interpretml
DataException: MissingColumnsInData, Expected column(s) 0 not found in fitted data
I have a model (fit during an Azure automated ml run) that predicts on a dataframe just fine, but fail when the model and dataframe are passed to interpret functions such as PartialDependence, complaining of missing columns.
# link to an AutoMLRun
from azureml.train.automl.run import AutoMLRun
automl_run = AutoMLRun(experiment=experiment, run_id='AutoML_ae0c7f63-a1b7-4892-af3e-92b79cdcf282')
# grab the best run and model
best_run, best_model = automl_run.get_output()
# get the test dataframe
from azureml.core import Dataset
test_dataset = Dataset.get_by_name(workspace=ws, name='employee_turnover_test')
df_test = test_dataset.to_pandas_dataframe()
y_col = ['EmployeeLeft']
x_col = ['City', 'EmailDomain', 'HiredthroughSMTP', 'ManagerRatingOfLikelihoodToLeave',
'MarkedForPHTProgram', 'MostRecentPerformanceEvaluation', 'SocialMediaActivity',
'Survey_AttitudeTowardWorkType', 'Survey_AttitudeTowardWorkload', 'Survey_RelativePeerAverageAttitudeTowardManager']
x_test = df_test.loc[:,x_col]
y_test = df_test.loc[:,y_col]
# confirm the model predicts on test dataframe
pred = best_model.predict_proba(x_test)
pred
# try feeding the model's predict_proba method and test dataframe to PartialDependence
from interpret.blackbox import PartialDependence
pdp = PartialDependence(predict_fn=best_model.predict_proba, data=x_test)
Full error message:
---------------------------------------------------------------------------
DataException Traceback (most recent call last)
<ipython-input-14-0bf5382bcb1b> in <module>
1 from interpret.blackbox import PartialDependence
2
----> 3 pdp = PartialDependence(predict_fn=best_model.predict_proba, data=x_test)
4 pdp_global = pdp.explain_global(name='Partial Dependence')
/anaconda/envs/azureml_py36/lib/python3.6/site-packages/interpret/blackbox/partialdependence.py in __init__(self, predict_fn, data, sampler, feature_names, feature_types, num_points, std_coef)
43 data, None, feature_names, feature_types
44 )
---> 45 self.predict_fn = unify_predict_fn(predict_fn, self.data)
46 self.num_points = num_points
47 self.std_coef = std_coef
/anaconda/envs/azureml_py36/lib/python3.6/site-packages/interpret/utils/all.py in unify_predict_fn(predict_fn, X)
210 def unify_predict_fn(predict_fn, X):
211 predictions = predict_fn(X[:1])
--> 212 if predictions.ndim == 2:
213 new_predict_fn = lambda x: predict_fn(x)[:, 1] # noqa: E731
214 return new_predict_fn
/anaconda/envs/azureml_py36/lib/python3.6/site-packages/sklearn/utils/metaestimators.py in <lambda>(*args, **kwargs)
114
115 # lambda, but not partial, allows help() to work with update_wrapper
--> 116 out = lambda *args, **kwargs: self.fn(obj, *args, **kwargs)
117 # update the docstring of the returned function
118 update_wrapper(out, self.fn)
/anaconda/envs/azureml_py36/lib/python3.6/site-packages/sklearn/pipeline.py in predict_proba(self, X)
469 Xt = X
470 for _, name, transform in self._iter(with_final=False):
--> 471 Xt = transform.transform(Xt)
472 return self.steps[-1][-1].predict_proba(Xt)
473
/anaconda/envs/azureml_py36/lib/python3.6/site-packages/azureml/automl/core/shared/logging_utilities.py in debug_log_wrapped(self, *args, **kwargs)
299 def debug_log_wrapped(self: Any, *args: Any, **kwargs: Any) -> Any:
300 self._logger_wrapper(log_level, "Starting {} operation of {}.".format(f.__name__, self.__class__.__name__))
--> 301 r = f(self, *args, **kwargs)
302 self._logger_wrapper(log_level, "{} {} operation complete.".format(self.__class__.__name__, f.__name__))
303 return r
/anaconda/envs/azureml_py36/lib/python3.6/site-packages/azureml/automl/runtime/featurization/data_transformer.py in transform(self, df)
406 if self._columns_types_mapping is not None:
407 df = self._check_columns_names_and_convert_types(
--> 408 df, self._columns_types_mapping
409 )
410
/anaconda/envs/azureml_py36/lib/python3.6/site-packages/azureml/automl/runtime/featurization/data_transformer.py in _check_columns_names_and_convert_types(df, columns_types_mapping)
697 columns=col,
698 data_object_name="fitted data",
--> 699 reference_code=ReferenceCodes._DATA_TRANSFORMER_TRANSFROM_COLUMN_NOT_FOUND,
700 )
701 )
DataException: DataException:
Message: Expected column(s) 0 not found in fitted data.
InnerException: None
ErrorResponse
{
"error": {
"code": "UserError",
"message": "Expected column(s) 0 not found in fitted data.",
"target": "X",
"inner_error": {
"code": "BadArgument",
"inner_error": {
"code": "MissingColumnsInData"
}
},
"reference_code": "17049f70-3bbe-4060-a63f-f06590e784e5"
}
}
Overall the Azure platform works so in this case without seeing your exact code I would suspect that you have created an index column. When it tells you '0' was not found it means that you have added this column when attempting to predict using the model and this column name was not present within the training data at the point in time of fitting. The alternative is that you have passed data to model.predict() with no column names.
I encountered the same Expected column(s) <column_name> (label_column_name) not found in X. issue when running a regression automated ML in a pipeline using AutoMLStep and reading the training_data using OutputFileDatasetConfig.
Retrieving the training_data using a Dataset object solved the issue.
train_data = Dataset.get_by_name(
workspace=ws,
name="dataset_name")
automl_config = AutoMLConfig(
training_data=train_data,
...)
train_step = AutoMLStep(
automl_config=automl_config,
...)
azureml-sdk v.1.38.0
