8️⃣ Deploy a model and set up the feedback loop

From the model version view, you can now deploy your model to make it able to perform predictions by clicking on “Deploy”.

The deployment can be done on the OVH infrastructure provided by Picsellia or directly on your own infrastructure. During the trial period, the only deployment available is the one on Picsellia infrastructure.

Once the confidence threshold is defined, the model is deployed. From the “Deployment” view you can access all your deployments and select anyone to access its monitoring dashboard.

This view allows you to access many metrics related to your model performances such as predictions latency, heatmap, KS Drift, Outlier score, mAP, and global distribution…. By the way, you can find details about them over there.

Following the deployment, those metrics are supposed to be empty until predictions are done by our model.

Before making predictions, it is useful to set up the feedback loop, it will allow the monitoring dashboard to compute supervised metrics such as KS Drift.

To do it, you need to access the “Settings” view of your deployment. To activate the feedback loop, you have to select the dataset used to train the monitored model and save the configuration

We advise waiting for a few minutes between the activation of the feedback loop and the first predictions.

The continuous training and continuous deployment can also be initiated from this “Settings” view, once activated, it will allow the addition of predicted data in an existing dataset, retraining the model with the enriched dataset, and redeploying the new model version generated.

The main idea of this loop is to ensure that the model is always improving itself by leveraging human-reviewed production data. It is important to know that this new training and deployment loop is activated when the number of predictions reviewed by a human and added to the dataset is reaching a threshold defined in the continuous training settings.

From the “Settings”’ view you can also set up alerts to be informed if a metric is reaching a determined threshold.