View a markdown version of this page

Interpreting the answers - Elemental Inference

Interpreting the answers

Use the following information to interpret the answers.

Scores

Each score is a number from 0 to 1. Scores are relative model scores, not calibrated probabilities. For example, a score of 0.8 doesn't mean that something is 80% likely. Use scores to rank or compare results. To choose thresholds, test them on your own content. Elemental Inference doesn't apply a threshold.

chooseOne answers

The selectedOptionName field contains the option that the model picks. The options list contains a score for each option.

yesNoPerOption answers

Elemental Inference scores each option independently. Several options can have high scores at the same time.

Shot answers

The answers from the shot question set appear on items with the type SHOT. Each item has the latest answers for the shot. The answers are about the shot as a whole, up to that point.

Scene answers

The answers from the scene question set appear on items with the type SCENE. Elemental Inference answers the questions periodically while a scene is in progress, and again when the scene ends. A SCENE item can have no customQuestions object until Elemental Inference first answers the questions for that scene. The final item for a scene has the final answers.

An item can also have no customQuestions object when Elemental Inference couldn't answer the questions for that shot or scene. For example, the analysis failed or returned unusable answers, or Elemental Inference couldn't read the version. A missing customQuestions object doesn't mean that the feed has no customer decisions.

To find out which questions produced an answer, use the questionSetVersion and sha256 fields. The questions of a published version never change. DRAFT can change over time, so the sha256 value tells you which content of DRAFT produced the answer. To compare, run get-question-set and check the sha256 value in the response.