Checking your AI Coverage

The last step before we name your AI is the coverage check.

This checks how well you've covered some of your input and output values with your training data.

If you've got discrete inputs (values that move between 1 and 0 for example), you'll be told if you're training data doesn't cover specific numbers.

For analogue values, it checks to see if you've got numbers that are high (in the top third of the range), moderate (in the middle third of the range) and low (at the bottom third of the range).

It's not unusual to have some gaps in your coverage - in some processes, you may simply never experience low values for flow or temperature. But the accuracy of the AI will be poor when you experience values that the AI hasn't been exposed to. So it's worth while resolving any coverage issues when you can.

The second-last step is giving your AI a name.