Your Notebook is now a Draft.

Once it's ready, please submit it for review by our team of Community Moderators. Thank you!


This content now needs to be approved by community moderators.


This essay was submitted and is waiting for review.

News» {{qctrl.question.primary_project.name}}

New feature: Multi-modal predictions

by max.wainwright {{qctrl.question.publish_time | dateStr}} Edited on {{qctrl.question.edited_time | dateStr}} {{"estimatedReadingTime" | translate:({minutes: qctrl.question.estimateReadingTime()})}}
  • Facebook
  • Twitter

  • Continuous predictions are much more powerful and much more nuanced than binary predictions. Rather than specifying whether or not something does happen, you can specify how much of something happens, or when it happens, if it happens at all. The simple controls available thus far have allowed predictors to specify central values and uncertainty ranges for their predictions, but this is often not enough. Sometimes continuous predictions need to have fat tails, and sometimes the most likely values lie on opposite sides of the prediction range. Until now, it's been very difficult to accurately represent these predictions.

    The new multi-modal prediction controls allow you to add up to 5 different bell curves to your prediction distribution. If you have only one prediction component, continuous predictions act just as they did before. But if you need more than one component, you can click the “add component” button and adjust the weights to fine-tune your prediction. More fine-tuned control means (potentially!) more points, so update your old continuous predictions now to maximize your score!

    Note that this feature is complete for now, but it's not complete forever. We welcome any feedback on how to improve it or make it more intuitive. Tentative plans for improvement already include

    • a more interactive probability distribution graph so that you can see exactly how much probability you have assigned to each value;
    • a cumulative distribution graph in addition to the probability distribution; and
    • potentially textual input for prediction ranges, although we need to make sure the interface isn't too inelegant.
    Submit Essay

    Once you submit your essay, you can no longer edit it.