Hacker News Reader

Comment by RogerL | Hacker News Reader

I'll be shameless and point you to my book on Kalman filtering which I wrote in IPython Notebook, which allows you to experiment within your browser.

https://github.com/rlabbe/Kalman-and-Bayesian-Filters-in-Pyt...

View on HN

I've been using your book to learn about unscented Kalman filters for something I'm working on, and I need to take this opportunity to say thank you! It would've taken me much longer to understand it all without your book - you've saved me much time and frustration.

This looks incredible, and I don't say that lightly. Thanks for posting it!

Looks like a very good resource, however which prior knowledge does one need to have to understand this book?

e.g., do I need to know bayes' rule? bayesian inference? etc.

Nice. Julia version here: https://github.com/wkearn/Kalman.jl

What do you mean by "Julia version?" I thought you meant that you were linking to a version of the ipython-notebook-based Kalman Filter textbook that had been ported to Julia. But... you seem to only be linking to a Julia library that implements Kalman Filters.

yes, that's what I meant, there's a Julia library that implements Kalman filters.

Obviously the important thing (and the one being ported) is the Kalman filter, which is what we're discussing in this post.

Not specifically iPython notebook versions of them.

Well, I was sorta interested until this:

>The key insight to this entire book is in the next paragraph. Read it carefully!

Okay. What is it!!!

>If we only take data from the measurement then the prediction will not affect the result. If we only take data from the prediction then the measurement will be ignored. If this is to work we need to take some kind of blend of the prediction and measurement (I've italicized the key point).

Unless I am missing something, this "key insight" is redundant and has essentially zero information content, apart from the assertion that to get a good result it needs to be a function of two variables, measurement and prediction. Not one. not the other. But two variables, and that's what they are called, yessir.

I think what you're missing is that the ratio is updated continuously.

Reply on news.ycombinator.com