Processing a Suggestion to Add Pairs
Each suggestion is displayed as a link on the Training tab. When you click this link, the Learn UI shows the Pair Selection tab which is specially configured for processing that suggestion:
|
1.
|
The data table is filtered to display records for the first subset in the family of subsets for this suggestion. |
|
2.
|
The caption of the table indicates the current subset. For example, “Add more pairs to subset 0101111 (2 of 15)”. The binary code indicates the present (1) and empty (0) feature scores in this subset. |
|
3.
|
The Previous and Next buttons are displayed above the data table, if the suggestion has more than one subset. You can use these buttons to filter the data table for each individual subset in the family of subsets for the current suggestion. |
|
4.
|
Use the Reset button to go back to the unfiltered data table and exit the suggestion processing mode. |
You should add a few training record pairs to the current subset. Click Next to see the filtered records for the next subset, and so on until all the subsets in the selected suggestion have been processed. A warning is displayed while proceeding to the next subset if no record pairs were added to the training dataset for the current subset.
Some subsets might have zero or only one record in the table, which is not enough to create a record pair. An appropriate message is displayed, and such subsets can be skipped. You might be able to manually find records in the table that have different empty and non-empty fields. Then you can still create a record pair from the given subset by combining such records.
Note: New pairs are added randomly to either the Training dataset or the Validation dataset. It is the Training dataset that must actually change to see different model training results for that subset. After saving a record pair, the Pair Selection tab indicates the dataset the last pair has been added to.