denizyuret
denizyuret
@Kausta found a bug in the cat/uncat higher order gradients implementing R1 regularization. I am moving from email to this github issue to follow up. Here is his error description:...
This should replace the multiply implemented functionality in: * AutoGrad: params.jl * Knet: cubytes.jl * Knet: serialize.jl * Knet: gcnode_kptr.jl An implementation exists in: * Knet/models21/deepmap.jl But this should be...
- [ ] SGD learning rate scheduler - [ ] Global gradient clip - [ ] Default to override param.opt rather than keep
@barisbatuhan says: I have implemented a residual block myself in my project and I am having an issue in the residual block. If I only predict by giving an input...
``` kx::Param{KnetArray} kx[:] = 0 # 29us -- setindex!(kx, 0, :) !!! illegal on ax, crashes on cx kx[:] .= 0 # 34us -- materialize!(dotview(kx,:),broadcasted(identity,0)) kx .= 0 # 1380us...
Multiple x sizes can lead to the same y size in conv and pool operations. Our current API picks one in the deconv and unpool implementations. Check out whether other...
Current design: * parameters have an opt field specifying the optimization algorithm/options. * by default these are initialized as `nothing`. * call to a minimizer (sgd, adam etc.) sets the...
## CuArray compatibility and speed on Knet tutorial/examples ### Knet/tutorial - [x] 15.quickstart - [x] 23.learning - [x] 30.lin - [x] 40.mlp - [x] 50.cnn - [x] 60.rnn - [x]...