Preprocessing fails in Perform mode #801
Description
Preprocessing seems to fail with text data.
This happens only in Perform mode, not Explain mode. In particular but not only with XgBoost.
Console output:
The task is binary_classification with evaluation metric average_precision
AutoML will use algorithms: ['Xgboost']
AutoML will ensemble available models
AutoML steps: ['simple_algorithms', 'default_algorithms', 'not_so_random', 'golden_features', 'insert_random_feature', 'features_selection', 'hill_climbing_1', 'hill_climbing_2', 'ensemble']
Skip simple_algorithms because no parameters were generated.
* Step default_algorithms will try to check up to 1 model
There was an error during 1_Default_Xgboost training.
Please check automl_results/signup_1hour-profile_changes_v2-bad_actors_v3-20250621-20250721/Perform/errors.md for details.
Error file contents:
## Error for 1_Default_Xgboost
Found array with 0 sample(s) (shape=(0, 100)) while a minimum of 1 is required by TfidfTransformer.
Traceback (most recent call last):
File ".venv/lib/python3.10/site-packages/supervised/base_automl.py", line 1183, in _fit
trained = self.train_model(params)
File ".venv/lib/python3.10/site-packages/supervised/base_automl.py", line 391, in train_model
self.keep_model(mf, model_subpath)
File ".venv/lib/python3.10/site-packages/supervised/base_automl.py", line 294, in keep_model
self._base_predict(self._one_sample, model)
File ".venv/lib/python3.10/site-packages/supervised/base_automl.py", line 1474, in _base_predict
predictions = model.predict(X)
File ".venv/lib/python3.10/site-packages/supervised/model_framework.py", line 447, in predict
X_data, _, _ = self.preprocessings[ind].transform(X.copy(), None)
File ".venv/lib/python3.10/site-packages/supervised/preprocessing/preprocessing.py", line 360, in transform
X_validation = tt.transform(X_validation)
File ".venv/lib/python3.10/site-packages/supervised/preprocessing/text_transformer.py", line 36, in transform
vect = self._vectorizer.transform(x)
File ".venv/lib/python3.10/site-packages/sklearn/feature_extraction/text.py", line 2129, in transform
return self._tfidf.transform(X, copy=False)
File ".venv/lib/python3.10/site-packages/sklearn/feature_extraction/text.py", line 1700, in transform
X = validate_data(
File ".venv/lib/python3.10/site-packages/sklearn/utils/validation.py", line 2954, in validate_data
out = check_array(X, input_name="X", **check_params)
File ".venv/lib/python3.10/site-packages/sklearn/utils/validation.py", line 1128, in check_array
raise ValueError(
ValueError: Found array with 0 sample(s) (shape=(0, 100)) while a minimum of 1 is required by TfidfTransformer.
Please set a GitHub issue with above error message at: https://github.com/mljar/mljar-supervised/issues/new
In Explain mode things are fine:
AutoML directory: automl_results/signup_1hour-profile_changes_v2-bad_actors_v3-20250621-20250721/Explain
The task is binary_classification with evaluation metric average_precision
AutoML will use algorithms: ['Xgboost']
AutoML will ensemble available models
AutoML steps: ['simple_algorithms', 'default_algorithms', 'ensemble']
Skip simple_algorithms because no parameters were generated.
* Step default_algorithms will try to check up to 1 model
1_Default_Xgboost average_precision 0.802009 trained in 31.98 seconds
* Step ensemble will try to check up to 1 model
AutoML fit time: 35.37 seconds
AutoML best model: 1_Default_Xgboost
2025-07-24 09:14:25,482 automl_trainer.train INFO Evaluating model on test set...
2025-07-24 09:14:27,847 automl_trainer.train INFO Test Accuracy: 0.9549
2025-07-24 09:14:29,984 automl_trainer.train INFO Test PR AUC (Average Precision): 0.8071
2025-07-24 09:14:29,989 automl_trainer.train INFO Test ROC AUC: 0.9621
2025-07-24 09:14:29,989 automl_trainer.train INFO Training optimized for: average_precision
Versions:
$ pip list | grep -E "(mljar-supervised|pandas|scikit-learn|numpy|click)"
click 8.2.1
mljar-supervised 1.1.18
numpy 1.26.4
pandas 2.3.1
scikit-learn 1.7.1