CODEWITHSUNDEEP
azuredecision-treepredictioncortana-intelligenceazure-machine-learning-service
Mikee
Mikee

Reputation: 854

Azure: problems with scoring a decision forest model

I built a two-class decision forest model in the azure ML studio by splitting a dataset into two.

I was happy with the model scoring and evaluation so I brought in a new dataset (with same variables, same data class and same source) for scoring.

I then got an error that the variables in the new scoring dataset was not categorical and I should use 'edit metadata' to make it categorical (I'm not sure sure why). I did this and this immediately made the model evaluation after scoring to be very poor.

How can I solve this problem?

A solution I have in mind but cannot figure out to do it is to avoid using the 'edit metadata' function so I can get same results as I got from the split data evaluation.

Error message:

 Error: Error 1000: AFx Library library exception: Feature 'Age' is of type: 'Numeric' which is not implicitly convertible to type: 'Categorical'. Please use the Metadata editor to explicitly convert the type.

Upvotes: 1

Views: 622

Answers (1)

hoaphumanoid
hoaphumanoid

Reputation: 997

I'm not sure what algorithm you are using, but what the system is saying is that it needs a categorical value to compute its internal parameters.

The low score you are getting probably has to do with the new database and the model you are using to get the score. Edit metadata module shouldn't affect the performance.

Upvotes: 1

Related Questions