Calibration plot

aksaha

What would be the best way to create Calibration plot for a Deep Learning Prediction model for a binary classifier?
The calibration curve works by sorting the probabilities assigned to the records being predicted by the probability reported by the classifier. It then bins the values, and calculates two things.
Thanks

rdesai

Sure,  So we don't have an extra operator for that,  but you should be able to create a process to get the data in the right shape and use standard plotting graphs.  The probs are being given as the confidence values in the output. You can either 
1) Use Generate Attribute to create a process that get's you the data into the right format
2) Since the output contains all probabilities for all class values you can create the plot by using  threshold operators to see how balanced the results are from the classifiers. 

I hope this helps!! 

sgenzer

cc @rdesai ?

rdesai

Sure,  So we don't have an extra operator for that,  but you should be able to create a process to get the data in the right shape and use standard plotting graphs.  The probs are being given as the confidence values in the output. You can either 
1) Use Generate Attribute to create a process that get's you the data into the right format
2) Since the output contains all probabilities for all class values you can create the plot by using  threshold operators to see how balanced the results are from the classifiers. 

I hope this helps!!