Problem in automatic gradient calculation using Deep Learning Toolbox
1 次查看(过去 30 天)
显示 更早的评论
Hello,
Could anyone please tell me why the dlgradient function is throwing the error? It seems to me that somehow the loss is not getting traced to the input weights? Could anyone suggest why it is happening?
x0 = dlarray([1,2,3],'BC');
y0 = dlarray(14,'BC');
fcnn_graph = layerGraph;
layers = [featureInputLayer(3,"Normalization","none","Name","InputLayer")
fullyConnectedLayer(5,"WeightsInitializer","glorot","BiasInitializer","ones","Name","fc1")
tanhLayer("Name","active_1")
fullyConnectedLayer(5,"WeightsInitializer","glorot","BiasInitializer","ones","Name","fc2")
tanhLayer("Name","active_2")
fullyConnectedLayer(1,"WeightsInitializer","glorot","BiasInitializer","ones","Name","fc3")
];
fcnn_graph = addLayers(fcnn_graph,layers);
dlnet = dlnetwork(fcnn_graph);
[gradients,state,loss] = modelGradients(dlnet,x0,y0);
function [gradients,state,loss] = modelGradients(dlnet,X,Y)
[YPred,state] = forward(dlnet,X);
loss = crossentropy(YPred,Y);
gradients = dlgradient(loss,dlnet.Learnables);
loss = double(gather(extractdata(loss)));
end
0 个评论
采纳的回答
Philip Brown
2021-9-9
To ensure tracing is happening, you need to pass your modelGradients function to dlfeval - see this doc page for more details. Replace your line:
[gradients,state,loss] = modelGradients(dlnet,x0,y0);
with
[gradients,state,loss] = dlfeval(@modelGradients, dlnet,x0,y0);
This will ensure that your modelGradients function has tracing between the input weights and the loss.
4 个评论
Philip Brown
2021-9-9
You can use the Learnables property of the dlnetwork; this stores them as a table. For example, for the 'fc1' layer weights, you can use:
dlnet.Learnables{1,3}{1}
更多回答(0 个)
另请参阅
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!