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.

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

Jump to bottom

L2 Decay for Optimizers #269

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

L2 Decay for Optimizers #269

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
96d4ffb
L2 Decay added to adam
cangumeli Feb 13, 2018
7cd74db
Empty space removal
cangumeli Feb 13, 2018
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
12 changes: 10 additions & 2 deletions src/update.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -237,7 +237,7 @@ Rmsprop(; lr=0.001, gclip=0, rho=0.9, eps=1e-6)=Rmsprop(lr, gclip, rho, eps, not


"""
Adam(;lr=0.001, gclip=0, beta1=0.9, beta2=0.999, eps=1e-8)
Adam(;lr=0.001, gclip=0, beta1=0.9, beta2=0.999, eps=1e-8, l2decay=0)
update!(w,g,p::Adam)

Container for parameters of the Adam optimization algorithm used by
Expand Down Expand Up @@ -280,9 +280,10 @@ type Adam
t::Int
fstm
scndm
l2decay::AbstractFloat
end

Adam(; lr=0.001, gclip=0, beta1=0.9, beta2=0.999, eps=1e-8)=Adam(lr, gclip, beta1, beta2, eps, 0, nothing, nothing)
Adam(; lr=0.001, gclip=0, beta1=0.9, beta2=0.999, eps=1e-8, l2decay=0)=Adam(lr, gclip, beta1, beta2, eps, 0, nothing, nothing, l2decay)


"""
Expand Down Expand Up @@ -377,6 +378,7 @@ for T in (Array{Float32},Array{Float64},KnetArray{Float32},KnetArray{Float64});
end

function update!(w::$T, g::$T, p::Adam)
l2decay!(w, g, p)
gclip!(g, p.gclip)
if p.fstm===nothing; p.fstm=zeros(w); p.scndm=zeros(w); end
p.t += 1
Expand Down Expand Up @@ -487,6 +489,12 @@ function gclip!(g, gclip)
end
end

function l2decay!(w, g, o)
o.l2decay == 0 && return g
axpy!(o.l2decay, w, g)
end


"""
optimizers(model, otype; options...)

Expand Down