Approving and Implementing a Model

A machine learning model in LityxIQ can be setup to require approvals before being implemented into production.  This step is not required, but can be used to help ensure that models are reviewed and approved before production-level data is scored based on it.

Once a model has been successfully created in LityxIQ, click on the model to highlight it. Then click or right-click and select Implement & Monitor -> Settings.

 

The first tab, Approvals & Production, is used to set approvers and put models into production.  It has three sections, described below.

 

Set Model Approvers

The Approvers area of the tab allows you to set the approvers for this model.  To add approvers, select them from this list and click Save Approver Changes.

 

Select a Version and Iteration for Approval

The Approval Requests section lets you decide which version and iteration of the model is to be requested for approval.  Select the version and iteration of the model you wish to use and then click on the Select button below to request that it be put it into production.

If there are approvers set for the model, they will automatically receive an email asking them to approve it.  The approver(s) will then need to log in to LityxIQ, review the model as they deem appropriate, and in this dialog box click the Approve button if they do wish to approve it.

Special note regarding "Always most recent and best"

There is a special option available in the Select Version for Production list named "Always most recent and best". 

This can be selected for production instead of a specific version and iteration of the model.  If this option is selected, whenever the model is scored against another dataset, LityxIQ will use the most recent version of the model that has been built, and determine which iteration of that version performed the best (according to the selected performance metric for that model).  This option allows a model's production version to be completely dynamic.  It may often change based on recent model runs, but will always be based on the best performing algorithm.  The downside is that there will be much less, or perhaps even no time, for others to review the model to ensure it meets all qualitative criteria necessary for production models.

 

Put the Approved Model Version into Production

The Production Status section is used to remove a version from production, or to put an approved version of the model into production.

If the model is not currently in production, this section will look like the following.  In this case, you can click the button to put the approved version into production.

If the model is currently in production, then you will have the option to remove the current version from production, as shown below.

Once you have removed the current production version, the interface will update.  If a new version has been approved, you will be able to now put it into production.