Linear Learner - Multiclass classification - Prediction Scores

Question

Hi,  
I am using the linear learner algo for multiclass classification with 5 classes and 9 features  
  
I'm trying to figure out what the scores relate to? After getting the predicted_label result I get 5 scores....eg....  
  
{"score": \[0.0003054474655073136, 0.8110334873199463, 0.0521857813000679, 0.1320497989654541, 0.004425559192895889], "predicted_label": 1.0}  
  
or  
  
{"score": \[0.0001415203878423199, 0.11968196928501129, 0.8732749223709106, 0.006143826059997082, 0.0007577905780635774], "predicted_label": 2.0}  
  
Does each score relate to a class? I have tried searching the internet for an explanation but so far cannot find anything, hopefully someone can enlighten me  
  
Regards

Answer

Thanks for using SageMaker. The scores returned by multiclass models are the predicted probabilites for each class.   
  
In your first example, the model is predicting class 1 with 81% confidence, and in your second example, the model is predicting class 2 with 87% confidence.  
  
In general, to interpret the `score` from classification models, you have to consider the loss function. If the loss is `logistic` for binary classification or `softmax_loss` for multiclass classification, then the `score` can be interpreted as a probability. These are the default losses used by Linear Learner. But if the loss is `hinge_loss`, then the `score` cannot be interpreted as a probability. This is because hinge loss corresponds to a Support Vector Classifier, which does not produce probability estimates.

Answer

Great, many thanks for the explanation

Linear Learner - Multiclass classification - Prediction Scores

Relevant content