Error When Doing Test Request-Response

Question

Error When Doing Test Request-Response

Dana Shields 26

I'm doing a Test Request-Response in Azure Machine Learning Studio (Classic), and am getting the following error below. The model trained and tested fine. Why would this error now be throwing when deployed as a web service?

Error Message: Score Model (AFx Library) : table: The data set being scored must contain all features used during training, missing feature(s): 'Rating_HashingFeature_485', 'Rating_HashingFeature_148', 'Rating_HashingFeature_536', 'Rating_HashingFeature_437', 'Preprocessed Review_HashingFeature_414', 'Preprocessed Review_HashingFeature_666', 'Preprocessed Review_HashingFeature_413', 'Preprocessed Review_HashingFeature_522', 'Preprocessed Review_HashingFeature_602', 'Preprocessed Review_HashingFeature_746'.
Site Path: /workspaces/a5be8959d5e047d3af0d350190323087/webservices/ba9b6b4066df42e5bb5edd9c0ad9062e/endpoints/default/test/rrs
Activity ID: 82e53138-b3b9-4a94-8695-0b8152c505ac
Request ID: 3f55ed33-f261-4103-ba76-9a36e3a3b03c
Workspace ID: a5be8959d5e047d3af0d350190323087
Workspace Type: Free
User Role: Owner
Tenant ID: 6cacd170-f897-4b19-ac58-46a23307b80a

GiftA-MSFT 11,181 Reputation points Moderator

2020-10-06T22:43:17.363+00:00

Thanks for reaching out. Did you include the above specified missing features as input columns when you scoring?
Dana Shields 26 Reputation points

2020-10-07T13:43:10.207+00:00
Dana Shields 26 Reputation points

2020-10-07T13:44:18.86+00:00

See the above. Filter-based feature selection action chose those hashing columns, split data included them, and train/score model used them. Screen shot below is of the output of Score Model.

Everything worked perfectly during the training/testing instantiation of the experiment.
Dana Shields 26 Reputation points

2020-10-07T13:48:55.403+00:00

Output of Score Model action:
Carlo Marques 21 Reputation points

2020-10-12T14:53:18.43+00:00

I have the same question.
So, I created an experiment, based on one of many examples and picked a dataset that required some transformations, columns choosing and categorical features. The model creation worked just fine, with only smaller hiccups. However as I deployed the webservice, the API is requesting the new column data set (created after the feature transformation) and not the original data set. This does not serve my purpose, as my aim for creating a service that would adhere to the original dataset features.

What am I doing wrong? Or shall I be required to implement the data transformation. Also, the feature I am trying to predict is also showing up as a input feature, and that does not make sense.

Any pointers?

In another experiment I see the "apply transformation" block being used. Not all transformation gets carried into it.
Carlo Marques 21 Reputation points

2020-10-12T14:53:57.187+00:00

I have the same question.
So, I created an experiment, based on one of many examples and picked a dataset that required some transformations, columns choosing and categorical features. The model creation worked just fine, with only smaller hiccups. However as I deployed the webservice, the API is requesting the new column data set (created after the feature transformation) and not the original data set. This does not serve my purpose, as my aim for creating a service that would adhere to the original dataset features.

What am I doing wrong? Or shall I be required to implement the data transformation. Also, the feature I am trying to predict is also showing up as a input feature, and that does not make sense.

Any pointers?

In another experiment I see the "apply transformation" block being used. Not all transformation gets carried into it.
GiftA-MSFT 11,181 Reputation points Moderator

2020-10-12T16:04:53.843+00:00

Thanks for reaching out. I encourage you to post this comment as a new question, and share the experiment you are working on or steps to reproduce this issue. Thanks.

1 answer

Your answer

GiftA-MSFT 11,181 Reputation points Moderator

2020-10-06T22:43:17.363+00:00

Thanks for reaching out. Did you include the above specified missing features as input columns when you scoring?
Dana Shields 26 Reputation points

2020-10-07T13:43:10.207+00:00
Dana Shields 26 Reputation points

2020-10-07T13:44:18.86+00:00

See the above. Filter-based feature selection action chose those hashing columns, split data included them, and train/score model used them. Screen shot below is of the output of Score Model.

Everything worked perfectly during the training/testing instantiation of the experiment.
Dana Shields 26 Reputation points

2020-10-07T13:48:55.403+00:00

Output of Score Model action:
Carlo Marques 21 Reputation points

2020-10-12T14:53:18.43+00:00

I have the same question.
So, I created an experiment, based on one of many examples and picked a dataset that required some transformations, columns choosing and categorical features. The model creation worked just fine, with only smaller hiccups. However as I deployed the webservice, the API is requesting the new column data set (created after the feature transformation) and not the original data set. This does not serve my purpose, as my aim for creating a service that would adhere to the original dataset features.

What am I doing wrong? Or shall I be required to implement the data transformation. Also, the feature I am trying to predict is also showing up as a input feature, and that does not make sense.

Any pointers?

In another experiment I see the "apply transformation" block being used. Not all transformation gets carried into it.
Carlo Marques 21 Reputation points

2020-10-12T14:53:57.187+00:00

I have the same question.
So, I created an experiment, based on one of many examples and picked a dataset that required some transformations, columns choosing and categorical features. The model creation worked just fine, with only smaller hiccups. However as I deployed the webservice, the API is requesting the new column data set (created after the feature transformation) and not the original data set. This does not serve my purpose, as my aim for creating a service that would adhere to the original dataset features.

What am I doing wrong? Or shall I be required to implement the data transformation. Also, the feature I am trying to predict is also showing up as a input feature, and that does not make sense.

Any pointers?

In another experiment I see the "apply transformation" block being used. Not all transformation gets carried into it.
GiftA-MSFT 11,181 Reputation points Moderator

2020-10-12T16:04:53.843+00:00

Thanks for reaching out. I encourage you to post this comment as a new question, and share the experiment you are working on or steps to reproduce this issue. Thanks.

Answer 1

GiftA-MSFT 11,181 Moderator

Can you try creating a new experiment, do you observe the same error?

Dana Shields 26 Reputation points

2020-10-12T15:47:47.797+00:00

The error magically disappeared, even on the initial experiment.
GiftA-MSFT 11,181 Reputation points Moderator

2020-10-12T15:52:50.12+00:00

Thanks for the feedback. It was probably an intermittent issue. Sorry for the inconvenience, let us know if this happens again or frequently. Thanks.

Share via

Error When Doing Test Request-Response

1 answer

Your answer