Inferences

To access and create an inference navigate to AI > Inferences.

AI: Inferences list

Create inference

To create an inference, click the Blue plus button. This will open a form:

Inferences: Create inference form

• Name: Provide a name for the inference.
• Model config: Select a model config in which the desired model was created.
• Model: Choose a trained model.
• AI runner: Select a runner on which the AI is to be executed.
• DataForge User group: Select a DataForge user group. Similar to the reporting feature, you’ll need to provide a user group, so that the service user can provide access to data on the Zabbix server.

Click Create to continue.

Configure inference

To access the config of a model, just click on an entry in the list.

General

This section allows the general configuration of your inference.

Inferences config: General

• Name: The name for the inference.
• Description: You can provide a description for your inference here.
• Switch to latest model: A toggle which, when active, ensures that the most recently generated model is used for inference.
• Host name: Specify the Zabbix host that the reconstruction loss should be sent to.
• Model: Displays the selected model.
• AI runner: Select an AI runner. AI runners are configured by DataForge managers in the server configuration.
• Loss functions: Select the loss functions you want to use.
• Loss item key prefix: The Zabbix item key prefix to which the name of the function is appended (For example: If your loss item key prefix is example.loss and you select the loss function MSE, then your metrics will be sent to: example.loss.mse).

The host and the items are configured in Zabbix and are used to receive the data that the model calculates. To do this, the host name of the configured host and an item key must be provided in the inference configuration. The loss data that the model calculates can then be used as usual in Zabbix to trigger actions at certain thresholds, or for other calculations.

Zabbix: AI-host

Zabbix: loss items with prefix

Inference loss

This diagram shows the calculated data based on the selected loss function. This is also the data that is transmitted to Zabbix. You can select the time period and the loss function above the graph. The graph will be updated accordingly.

Inferences config: Inference details