Choosing AI Inputs

Now it's time to choose which inputs you'd like to use.

Cognition uses its knowledge of your plant to determine a likely set of points that might be relevant to your AI.

Unless you're trying to improve performance, you should generally leave most of these options checked, if you think you can find suitable examples of their values.

Adding inputs increases the capability of the AI to find new and unexpected interactions between your properties. You never know if wind-direction or humidity are going to have an unexpected effect on your process, or if night-shift tend to run everything a little harder than the day-shift.

While you should avoid having hundreds of inputs, you should certainly aim for as many possibly-relevant inputs as you can.

If you have some properties that change very slowly - such a property containing the day of the week - it might be a pain to find suitable training data for each possible value. To ensure your AI is reliable, it might be worth turning those inputs off.

Some values are analytic values - they're based on other information that is already available. An example of this might be daily, monthly and year-to-date totals. These values aren't going to be relevant to an AI, so should be removed.

Some values are only updated very slowly, such as how long tasks take. It may be difficult to get a good set of training samples covering a wide range of values, so in some cases we'd suggest removing these inputs.

Once you've selected your inputs, it's time to set up your outputs.