Does Snips maintain any metrics on intent prediction accuracy?
I’ve got training data for about 45 intents. I’d like to run through a battery of phrase classifications and get an average for the confidence for each intent. Then I could concentrate on increasing the accuracy of just the low scoring intents. Does such an animal exist?
If not, I will write a test frame that will take a phrase and the intent that should be returned. It’ll spit out the resulting average for each intent, a list of errors, and whatever I can think of that might also be useful.