Skip to content
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

#131 - Matrix power via square decomposition, v2 #134

Merged
merged 2 commits into from
Feb 12, 2020
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
22 changes: 20 additions & 2 deletions src/power.jl
Original file line number Diff line number Diff line change
Expand Up @@ -67,11 +67,12 @@ function IntervalMatrixPower(M::IntervalMatrix{T}) where {T}
return IntervalMatrixPower(M, M, 1)
end

function IntervalMatrixPower(M::IntervalMatrix{T}, k::Int) where {T}
function IntervalMatrixPower(M::IntervalMatrix{T}, k::Int;
algorithm::String="power") where {T}
Copy link
Member

Choose a reason for hiding this comment

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

Should algorithm be a field of the type?

Copy link
Member Author

Choose a reason for hiding this comment

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

I was also thinking about that. It is more general to let you choose at every step, but we could at least have a fallback algorithm defined in the type maybe?

Copy link
Member

@mforets mforets Feb 11, 2020

Choose a reason for hiding this comment

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

At first sight it looks better to me if the algorithm is in the type, since it is a property of how to get to M^k, and also how to get higher powers. (Of course i'm assuming that there is no mixed algorithm, but the "mixed" algorithm can be an algorithm by itself).

Copy link
Member

Choose a reason for hiding this comment

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

Anyway, we can revisit this later.

Copy link
Member Author

@schillic schillic Feb 12, 2020

Choose a reason for hiding this comment

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

(Of course i'm assuming that there is no mixed algorithm, but the "mixed" algorithm can be an algorithm by itself).

Yes and no. If the mixed algorithm follows a strict pattern, it can be added as another algorithm. But adding a new algorithm for any pattern is not sustainable. Furthermore, you can imagine patterns that depend on user knowledge (like "next I need more precision") that you cannot express in a static algorithm. So I conclude that a String as a field is not sufficient. If we make the algorithm a field, it should be its own struct <:IntervalMatrixPowerAlgorithm that you can influence from outside.

Copy link
Member Author

Choose a reason for hiding this comment

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

I opened #139.

@assert k >= 1 "matrix powers must be positive"
pow = IntervalMatrixPower(M, M, 1)
@inbounds for i in 1:(k-1)
increment!(pow)
increment!(pow; algorithm=algorithm)
end
return pow
end
Expand All @@ -93,6 +94,7 @@ Increment a matrix power in-place (i.e., storing the result in `pow`).
* `"multiply"` -- fast computation using `*` from the previous result
* `"power"` -- recomputation using `^`
* `"intersect"` -- combination of `"multiply"` and `"power"`
* `"sqrt"` -- decompose `k = a² + b`

### Output

Expand All @@ -113,6 +115,8 @@ function increment!(pow::IntervalMatrixPower; algorithm::String="intersect")
pow.Mᵏ = _eval_power(pow)
elseif algorithm == "intersect"
pow.Mᵏ = _eval_intersect(pow)
elseif algorithm == "sqrt"
pow.Mᵏ = _eval_sqrt(pow)
else
throw(ArgumentError("algorithm $algorithm is not available; choose " *
"from 'multiply', 'power', 'intersect'"))
Expand Down Expand Up @@ -164,6 +168,20 @@ function _eval_intersect(pow::IntervalMatrixPower)
return intersect(_eval_multiply(pow), _eval_power(pow))
end

function _eval_sqrt(pow::IntervalMatrixPower; algorithm::String="power")
# decompose k = a² + b with a, b being integers
k = pow.k
a = floor(Int, sqrt(k))
b = k - a^2

# recursively compute M^a and M^b
Mᵏ = square(get(IntervalMatrixPower(pow.M, a; algorithm=algorithm)))
Copy link
Member

@mforets mforets Feb 11, 2020

Choose a reason for hiding this comment

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

Is line 178 the same as get(IntervalMatrixPower(pow.M, a*a) ? (Also, this is the functionality that i was asking in this comment. I'll create an issue).

EDIT: This is now #138.

Copy link
Member

Choose a reason for hiding this comment

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

My question still is why not use get(IntervalMatrixPower(pow.M, a*a; algorithm=algorithm))) here.

Copy link
Member Author

@schillic schillic Feb 12, 2020

Choose a reason for hiding this comment

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

why not use get(IntervalMatrixPower(pow.M, a*a; algorithm=algorithm))) here.

Because it is not equivalent. We do not use square for arbitrary square numbers but only for powers of 2.

Copy link
Member Author

Choose a reason for hiding this comment

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

Oh, damn, your version was the correct one:
A^(a²) = A^(a*a) = (A^a)^a != (A^a)² for a != 2

if b > 0
Mᵏ *= get(IntervalMatrixPower(pow.M, b; algorithm=algorithm))
end
return Mᵏ
end

"""
base(pow::IntervalMatrixPower)

Expand Down
1 change: 1 addition & 0 deletions test/runtests.jl
Original file line number Diff line number Diff line change
Expand Up @@ -142,4 +142,5 @@ end
increment!(pow) # power of two
increment!(pow, algorithm="power")
increment!(pow, algorithm="intersect")
increment!(pow, algorithm="sqrt")
end