====AI Prediction====

AI Prediction is used when you want to try to predict what specific capture values should be based on other capture values.

Most often, you're trying to predict a //result// from something you know before-hand, such as...

  * Determining product quality from the current system state,
  * Calculating an ideal system setup based on the next product information, or
  * Calculating a virtual sensor value based on manual recordings.

See [[AI vs Fuzzy Matching|AI vs Fuzzy Matching]] for a discussion on some key benefits and down-sides of using AI vs using a 'lookup' or 'fuzzy matching' technique.

===Multiple Models Per Capture===

A single type of Capture may have several different AI models, each answering a different type of question.

===Creating an AI Model===

To build an AI model, you press the **AI Model** button under **Search** in the [[the_capture_list|capture list]].

Then - as an Administrator - press **Add New AI Model**.

You'll then be prompted to select the //inputs// and //outputs// of your model. The inputs are the things you're going to know ahead of time, and the outputs are the things you **want** to know.

In our example cupcake bakery, the things we would know before or during production would be...

  * Product Weight,
  * Product Type (ie. Blueberry, Chocolate, Bananna).
  * Product Style (ie, Plain, Deluxe, Fudgy, Gluten Free),
  * Ambient Temperature
  * Ambient Humidity

...while the things we //want to know// - that are normally measured by our QA team - will be...

  * Surface Darkness,
  * Internal Temperature,
  * Texture/Crumble,
  * QA Pass or Fail

Once these are all selected, press the button to build the AI. 

A ZIP file will be downloaded.

===Training the AI===

The ZIP file contains almost everything you need to build an AI model.

Training AI takes a considerable amount of computing power and time. 

Because ARDI servers are often used for applications where data timing is important, we don't suggest training complex AIs on the ARDI server itself. Instead, we give you the tools needed to build the AI on any machine of your choice.

See [[Building an AI Model]] for more details of the process.

===Publishing an AI===

Once your AI has been trained, you'll have two important files.

  * model.keras
  * model.json

Go back to the Capture addon, go to **Search|AI Model**, and press the **Upload New AI Model** button.

Here you'll be able to create a //name// for your AI model (describing what the model does, such as 'Predict Quality Outcomes') and fields to upload both your //keras// and //json// files.

Once you press **Upload**, the files will be updated and you'll be able to run the model from the list.

===Live AI===

To make your AI run **live** - automatically updating with the rest of your process data - you can [[create a ModelHost service]].