Skip to content

Conversation

@schmidtjonathan
Copy link
Collaborator

Closes #237

@schmidtjonathan schmidtjonathan added documentation Improvements or additions to documentation filtsmooth Issues related to filtering and smoothing labels Dec 22, 2020
@schmidtjonathan schmidtjonathan self-assigned this Dec 22, 2020
@review-notebook-app
Copy link

Check out this pull request on  ReviewNB

See visual diffs & provide feedback on Jupyter Notebooks.


Powered by ReviewNB

@@ -0,0 +1,4552 @@
{
Copy link
Collaborator

@pnkraemer pnkraemer Feb 17, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Applied Stochastic Differential Equations

Bayesian Filtering and Smoothing


Reply via ReviewNB

@@ -0,0 +1,4552 @@
{
Copy link
Collaborator

@pnkraemer pnkraemer Feb 17, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

the align block somehow does not render well -- at least in ReviewNB.

also: "Note that this can be generalized to a linear, time-varying state space model"


Reply via ReviewNB

@@ -0,0 +1,4552 @@
{
Copy link
Collaborator

@pnkraemer pnkraemer Feb 17, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I dont think the inline comments here are necessary -- it is just two lines with descriptive variable names


Reply via ReviewNB

@@ -0,0 +1,4552 @@
{
Copy link
Collaborator

@pnkraemer pnkraemer Feb 17, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Perhaps add some explanation aka "This example thus requires a discrete, lti, Gaussian transition. In Probnum, this is the "DiscreteLTIGaussian" object, which needs to be initialised with the state transition, shift vector and process noise covariance matrix."

What do you think? I suppose this would apply to the blocks below, too.

It is also fine though if you dont wanna do it.


Reply via ReviewNB

@@ -0,0 +1,4552 @@
{
Copy link
Collaborator

@pnkraemer pnkraemer Feb 17, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

linear, time-invariant


Reply via ReviewNB

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

For the sake of consistency, I would prefer that to stay capitalized, if that's ok with you.

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I did not mean capitalisation, but the comma and the dash: Linear, Time-Invariant

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Alright :)

@@ -0,0 +1,4552 @@
{
Copy link
Collaborator

@pnkraemer pnkraemer Feb 17, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

see my "comments" comment above


Reply via ReviewNB

@@ -0,0 +1,4552 @@
{
Copy link
Collaborator

@pnkraemer pnkraemer Feb 17, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'd say here we would need some explanation of what is going on? I.e. statespace.generate samples from the state space model.


Reply via ReviewNB

@@ -0,0 +1,4552 @@
{
Copy link
Collaborator

@pnkraemer pnkraemer Feb 17, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

sampling and interpolation? Not prediction?


Reply via ReviewNB

@@ -0,0 +1,4552 @@
{
Copy link
Collaborator

@pnkraemer pnkraemer Feb 17, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

have you tried plotting the uncertainty sausages in the same plot as the measurents? Would be good to see if their width is appropriate or not


Reply via ReviewNB

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

They _are_plotted in every plot. For the components that are observed, the uncertainty is just lower (above row of plots)

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ah okay my bad, I did not see that.

@@ -0,0 +1,4552 @@
{
Copy link
Collaborator

@pnkraemer pnkraemer Feb 17, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

"Only the dynamics model changes' -- well, the data does too?

"linear, time-invariant stochastic differential equation (SDE)".

Please remind me that we should discuss whether we want to talk about "Wiener processes", or "Brownian motion", and which letters we use in the notation.


Reply via ReviewNB

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think talking about Brownian Motion might be more consistent, cf. IBM. What do you think?

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

in this case yes. Though, on a side note, I personally prefer Wiener process (as this is also how I write my papers about this stuff, and AFAIK, nathanael does too), but this is beyond this PR. There is a discussion in the ODE notation about this issue, so we can continue there,.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Agreed.

@review-notebook-app
Copy link

review-notebook-app bot commented Feb 17, 2021

View / edit / reply to this conversation on ReviewNB

pnkraemer commented on 2021-02-17T07:46:30Z
----------------------------------------------------------------

maths doesnt render well


schmidtjonathan commented on 2021-02-23T10:23:49Z
----------------------------------------------------------------

That's an issue with ReviewNB, I think. Because in the notebook, it does.

pnkraemer commented on 2021-02-23T10:49:13Z
----------------------------------------------------------------

this is what I get on readthedocs. I dont think the "$" should be there

schmidtjonathan commented on 2021-02-23T10:56:11Z
----------------------------------------------------------------

Oh damn, my bad! Fixed.

Thanks for catching that!

@review-notebook-app
Copy link

review-notebook-app bot commented Feb 17, 2021

View / edit / reply to this conversation on ReviewNB

pnkraemer commented on 2021-02-17T07:46:30Z
----------------------------------------------------------------

sounds a bit salty ;)


schmidtjonathan commented on 2021-02-23T10:24:25Z
----------------------------------------------------------------

Does it? :D Not intended

pnkraemer commented on 2021-02-23T10:50:20Z
----------------------------------------------------------------

hmmmm now that I read it again, I am less sure about my comment... maybe I was in a petty mood last time, feel free to ignore this :D

schmidtjonathan commented on 2021-02-23T10:57:20Z
----------------------------------------------------------------

Alright :D

@review-notebook-app
Copy link

review-notebook-app bot commented Feb 17, 2021

View / edit / reply to this conversation on ReviewNB

pnkraemer commented on 2021-02-17T07:46:31Z
----------------------------------------------------------------

maybe a side remark as to what happens with less easy-to-derive jacobians? I.e. use UKF or AD framework of your choice.


@review-notebook-app
Copy link

review-notebook-app bot commented Feb 17, 2021

View / edit / reply to this conversation on ReviewNB

pnkraemer commented on 2021-02-17T07:46:31Z
----------------------------------------------------------------

samplign and interpolation?


@review-notebook-app
Copy link

review-notebook-app bot commented Feb 17, 2021

View / edit / reply to this conversation on ReviewNB

pnkraemer commented on 2021-02-17T07:46:32Z
----------------------------------------------------------------

maybe label with "95%" stdev, instead of 1.96?


schmidtjonathan commented on 2021-02-23T10:34:07Z
----------------------------------------------------------------

I prefer the version as is, but I'm ok with changing it, if you wish :)

pnkraemer commented on 2021-02-23T10:51:23Z
----------------------------------------------------------------

I think it is personal preference, so your call. I personally prefer the percentage, because it is easier to parse, but I guess one can also argue for the 1.96 thing to be more representative of what is going on. If you prefer this, go ahead with it

@pnkraemer
Copy link
Collaborator

Overall, it looks great! Thanks for your efforts on this!

Most of my comments are small things. I only commented some things once, whereas it could be said about the entire PR. Do with this information what you like :)

@pnkraemer pnkraemer mentioned this pull request Feb 19, 2021
Copy link
Collaborator Author

That's an issue with ReviewNB, I think. Because in the notebook, it does.


View entire conversation on ReviewNB

Copy link
Collaborator Author

Does it? :D Not intended


View entire conversation on ReviewNB

Copy link
Collaborator Author

I prefer the version as is, but I'm ok with changing it, if you wish :)


View entire conversation on ReviewNB

Copy link
Collaborator

this is what I get on readthedocs. I dont think the "$" should be there


View entire conversation on ReviewNB

Copy link
Collaborator

hmmmm now that I read it again, I am less sure about my comment... maybe I was in a petty mood last time, feel free to ignore this :D


View entire conversation on ReviewNB

Copy link
Collaborator

I think it is personal preference, so your call. I personally prefer the percentage, because it is easier to parse, but I guess one can also argue for the 1.96 thing to be more representative of what is going on. If you prefer this, go ahead with it


View entire conversation on ReviewNB

Copy link
Collaborator Author

Oh damn, my bad! Fixed.

Thanks for catching that!


View entire conversation on ReviewNB

Copy link
Collaborator Author

Alright :D


View entire conversation on ReviewNB

@schmidtjonathan
Copy link
Collaborator Author

schmidtjonathan commented Feb 23, 2021

@pnkraemer
Concerning the align environments:
Removing the $ caused tox to throw warnings, so I added $$ instead, and now it seems to work. Would you mind checking if it's ok?

Copy link
Collaborator

@pnkraemer pnkraemer left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks good, thanks for your work on this!

@schmidtjonathan schmidtjonathan merged commit 17b3874 into probabilistic-numerics:master Feb 24, 2021
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

documentation Improvements or additions to documentation filtsmooth Issues related to filtering and smoothing

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Example notebooks for filtering

2 participants