uw-amo / robustdmd.jl Goto Github PK

Hey Travis,

I was trying your code out and found that I had to first convert the input data to Float64 to avoid an a type error with gemm. It might be nice to add some more methods that can take care of this.

I needed to install StatsBase to run this code on Julia v0.6 (not sure about v0.5).

I hope you don't mind these issues I am opening up. The exact DMD function is pretty slow for me because my data is about 15000 samples with 2500 features. I did some profiling and the lines involving transpose are the most expensive steps. I think julia arrays are memory-contiguous along the columns, so one solution would be to assume that the data is already given with time as the second index. Most datasets I have worked with are stored on disk in this way anyway.
I am not sure what the best choice is in terms of memory layout for the optimized DMD solvers though.