search

XtractOpen / Meganet.jl

A fresh approach to deep learning written in Julia
http://www.xtract.ai/
Other
14 stars 9 forks source link

Parameterize Q in NN #51

Open klensink opened 6 years ago

klensink commented 6 years ago

https://github.com/XtractOpen/Meganet.jl/blob/master/src/integrators/NN.jl#L82

This line would really benefit from parameterizing q in NN.

Right now this.q * Yd is unstable and the append is slow.

Would also help to pre-allocate Ydata to it's full size. One would just need to first count the non-zero entries in this.outTimes with something like: length(find(x -> x==1, this.outTimes))

lruthotto commented 6 years ago

Sounds like a good idea. Not sure how to parametrize Q in a way that still allows it to be an identity operator. Is this really unstable?