May 15th, 2022
- Model Retraining: Added ability to retrain models, choosing the date range of serving data to train with. Check the updated requirements in model retraining.
March 10th, 2022
🚨 BREAKING CHANGES
- Models and Deployments: Created a separate entity for models inside the deployment. Now users create deployments and add models to the deployment. With this change, the deployment is the entity you communicate with to interact with model, and the model the containerized format of your trained production-ready model.
- Challenger Models: Introduced ability to add challenger models to deployments. Challenger models receive the same traffic as the live model allowing users to directly compare a model and its variations in a production environment.
- API Monitor: Added CPU and RAM usage graphs to API Monitor. Users can now see how much resources their models consume.
- Activity: Added "Activity Card" in deployments overview page.
- Training Data: Fixed issue of upload failure with large files. Users can now upload up to 2GB of training data without facing timeout.
December 30th, 2021
- Model Monitor: added api endpoint to send prediction feedback/ground truth. You can now view all predictions and their feedback (if any) in the predictions tab.
- Model Monitor: added ability to retrieve evaluation metrics from deployment. You can now view your the metrics (if any) in the predictions tab.
- API Monitor: added status code and remote address filter to request logs.
- Data Statistics: added "other" category in histogram.
- Data Statistics: added support for datetime features.
- Users: Added sent invitations to users page.
- Konan Template Deployments: Initial release of Konan Template Deployments, a tool developed to help you bootstrap your first deployment on Konan.
- API Monitor: show user email instead of ID in "Requested by" column in request logs table.
November 18th, 2021
- Added serving data statistics. You can now see data profiling on serving data in the data statistics tab.
- Added data drift to highlight statistical differences between training data features and serving data.
- Added functionality to upload training data to be used in calculating data drift.
October 1st, 2021
🚨 BREAKING CHANGES
- Removed "projects". You can now directly create deployments. Old deployments belonging to projects have been moved to their own deployments page.
- Changed url scheme of Konan's API endpoints.
- Integrated Konan's Container Registry (KCR), where you can host your container images.
- Added live container logs tab to deployments, where you can monitor your deployments' historical and live logs.
- Added delete deployment functionality.
- Initial release of Konan's Python SDK. Konan SDK allows you to easily integrate Konan deployments into your infrastructure. You can still integrate your deployments using Konan's API.
- Fixed average response time equation in deployments page.
- Fixed inability to choose "today" in date picker.
- Increased time before scaling down deployments. Continuous traffic should now keep a deployment up to avoid cold-start penalty.
August 15th, 2021
- Organizations: Organization groups a set of individuals under the same entity. Members of the same organization see the same projects.
- Projects: A project groups different versions of a deployment together. It has only one active model which is the live model. Previously deployed models are archived under the old model section.
- Deployments: Deployments are your Konan-hosted production-ready machine learning models that help you take data-driven business decisions.