\lib\site-packages\sklearn\tree\_classes.py", line 1252, in fit
    super().fit(
  File "C:\Users\egor.kraev\.conda\envs\generic3.9\lib\site-packages\sklearn\tree\_classes.py", line 157, in fit
    X, y = self._validate_data(X, y,
  File "C:\Users\egor.kraev\.conda\envs\generic3.9\lib\site-packages\sklearn\base.py", line 431, in _validate_data
    y = check_array(y, **check_y_params)
  File "C:\Users\egor.kraev\.conda\envs\generic3.9\lib\site-packages\sklearn\utils\validation.py", line 63, in inner_f
    return f(*args, **kwargs)
  File "C:\Users\egor.kraev\.conda\envs\generic3.9\lib\site-packages\sklearn\utils\validation.py", line 720, in check_array
    _assert_all_finite(array,
  File "C:\Users\egor.kraev\.conda\envs\generic3.9\lib\site-packages\sklearn\utils\validation.py", line 103, in _assert_all_finite
    raise ValueError(
ValueError: Input contains NaN, infinity or a value too large for dtype('float64').
python-BaseException

No repository file or test is named. Start by reproducing SparseLinearDML with the reported model settings and inspect the scikit-learn convergence warnings and traceback; done means identifying why the fit produces divide-by-zero and NaN values and adding a verified regression test or documented resolution.