Problem with fitting DPsynthpop

[haaksb]

Hi

I have been testing dpart, and it looks very interesting. I have, however, run into a problem when running .fit on DPsynthpop, here is the code:

from dpart.engines import DPsynthpop
pb_model = DPsynthpop(epsilon=0.4)
pb_model.fit(df)

Error:

File c:\Users\hakonbo\Anaconda3\lib\site-packages\diffprivlib\models\logistic_regression.py:224, in LogisticRegression.fit(self, X, y, sample_weight)
    220     warnings.warn("Data norm has not been specified and will be calculated on the data provided.  This will "
    221                   "result in additional privacy leakage. To ensure differential privacy and no additional "
    222                   "privacy leakage, specify `data_norm` at initialisation.", PrivacyLeakWarning)
    223     self.data_norm = np.linalg.norm(X, axis=1).max()
--> 224 X = self._clip_to_norm(X, self.data_norm)
    226 self.multi_class = _check_multi_class(self.multi_class, solver, len(self.classes_))
    228 n_classes = len(self.classes_)
...
--> 159     raise ValueError(f"Clip value must be strictly positive, got {clip}.")
    161 norms = np.linalg.norm(array, axis=1) / clip
    162 norms[norms < 1] = 1

ValueError: Clip value must be strictly positive, got 0.0.

I have traced the error to the self.root stuff in dpart.py, which creates a column with only zeroes which then gives an error. I dont quite understand its function, or how the first column in the data set is supposed to be synthezised, so that is as far as I got. I would appreciate any help.

On a separate point, it seems that there is suppsed to be a "method" command that determines how each column is fitted, but I dont understand how to use it and cant find documentation on it. I would appreciate some pointers on that as well.

Hope you can take the time to help me.