-
-
Notifications
You must be signed in to change notification settings - Fork 210
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
Change controller handling with the adaptive option #2284
base: master
Are you sure you want to change the base?
Change controller handling with the adaptive option #2284
Conversation
Co-authored-by: Hendrik Ranocha <[email protected]>
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.
Why does this file exist at all?
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.
It looks like this was a mistake in the big refactors, we should delete it.
if adaptive | ||
controller = default_controller(_alg, cache, qoldinit, beta1, beta2) | ||
else | ||
controller = NonAdaptiveController() | ||
end |
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.
This will introduce an additional type instability since the type of controller
is determined by the value of adaptive
. I cannot really judge how severe this issue is right now. Others probably have more to say about this.
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.
Why is there a type instability ?
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.
julia> function foo(; adaptive = true)
if adaptive
return 1.0
else
return 1
end
end
foo (generic function with 1 method)
julia> @code_warntype foo(adaptive = true)
MethodInstance for Core.kwcall(::@NamedTuple{adaptive::Bool}, ::typeof(foo))
from kwcall(::NamedTuple, ::typeof(foo)) @ Main REPL[1]:1
Arguments
_::Core.Const(Core.kwcall)
@_2::@NamedTuple{adaptive::Bool}
@_3::Core.Const(foo)
Locals
adaptive::Union{}
@_5::Bool
Body::Union{Float64, Int64}
1 ─ Core.NewvarNode(:(adaptive))
│ Core.NewvarNode(:(@_5))
│ %3 = Core.isdefined(@_2, :adaptive)::Core.Const(true)
└── goto #3 if not %3
2 ─ (@_5 = Core.getfield(@_2, :adaptive))
└── goto #4
3 ─ Core.Const(:(@_5 = true))
4 ┄ %8 = @_5::Bool
│ %9 = (:adaptive,)::Core.Const((:adaptive,))
│ %10 = Core.apply_type(Core.NamedTuple, %9)::Core.Const(NamedTuple{(:adaptive,)})
│ %11 = Base.structdiff(@_2, %10)::Core.Const(NamedTuple())
│ %12 = Base.pairs(%11)::Core.Const(Base.Pairs{Symbol, Union{}, Tuple{}, @NamedTuple{}}())
│ %13 = Base.isempty(%12)::Core.Const(true)
└── goto #6 if not %13
5 ─ goto #7
6 ─ Core.Const(:(Base.kwerr(@_2, @_3)))
7 ┄ %17 = Main.:(var"#foo#1")(%8, @_3)::Union{Float64, Int64}
└── return %17
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.
Note the Union{Float64, Int64}
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.
Ok, but is there no type instability due to the fact that the type of the controller depends also on the algorithm is adaptive is true ?
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.
@ChrisRackauckas @oscardssmith What do you think about this ?
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.
it is a minor problem. Are there settings of the normal PI controllers that will make them not adapt ever? It feels like there should be some combination of 0 and Inf that make it never adapt.
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 do not know, is it related to my changes ?
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.
A better strategy would be to add traits to controllers for overrides_adaptive(controller)
or something, which would allow the default controller to stay the same (making it better for compilation) while allowing the alternative controllers to bypass the if adaptive
checks by making them if adaptive || overrides_adaptive(controller)
.
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.
@ChrisRackauckas I thought over this but for each things I tried, I do not succeed to prevent from a type instability. Can you explain a little more ? I think I have missunderstood something but I do not know what.
Checklist
contributor guidelines, in particular the SciML Style Guide and
COLPRAC.
Additional context
One can now use a controller with the option adaptive set to false.