Bump compats and update tutorials for Optimization v4 #950
Conversation
|
This is stalled and will likely lead to more people complaining about the Optimization change can we get a decision for it? |
|
@avik-pal update Boltz? |
|
Some tests still fail for DataInterpolations, let me dev and patch those shouldn't be hard. |
| plt = scatter(tsteps, ode_data[1, :]; label = "Data") | ||
| # plot the different predictions for individual shoot | ||
| plot_multiple_shoot(plt, preds, group_size) |
There was a problem hiding this comment.
I don't see how this is going to work if we don't have the predictions?
| println(l) | ||
| # plot current prediction against data | ||
| if doplot | ||
| pred = predict_neuralode(state.u) |
There was a problem hiding this comment.
This is not a viable solution, it doubles the cost. We should show a better standard way to do this.
There was a problem hiding this comment.
It was the same before the gradient and loss were evaluated separately so I figured it's a fair tradeoff, the current alternative is to use a global variable.
Down the line I plan to improve OptimizationState so it can be hacked for these things, I am not sure how it should be yet but should be possible
There was a problem hiding this comment.
We should show a current best practice that does not involve recomputing since that can be quite a regression.
|
What's the state of this? |
Vaibhavdixit02
force-pushed
the
docsupdates
branch
from
70c4fb3 to
d20f849
Compare
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
|
Yay! 🎉 |
Checklist
contributor guidelines, in particular the SciML Style Guide and
COLPRAC.
Additional context
Add any other context about the problem here.