TastyPancakes/Flux.jl
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

add docsting to train!

Browse Source
This commit is contained in:
CarloLucibello 2017-10-11 13:26:40 +02:00 committed by Mike J Innes
parent 6d3a2a2210
commit 041079237e
1 changed files with 11 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
src/optimise/train.jl
View File

@ -4,10 +4,19 @@ using Flux.Tracker: back!
tocb(f) = f tocb(f) = f
tocb(fs::AbstractVector) = () -> foreach(call, fs) tocb(fs::AbstractVector) = () -> foreach(call, fs)
"""
train!(loss, data, opt; cb = () -> ())
For each datapoint `d` in `data` computes the gradient of `loss(d...)` through
backpropagation and calls the optimizer `opt` and the callback `cb`
(i.e. `opt()` and `cb()`).
Multiple callbacks can be passed to `cb` as an array.
"""
function train!(loss, data, opt; cb = () -> ()) function train!(loss, data, opt; cb = () -> ())
cb = tocb(cb) cb = tocb(cb)
@progress for x in data @progress for d in data
l = loss(x...) l = loss(d...)
isinf(l.data[]) && error("Loss is Inf") isinf(l.data[]) && error("Loss is Inf")
isnan(l.data[]) && error("Loss is NaN") isnan(l.data[]) && error("Loss is NaN")
back!(l) back!(l)