AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |
Back to Blog
Computrainer studio1/6/2023 ![]() And then for my very next sample file (from a different vendor, etc) my learned model created all sorts of nonsensical artifacts. I finally gave up because I was literally unable to fix the issues that I had with my own ride files without resorting to machine learning. I couldn't find a sane way to parameterize the probability matrix to account for all the different data sources we have, its very much a science project to come up with a correct model for kalmann. It is provably optimal when the model is correct, but if you don't tune it correctly I found it will confidently deliver a giant mess with all sorts of magical thinking. What I found is that it for our general use it depended too heavily on having a correct probability model. ![]() One of my first attempts to implement smoothing was a kalman filter. Kalmann is absolutely the gold standard for removing noise.
0 Comments
Read More
Leave a Reply. |