Unknown label type in autosplit

[rogue26]

I get the following error when trying to use the autosplit feature, regardless of whether I split on a numerical or categorical variable.

2019-12-02 20:08:06,514 ERROR Traceback (most recent call last):
  File "<string>", line 182, in auto_split
  File "/home/dataiku/dss/plugins/installed/decision-tree-builder/python-lib/dku_idtb_decision_tree/autosplit.py", line 24, in autosplit
    return compute_splits(df[[feature]], df[target], max_splits)
  File "/home/dataiku/dss/plugins/installed/decision-tree-builder/python-lib/dku_idtb_decision_tree/autosplit.py", line 58, in compute_splits
    tree_estimator.fit(feature_df, target_col)
  File "/home/dataiku/dataiku-dss-5.1.2/python.packages/sklearn/tree/tree.py", line 801, in fit
    X_idx_sorted=X_idx_sorted)
  File "/home/dataiku/dataiku-dss-5.1.2/python.packages/sklearn/tree/tree.py", line 140, in fit
    check_classification_targets(y)
  File "/home/dataiku/dataiku-dss-5.1.2/python.packages/sklearn/utils/multiclass.py", line 171, in check_classification_targets
    raise ValueError("Unknown label type: %r" % y_type)
ValueError: Unknown label type: 'continuous'

Could there be a problem in my data causing this. Alternatively, I wonder if the autosplit module should use DecisionTreeRegressor rather than DecisionTreeClassifier.

[AgatheG]

Hi,

This issue is due to your target feature containing floats. Autosplit would work with any discrete values (booleans, integers or strings).
If you are doing a multiclass classification with floats as classes, a workaround would be to convert your classes (for instance, add a string prefix in a Prepare recipe) so that they would be handled properly by scikit-learn.

Please note that for now, we do not support regression but only classification (binary or multiclass); thus we use DecisionTreeClassifier.