Error in python process: <class 'ValueError'>: Numeric feature score_2 is empty

Jarm93

Trying to run a Recommendation system, I received this error not allowing to run the job. Received this error message, checked the database and all fields are with information.

[01:16:18] [INFO] [dku.utils]  - *************** Recipe code failed **************[01:16:18] [INFO] [dku.utils]  - Begin Python stack[01:16:18] [INFO] [dku.utils]  - Traceback (most recent call last):[01:16:18] [INFO] [dku.utils]  -   File "/opt/dataiku/python/dataiku/container/exec_train_recipe.py", line 22, in <module>[01:16:18] [INFO] [dku.utils]  -     main(execution["contextPath"], desc['operationMode'])[01:16:18] [INFO] [dku.utils]  -   File "/opt/dataiku/python/dataiku/doctor/prediction/reg_train_recipe.py", line 204, in main[01:16:18] [INFO] [dku.utils]  -     collector_data = collector.build()[01:16:18] [INFO] [dku.utils]  -   File "/opt/dataiku/python/dataiku/doctor/preprocessing_collector.py", line 39, in build[01:16:18] [INFO] [dku.utils]  -     self.ret["per_feature"][vname] = self.get_feature_analysis_data(vname, per_feature_params)[01:16:19] [INFO] [dku.utils]  -   File "/opt/dataiku/python/dataiku/doctor/preprocessing_collector.py", line 77, in get_feature_analysis_data[01:16:19] [INFO] [dku.utils]  -     raise safe_exception(ValueError, u"Numeric feature {} is empty".format(safe_unicode_str(name)))[01:16:19] [INFO] [dku.utils]  - ValueError: Numeric feature score_2 is empty

Can anyone guide me with the steps to solve this error? Thanks

Operating system used: AWS

Alexandru

From the error shared it looks like Revie the score_2 column doesn't contain valid numeric data, can you please check your input data?

Ninai

Hello, I have the same problem, the feature data is numeric and there are no rows with missing data. Could you please share if and how you solved this @Jarm93 ?

Ninai

Hi again, it seems like this happened because I trained and deployed some models and then did some data preparation on the input data afterwards. I did some normalization, which led to a feature having values of type double when it was integer before.

The solution was to train and deploy the models again with the new input data.