Models, or machine learning-based decision processes are the basic atomic component that Superwise observes. The platform is agnostic to model type, use case, data type, and serving platform, however not all model types have ground truths. In those cases, performance metrics will not be available.
In Superwise - a model is an entity that represents a specific task/project that is then represented by different versions that are trained/re-trained over time (see versions). The model can be a simple PyTorch model or even an ensemble of algorithms and models that generate a prediction according to their input data.
Per model - we have several versions, out of which, only one is active and being monitored. On you re-train a model and deploy it in production - you need to activate it instead of the old version - so the model will be monitored.
Updated 7 months ago