krlennon / rude Goto Github PK

Hello Lennon,

My name is Elias Rodrigues and I am a postdoctoral researcher at School of Applied Mathematics at EMAp-FGV. It is the mainly think tank in Latin América.

My research field is non-Newtonian fluid mechanics, Thixotropy, and Fluid Dynamics. I have read your work at ArXiv (https://arxiv.org/abs/2210.04431) and it is impressive, congratulations!!.
I was wondering how can I have access to your doctoral thesis to read more about your work in details?
Best Regards

I am running on a Mac with Ventura OS and M1 chip. When reading the pre-trained weights, I get good results. When I train the network from scratch, I get a strange error when executing rude.jl from the folder Giesekus using Visual Studio Code:

#for k = range(start_at,length(protocols),step=1)
#for k = range(start_at,1,step=1)
k = 1
    loss_fn(θ) = loss_univ([θ; p_system], protocols[1:k], tspans[1:k], σ0, σ12_all, k)
    cb_fun(θ, l) = callback(θ, l, protocols[1:k], tspans[1:k], σ0, σ12_all, k)
    result_univ = DiffEqFlux.sciml_train(loss_fn, θi,
                         AMSGrad(),
                         cb = cb_fun,
                         allow_f_increases = false,
                         maxiters = 5)
                         #maxiters = 200)
    global θi = result_univ.minimizer
    @save "tbnn.bson" θi
#end

A stack trace is dumped in the terminal, but then the screen clears, and the error is gone. This is very disconcerting. I then reran the code from the terminal (as opposed to Visual Studio Code). Here is the stack trace, which is quite long, so I attached the file. Does this look like an insufficient memory error? Thanks.
stack_trace.txt