-
Notifications
You must be signed in to change notification settings - Fork 18
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
initial gifs for priors #45
Conversation
Looks mostly good. |
Seems to be a typo in the first sentence here: |
Could add: "... the physical parameter has the (provided) lower and upper bounds" |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I took a "full" read-through. It looks quite good now.
Would it help new users to be explicit about how the the push-forward should work?
For example, if I want to set up a bounded_below(0.0)
constraint, knowing the parameter is O(500)
, what should I set the mean of the normal to? (My understanding is log(500)
, but perhaps it would be useful to make this explicit?
Alternatively, would it be useful to add a few lines of code to show how you could plot the resulting distribution for given fixed parameters? That may be useful if I need to set up a specific problem.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think this is very helpful! It reads well and should provide good example on how to start specifying your priors. Really like the animations
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I like this a lot. One thing I was wondering is whether users would appreciate the ability to plot ParameterDistributions
, both in unconstrained an in constrained space. E.g., one could provide a function plot_pdf(xarray::Array{<:Real,1}, pd::ParameterDistribution, constrained::Bool=false, kw...)
that takes a parameter distribution and plots either the pdf of the unconstrained distribution or that of the constrained distribution. I haven't thought carefully about how best to implement this, but I think that people will often struggle to come up with parameters for the unconstrained prior (e.g., the mean of 0.5 and standard deviation of 1 used in the example), and being given an easy way of seeing what the distribution of the unconstrained pdf gets mapped into when sending it into the constrained space might help with this.
(This would probably be a separate PR.)
@bielim RE your point about plotting. I'm a little wary of it just because plotting is quite problem specific, and only clear in small dimensions. I think it would be worth looking being able to construct a |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM. Let's merge.
Yep, I agree that it would require a lot of thought and work to implement plotting capability in a reasonably general way. But perfect is the enemy of good, so maybe even if the implementation will be limited to simple cases (e.g., only for distributions from independent samples) it would still be useful. I think that in practice, the choice of priors is one of the most difficult aspects of using |
And this now looks good to me, too! Bors it in! |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
looks great!
Thanks everyone! |
bors r+ |
Build succeeded: |
Resolves #44
Check out the parameter distributions section in the documentation
TODO:
@example
with@setup
blocks, to avoid documenter bug