LityxIQ provides fully automated model version control and model management of all machine learning models in the platform. All models are tracked through their version history, allowing you to review and compare performance, or deploy any prior versions of the model.
Each model has a Current Version and a Production Version (if it has been put into production). These are easily viewed in the list of models.
The Current Version of a model is comprised of two components. Before the decimal point is the major model version, and after is the minor model version. The current version of a model may also have many iterations associated with it. Iterations are explained more below.
If a model has been put into production, it also has a Production Version, which is comprised of three components. The first two of these components are similar to the current version, representing the major and minor version of the model that has been put into production. The third component represents the Iteration of the model that has been put into production.
The Iteration component of a model version is generated when you execute a model. Each combination of algorithms and settings you have created for the model will result in an "iteration" being performed. A model may have just one iteration, or may have dozens. For example, if you executed 5 algorithms each with two combinations of transformation settings, you will in effect be executing 10 iterations of the model. If the model's current version is, say, 1.8, it will result in the model having iterations 1.8.1, 1.8.2, up through 1.8.10 available for review and for putting into production.
Here are some key points in how model versions are tracked and updated in LityxIQ.
- When a model is initially created, its Current Version is set to 1.0.
- When a model is executed, the results are tracked and tied to the Current Version at that time. Its Current Version is also revised by incrementing the minor version by 1. In other words, a model whose current version is 1.5 will have its current version updated to 1.6 after model execution. Note that this implies that the Current Version of the model contain the latest version of the model settings that have not yet been executed.
- You can also increment (or advance) the major model version of any model. Doing so will increment the first component of the current version by 1 and reset the second component (the minor version) to 0. For example, if you advance the major model version of a model whose current version is 1.6, its current version will move to 2.0. You can do this by selecting a model and clicking Advance Major Model Version from the Selected Model menu or from the right click menu.