Azure Machine Learning Results Interpretation

5
votes

I try to do an experiment in Azure Machine Learning whith a "Decision Forest Regression" Algorythm to predict Weather. enter image description here

I use the Weather Dataset that AML Studio suggested me (It's 400K rows of Wheater in a airport). enter image description here

I would like to predict the "DryBulbCelsus" column (it's values between 20 and 23), so I select the column in the Train Model. I run it everything goes well. But the problem is that I don't understand my score model. I have 2 more colums of results "Score Label Mean" and "Score Label Standard Deviation" with data that I don't understand. enter image description here

If someone work with AML and can explain me how I must interprete the data in result. Thank you !

2
azureregressionazure-machine-learning-studio
Check the settings in "Decision Forest Regression" model. 'Create Trainer Mode' should be set to "Single parameter". - Haritha Thilakarathne
@HarithaThilakarathne It's set to "Single parameter". I actually try to reproduce the experiment with another algorythm and the result is the same, the value are bigger that the original data. - Atloka
check whether the type of the data column is numerical and you've cleaned the missing values and the outliers. - Haritha Thilakarathne
@A.Philipps It too late to ask but did you find the way to read it. Cause Documentation says it should be score label. but I am getting the above same. Could you please reply if you got anything about it. - Er. Mohit Agrawal

2 Answers

6
votes

Some learners, specifically the Decision Forest family and Bayes Point Machine, are capable of estimating the uncertainty around the prediction.

The "Scored Label Mean" is the prediction, and "Scored Label Standard Deviation" is the uncertainty around that prediction.

0
votes

I was confused by this too, 'Scored Label Mean' is the mean of the scored labels from the different trees in the forest in a Decision Forest model, for example. And so is equivalent to the 'Scored Labels' output from a Liner Regression model, for example.