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

Added support for instance weights in dagnn.Loss #1112

Open
wants to merge 6 commits into
base: master
Choose a base branch
from
Open

Added support for instance weights in dagnn.Loss #1112

Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
2b8b85c
Merge pull request #1 from vlfeat/master
Sep 22, 2017
64f641d
Updated loss function to include weights for the dag.
Oct 6, 2017
1c2fc05
Changed softmax function to prevent values from ever being 0.
Nicholas-Schaub Oct 26, 2017
262a830
Changed dag Loss function to include weights, put limits on vl_nnsoft…
Nicholas-Schaub Oct 27, 2017
46fd2eb
Fixed dagnn Loss function.
Nicholas-Schaub Oct 27, 2017
4021f68
Reverted changes to vl_nnsoftmax
Feb 3, 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
30 changes: 27 additions & 3 deletions matlab/+dagnn/Loss.m
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,20 +12,44 @@

methods
function outputs = forward(obj, inputs, params)
outputs{1} = vl_nnloss(inputs{1}, inputs{2}, [], 'loss', obj.loss, obj.opts{:}) ;

% If there are 3 inputs, the third input should contain sample
% specific weights
weights = ones(size(inputs{2}));
i = find(strcmpi(obj.opts,'instanceWeights'));
if numel(inputs)==3
weights = inputs{3};
elseif ~isempty(i)
weights = obj.opts{i+1};
end

outputs{1} = vl_nnloss(inputs{1}, inputs{2}, [], 'loss', obj.loss, 'InstanceWeights', weights, obj.opts{:}) ;
obj.accumulateAverage(inputs, outputs);
end

function accumulateAverage(obj, inputs, outputs)
if obj.ignoreAverage, return; end;
if obj.ignoreAverage, return; end
n = obj.numAveraged ;
m = n + size(inputs{1}, 1) * size(inputs{1}, 2) * size(inputs{1}, 4);
obj.average = bsxfun(@plus, n * obj.average, gather(outputs{1})) / m ;
obj.numAveraged = m ;
end

function [derInputs, derParams] = backward(obj, inputs, params, derOutputs)
derInputs{1} = vl_nnloss(inputs{1}, inputs{2}, derOutputs{1}, 'loss', obj.loss, obj.opts{:}) ;

% If there are 3 inputs, the third input should contain sample
% specific weights
weights = ones(size(inputs{2}));
i = find(strcmpi(obj.opts,'instanceWeights'));
if numel(inputs)==3
weights = inputs{3};
derInputs{3} = [];
elseif ~isempty(i)
weights = obj.opts{i+1};
derInputs{3} = [];
end

derInputs{1} = vl_nnloss(inputs{1}, inputs{2}, derOutputs{1}, 'loss', obj.loss, 'InstanceWeights', weights, obj.opts{:}) ;
derInputs{2} = [] ;
derParams = {} ;
end
Expand Down