Validation Run

A training run is considered complete once we are able to run the validation set. We strongly encourage you to provide a validation set of up to 1000 records, and we will share the results of the validation with you before declaring the run as complete. If you provide us with the label values of these validation records, we can also augment the validation report with the matched/mismatched/degree of match values.

The validation run is not a mandatory step, but is useful for your team to detect or identify data drifts. These records must match in data precision, input and output node counts. However, to maintain data security, we will not be able to extract some specified index from the training records, since the training records are encrypted. By design, we do not decrypt and store data in a decrypted capacity in any form, which makes extracting a specific record difficult. Therefore, the validation run, if opted for, must be implemented on a standalone data set.