Interpreting Grids and Grid Results

When you click on the Grid Results item in the Model Action menu, you are taken to the Grid Results page, which has two tabs. The first tab is the Grid Table and has one row per grid experiment. You can download the results of a specific grid experiment and remove it from the table.

You can monitor the progress of the grid as it trains and evaluates each parameter combination.

The second tab is the Grid Result Table & Model Creation and lists the results of a combination of a grid experiment and a threshold search. Threshold searching will be discussed in a following section. A single grid parameter combination will appear in this table for each threshold that has been searched for. In the table below notice that all the rows on the first page are all based on grid experiment GRID_00. Each of these rows will have a unique threshold that was used in classification. Also, this table will list the metrics that have been evaluated for the grid parameter combination and for the grid parameter combination and threshold, depending upon the metric. Notice that all of the rows on this page have the same log_loss and auc_score. That is because those metrics do not depend upon the threshold. The accuracy and recall_N columns vary by row because these do depend upon the threshold. The threshold is listed after the metric columns.

Because this table usually contains many rows and since the most useful model for an application may depend upon several metrics, there are buttons for calling up models that let you specify multi-stage filtering and sorting.