Add kalman smoother and log likelihood #209
Conversation
Shunsuke-Hori
changed the title
Codecov Report
@@ Coverage Diff @@
## master #209 +/- ##
==========================================
+ Coverage 90.76% 90.93% +0.16%
==========================================
Files 25 25
Lines 1744 1776 +32
==========================================
+ Hits 1583 1615 +32
Misses 161 161
Continue to review full report at Codecov.
|
There was a problem hiding this comment.
This is great code and a useful contribution -- Makes it very easy to review when I don't have much to add to what you've done.
Thank you for adding a smoother. We can wait for @sglyon to approve, but I think this is ready to be merged (modulo adding a new line before the ##### Return in docs).
src/kalman.jl
Outdated
| forecast for period ``t`` observation conditional on ``t-1`` | ||
| observation. | ||
| - `y::AbstractVector`: observations at period ``t`` | ||
| ##### Returns |
There was a problem hiding this comment.
It improves readability to put a new line to separate the end of the inputs and the beginning of the returns.
There are a few other places this shows up in this file.
| B = G * Sigma' * G' + R | ||
| M = A * inv(B) | ||
| B = G * Sigma * G' + R | ||
| M = A / B |
There was a problem hiding this comment.
Good catch! This is better than taking the inverse.
|
@cc7768 updated. Thank you!! |
There was a problem hiding this comment.
The code is very clean (as usual 👍 ). Thanks for this contribution @Shunsuke-Hori
I haven't had time to check the math, but it looks right at first glance. If one other person could vouch for the validity I think we can merge.
Also, I think your hunch about performance/efficiency is probably correct. This was written more as a textbook implementation and likely leaves some performance on the table.
|
I double checked the math and equations. Also looks good to me. Really cool contribution all round as well. |
This PR adds Kalman smoother and a function to compute log-likelihood. I also made minor updates to existing code.
Ready for review
PS: I'm wondering if the current
kalman.jlis efficient and well documented enough.