Data Reconstruction fails when selected features doesn't include a categorical feature.

[nikml]

• nannyml version: 0.2.0 (from main branch)
• Python version: 3.10
• Operating System: Fedora 35

Description

Perform Multivariate Drift Analysis with Data Reconstruciton on a dataset without categorical features.

What I Did

I tried to run the code from Data Reconstruction Deep Dive but the code failed with:

# Let's compute univariate drift
rcerror_calculator = nml.DataReconstructionDriftCalculator(model_metadata=metadata, chunk_size=DPP)
rcerror_calculator.fit(reference_data=reference)
# let's compute (and visualize) results across all the dataset.
rcerror_results = rcerror_calculator.calculate(data=data)
rcerror_results.data
# let's create plot with results
figure = rcerror_results.plot()
figure.show()
figure.write_image(file="butterfly-multivariate-drift.svg")
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
/tmp/ipykernel_10304/2321058283.py in <cell line: 3>()
      1 # Let's compute univariate drift
      2 rcerror_calculator = nml.DataReconstructionDriftCalculator(model_metadata=metadata, chunk_size=DPP)
----> 3 rcerror_calculator.fit(reference_data=reference)
      4 # let's compute (and visualize) results across all the dataset.
      5 rcerror_results = rcerror_calculator.calculate(data=data)

~/Source/nannyml/nannyml/drift/base.py in fit(self, reference_data)
    165                 self.chunker = DefaultChunker(minimum_chunk_size=minimum_chunk_size)
    166 
--> 167         self._fit(reference_data)
    168 
    169     def _fit(self, reference_data: pd.DataFrame):

~/Source/nannyml/nannyml/drift/data_reconstruction/calculator.py in _fit(self, reference_data)
    101         # TODO: We duplicate the reference data 3 times, here. Improve to something more memory efficient?
    102         imputed_reference_data = reference_data.copy(deep=True)
--> 103         imputed_reference_data[selected_categorical_column_names] = self._imputer_categorical.fit_transform(
    104             imputed_reference_data[selected_categorical_column_names]
    105         )

~/.cache/pypoetry/virtualenvs/nannyml-RmJkXFBz-py3.10/lib64/python3.10/site-packages/sklearn/base.py in fit_transform(self, X, y, **fit_params)
    850         if y is None:
    851             # fit method of arity 1 (unsupervised transformation)
--> 852             return self.fit(X, **fit_params).transform(X)
    853         else:
    854             # fit method of arity 2 (supervised transformation)

~/.cache/pypoetry/virtualenvs/nannyml-RmJkXFBz-py3.10/lib64/python3.10/site-packages/sklearn/impute/_base.py in fit(self, X, y)
    317             Fitted estimator.
    318         """
--> 319         X = self._validate_input(X, in_fit=True)
    320 
    321         # default fill_value is 0 for numerical input and "missing_value"

~/.cache/pypoetry/virtualenvs/nannyml-RmJkXFBz-py3.10/lib64/python3.10/site-packages/sklearn/impute/_base.py in _validate_input(self, X, in_fit)
    285                 raise new_ve from None
    286             else:
--> 287                 raise ve
    288 
    289         _check_inputs_dtype(X, self.missing_values)

~/.cache/pypoetry/virtualenvs/nannyml-RmJkXFBz-py3.10/lib64/python3.10/site-packages/sklearn/impute/_base.py in _validate_input(self, X, in_fit)
    268 
    269         try:
--> 270             X = self._validate_data(
    271                 X,
    272                 reset=in_fit,

~/.cache/pypoetry/virtualenvs/nannyml-RmJkXFBz-py3.10/lib64/python3.10/site-packages/sklearn/base.py in _validate_data(self, X, y, reset, validate_separately, **check_params)
    564             raise ValueError("Validation should be done on X, y or both.")
    565         elif not no_val_X and no_val_y:
--> 566             X = check_array(X, **check_params)
    567             out = X
    568         elif no_val_X and not no_val_y:

~/.cache/pypoetry/virtualenvs/nannyml-RmJkXFBz-py3.10/lib64/python3.10/site-packages/sklearn/utils/validation.py in check_array(array, accept_sparse, accept_large_sparse, dtype, order, copy, force_all_finite, ensure_2d, allow_nd, ensure_min_samples, ensure_min_features, estimator)
    663 
    664         if all(isinstance(dtype, np.dtype) for dtype in dtypes_orig):
--> 665             dtype_orig = np.result_type(*dtypes_orig)
    666 
    667     if dtype_numeric:

<__array_function__ internals> in result_type(*args, **kwargs)

ValueError: at least one array or dtype is required

It's likely that the current implementation cannot handle the selected features not having categorical or not having continuous features included.