Add Hansen's Standardized Skewed T Distribution #86
Conversation
chm-von-tla
changed the title
|
The first commit is a minor style fix |
| Create a standardized skewed t distribution with `v` degrees of freedom and `λ` shape parameter. `ν,λ`` can be passed | ||
| as scalars or vectors. | ||
| """ | ||
| StdSkewT(ν,λ) = StdSkewT([ν λ]) |
| as scalars or vectors. | ||
| """ | ||
| StdSkewT(ν,λ) = StdSkewT([ν λ]) | ||
| StdSkewT(ν::Integer,λ::Integer) = StdSkewT(float(ν),float(λ)) |
There was a problem hiding this comment.
This can be removed. That conversion happens automatically.
There was a problem hiding this comment.
Actually, it doesn't
| """ | ||
| StdSkewT(ν,λ) = StdSkewT([ν λ]) | ||
| StdSkewT(ν::Integer,λ::Integer) = StdSkewT(float(ν),float(λ)) | ||
| StdSkewT(v::Vector{T},λ::Vector{T}) where {T} = StdSkewT{T}(v,λ) |
There was a problem hiding this comment.
This doesn't work because there is no such inner constructor. Also, it probably shouldn't work, so just remove it.
| StdSkewT(v::Vector{T},λ::Vector{T}) where {T} = StdSkewT{T}(v,λ) | ||
| StdSkewT(coefs::Vector{T}) where {T} = StdSkewT{T}(coefs) | ||
|
|
||
| (rand(d::StdSkewT{T})::T) where {T} = (ν=d.coefs[1];λ=d.coefs[2]; quantile(d,rand(1))) |
There was a problem hiding this comment.
rand(d::StdSkewT{T}) where {T} = (ν=d.coefs[1];λ=d.coefs[2]; quantile(d, rand()))
There was a problem hiding this comment.
Is this correct? I see ν and λ being assigned and then not used
| function quantile(d::StdSkewT{T}, q::T) where T | ||
| ν = d.coefs[1] | ||
| λ = d.coefs[2] | ||
| xa = a(d,d.coefs) |
| ν = d.coefs[1] | ||
| λ = d.coefs[2] | ||
| xa = a(d,d.coefs) | ||
| xb = b(d,d.coefs) |
hand optimized the logkernel function for StdSkewT instead of using the support
{a,b,c} functions which were wasteful since calls to them were repeated. A
speed improvement of 4x has been attained on `fit(GARCH{1,1},BG96)`
|
I incorporated the changes you requested, added tests (by basically imitating the tests for the other distributions), I added a mention of the distribution in the docs (I'm not familiar with the process so I may have made mistakes) and optimized the performance (see my last post on #87, and/or commit message for 49e7858) |
|
This is starting to look good. There's just a doctest that's still failing: https://travis-ci.org/github/s-broda/ARCHModels.jl/jobs/762500760 Do you know how to address this? You basically need to update the output in src/usage.md, lines 331-362. If not, I can merge it as is and fix it myself. Cheers |
|
No, I am not familiar with what's going on in the lines you mentioned. However I also spotted a different thing on those logs: after the error on line 990 it is complaining that the evaluated output differs from what is expected (there is a space missing in the function call) |
|
It would be better if you fixed it, I haven't used julia's documentation system before |
|
OK, I'll merge and fix it myself. Thanks for your contribution! |
|
Thank you for the advice along the way! |
No description provided.