Bug in 08 Kolmogorov.jl

[finmod]

In m = fmap(cpu , m) "cu" should be "cpu"

[finmod]

And another one further down:

prob = SDEProblem(f_val,g_val,u₀,tspan) instead of prob = SDEProblem(f,g,u₀,tspan)

For: @time sol = solve(prob, NeuralNetDiffEq.NNKolmogorov(m, opt, sdealg, ensemblealg), verbose = true, dt = dt,
dx = dx , trajectories = trajectories , abstol=1e-6, maxiters = 3000)

takes:
Current loss is: 288.77471066944065
Current loss is: 288.7747063768811
15327.516623 seconds (174.48 G allocations: 9.600 TiB, 11.89% gc time)
([105.83 108.06 … 105.32 107.2], Float32[11.2977295 10.500775 … 11.479994 10.80812])

4 hours and 15 minutes!!!

[ChrisRackauckas]

I haven't built it yet to see. @ashutosh-b-b that looks incorrect though if the loss doesn't change?

[finmod]

Also add 'using Distributions' to calculate mean(phi)

[ChrisRackauckas]

using Statistics I think

[ChrisRackauckas]

@ashutosh-b-b it's worth a few fixes here. Notice I refactored a bit in #169

[ChrisRackauckas]

rebuild! advanced/03-kolmogorov_equations

[ChrisRackauckas]

!rebuild advanced/03-kolmogorov_equations.jmd

[github-actions]

Created pipeline: see it here.

[ChrisRackauckas]

Fixed!

[finmod]

Great. 427 seconds by steeping back to CuArrays and CUDAnative. Hopefully, CUDA.jl on Windows will pass its tests soon.