Prior Knowledge: A Predictive Database For Developers

So, this seems fairly intuitive to me - mostly because I have been trying to build one of these for predicting strategic actions by states ie: when is x country going to launch a missile. I am curious what they are using as their heuristics for developing solutions and how they are determining confidence intervals (my guess: number of samples).

It is actually pretty easy to build a process that "fills in the blanks" the hard part, obviously is getting those blanks correctly. In my experience that takes a fairly deep set of expert opinions and coherent "causal" node programmatics which have assigned confidence intervals. Not to mention the fact that people are going to be scrutinizing this data and if it is wrong, they are going to lose faith in the system, and quickly.

When you talk about implementing this on databases however, either the blanks which are being filled need to not be critical processes which require precision or they need to iterate corrections very quickly. In the end I think this is useful for non-critical products which users can correct if necessary. If however they have a novel solution with a significant history of "right" answers, they may be onto something in machine learning.

Intriguing. The linked article mentions linear regression but this article:

http://blog.priorknowledge.com/blog/scaling-not-your-problem...

suggests something that handels scale, heterogeneous and missing data gracefully. Maybe Random Forests?

Their PreQL demo during the Disrupt was pretty amazing. Im looking forward to using their API for several ML experiments.

Prior Knowledge's Veritable database sounds intriguing. I do a bit of work on the analytics side with medical claims data. I wonder how effective Veritable would be against a large medical claims data set where you have multiple hospital claims, lab claims, pharmacy claims for each member.