ValueError: not enough values to unpack (expected 8, got 3)

[CarloNicolini]

I am using the ContinuosBinning with a number of categorical and numerical variables as X and a float bounded in [0,1] as y.
With some of the categorical variables I sometimes get the following error when fitting:

File ~/opt/miniconda3/envs/creditfactor/lib/python3.10/site-packages/optbinning/binning/continuous_binning.py:628, in ContinuousOptimalBinning._fit(self, x, y, check_input)
    623         [splits, n_records, sums, ssums, stds, min_t, max_t,
    624          n_zeros] = self._prebinning_refinement(
    625             user_splits, x_clean, y_clean, y_missing, x_special,
    626             y_special, y_others)
    627 else:
--> 628     [splits, n_records, sums, ssums, stds, min_t, max_t,
    629      n_zeros] = self._fit_prebinning(
    630         x_clean, y_clean, y_missing, x_special, y_special, y_others)
    632 self._n_prebins = len(n_records)
    634 self._categories = categories

ValueError: not enough values to unpack (expected 8, got 3)

Originally, I was using a BinningProcess + Scorecard pipeline, but going in the details of the issue I've found it originates in the ContinuosBinning class, so I've posted as a problem with it.
By changing the prebinning_method = "uniform" in ContinuosBinning the issue vanishes, but how do I pass this argument in the BinningProcess?

[guillermo-navas-palencia]

Hi @CarloNicolini.

I am unable to reproduce this error. I understand you are using Linux and Python 3.10, an environment tested with GitHub actions. You can pass specific parameters using binning_fit_params. For example

binning_fit_params = {v: {'prebinning_method': 'uniform'} for v in variables}

http://gnpalencia.org/optbinning/tutorials/tutorial_binning_process_FICO_xAI.html#Modeling

[CarloNicolini]

binning_fit_params = {v: {'prebinning_method': 'uniform'} for v in variables}

With this step, the error goes away for my categorical variables. However I am able to reproduce it when switching back to "prebinning_method": "cart".
This happens on Mac M1 and optbinning==0.16.1

[guillermo-navas-palencia]

Thanks. Could you please send a reproducible example? I am not a macOs user, but I could try that example via GitHub actions.

[guillermo-navas-palencia]

Hi @CarloNicolini.

The ContinuousOptimalBinning class had a bug when no prebinning splits were generated. The new release 0.17.0 will fix it. In the meantime, you can use this branch: https://github.com/guillermo-navas-palencia/optbinning/tree/feature/continuous_binning_sample_weight.

To confirm, you should see a similar message when setting verbose=True:

[CarloNicolini]

Thanks. Could you please send a reproducible example? I am not a macOs user, but I could try that example via GitHub actions.

Many thanks, I've just updated to 0.17.0 and it works without the "uniform" prebinning specification.