New, simpler pareto front optimization #133
Conversation
|
This seems wrong. Locally, the model is recovered. The docs do not recover the model, and tests recover a third one. |
|
Interesting. What are the sources of randomness here? If you set a random seed does it always give the same result, or is it something deeper? |
|
AFAIK there is none. That's what's bugging me the most. But I'll try that on Sunday / Monday. Did you try to run it on your machine by any chance? I will give this a shot on my old laptop as well, just in case. Even if I think the hardware should not be the issue. |
|
I don't know if this is a win, but after changing to Ubuntu 20.04 and Julia 1.5 the test is failing for me as well :). |
|
Looks like it's good now? |
|
Nope. Michaelis is still wrong by a small factor ( and again not on my laptop but just on travis ). See the log of the tests for that. But in general, yes. This seems to work and is a simple enough way while being flexible. |
|
Works! Woohooo! |
|
what was the trick? |
|
Honestly, I do not know yet. I am working towards that. I added the max convergence to This does not explain why it worked locally for me and not on Travis though... |
src/sindy/isindy.jl
Outdated
|
|
||
| @inbounds for i in 1:size(Ẋ, 1) | ||
| @simd for i in 1:size(Ẋ, 1) |
There was a problem hiding this comment.
this isn't a simd-able loop. The macro will just be a no-op
There was a problem hiding this comment.
Alright. I will change that.
#130