Andres Algaba

Hi @npatki, nice to meet you too! It's my pleasure; thanks to the team for the effort on SDV and the quick response. Yes, I agree. Is there anything which...

Okay, thanks, and no problem!

Hi @npatki, thanks for the feedback! It may indeed be something that belongs in the SDV library. I found: sdv-dev/SDV#913, for example. One obstacle I see is that the ``fit``...

Upon further inspection, I believe there may be a problem in: ```python if condition_column is not None and condition_value is not None: condition_info = self._transformer.convert_column_name_value_to_id( condition_column, condition_value) global_condition_vec = self._data_sampler.generate_cond_from_condition_column_info(...

Update: I believe a problem may reside in the ```generate_cond_from_condition_column_info``` of the ```DataSampler```. ``` python def generate_cond_from_condition_column_info(self, condition_info, batch): """Generate the condition vector.""" vec = np.zeros((batch, self._n_categories), dtype='float32') id_ =...

Hi @npatki, no problem, and thanks for the confirmation! Besides the change from ```_discrete_column_matrix_st``` to ```_discrete_column_cond_st``` (as mentioned above): https://github.com/sdv-dev/CTGAN/blob/5358af7cd653eb0c3a96f9671c90fbdde9672f45/ctgan/data_sampler.py#L153 (By the way, I found issue #169 talking about a...

Hi everyone, I believe that the initial solution from @tejuafonja is correct. See #236 for more details on this, and feel free to give some feedback there!

Hi everyone, would attributing the ``discriminator`` to ``self`` like the ``self._generator`` be a potential solution here? It can maybe be optional with ``False`` as a default?

Hi everyone, related to this, I was wondering whether it would also be helpful to split the G loss into the real generator and cross-entropy loss? The cross-entropy loss dominates...

Hi @Hamideh-Mehri, maybe you can look at my PR: https://github.com/sdv-dev/CTGAN/pull/236. I believe this fixes the issue.