Skip to content

included complex step finite method for the gradient computation #74

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.

Sign up for GitHub

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

Open
wants to merge 2 commits into
base: master
Choose a base branch
from
Open

included complex step finite method for the gradient computation #74

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
b2d008a
included complex step finite method for the gradient computation
flavioluiz Aug 15, 2015
6a5f478
solves a problem for derivative when x = 0
flavioluiz Aug 18, 2015
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
13 changes: 11 additions & 2 deletions src/finite_difference.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,7 +38,7 @@ end
macro complexrule(x, e)
x, e = esc(x), esc(e)
quote
$e = eps($x)
$e = eps($x + 1e-200)
end
end

Expand Down Expand Up @@ -129,8 +129,17 @@ function finite_difference!{S <: Number, T <: Number}(f::Function,
x[i] = oldx
g[i] = (f_xplusdx - f_xminusdx) / (epsilon + epsilon)
end
elseif dtype == :complex
xcomplex = x + 0.*im
for i = 1:n
@complexrule x[i] epsilon
oldxi = xcomplex[i]
xcomplex[i] = oldxi + epsilon * im
g[i] = imag(f(xcomplex)) / epsilon
xcomplex[i] = oldxi
end
else
error("dtype must be :forward or :central")
error("dtype must be :forward, :central or :complex")
end

return
Expand Down