Skip to content

Conversation

@frankschae
Copy link
Collaborator

Implements Proposition 5.6 from issue #10 .

The sampling appears to be a slightly shifted.

regression

@frankschae
Copy link
Collaborator Author

with different seed:
regression

@mschauer
Copy link
Owner

Wo wir aufpassen müssen: In einer früheren Version fehlt die Matrix inv(a) in der Gleichung für mu:
Screenshot 2020-11-16 at 11 06 19

Richtig ist

Screenshot 2020-11-16 at 11 04 04

@frankschae
Copy link
Collaborator Author

with smaller noise value:

par = [-0.2, 0.1, 0.09]

regression

@frankschae
Copy link
Collaborator Author

tend = 20 Plot
regression

dy = y2 - y
ds = t2 - t
#@show size(mu), size(Gϕ'), (dy - paramintercept(t, y)*ds)
mu = mu + Gϕ'*(dy - paramintercept(t, y)*ds)
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 with Gϕ' here, we are implementing the correct version which contains g(y, Y.prob.p, t)*g(y, Y.prob.p, t)'

WW = Γ + Ξ
WL = (cholesky(Hermitian(WW)).U)'
th° = WL'\(randn(size(μ))+WL\μ)
return th°
Copy link
Owner

Choose a reason for hiding this comment

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

Lets return Mitosis.Gaussian{(:F,:Γ)}(μ, WW) here.

Copy link
Owner

Choose a reason for hiding this comment

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

(I have to write the rand function for those still, though)

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 moved L30 and L31 to the tests and changed the return value.

y .= Y.u[1]

fjac!(ϕ, y, pest, t)
fill!(μ, zero(eltype(ϕ)))
Copy link
Owner

Choose a reason for hiding this comment

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

Let's allow μ and Γ arguments to conjugate, so several segments can be chained.

Copy link
Collaborator Author

Choose a reason for hiding this comment

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

Added them as:

.. , μ0=nothing, Γ0=nothing)

so if not specified they will be filled with zero , if specified the values are written into μ and Γ.

@mschauer
Copy link
Owner

Looks good.

regression

@frankschae frankschae merged commit 5b37350 into main Nov 16, 2020
@frankschae frankschae deleted the regression branch November 16, 2020 12:57
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants