Out of place optimizations #107
Conversation
src/jacobians.jl
Outdated
| dx = (vecfx1-vecfx)/epsilon | ||
| J = J + _make_Ji(J, eltype(x), dx, color_i, nrows, ncols) | ||
| if jac_prototype isa Nothing | ||
| J = hcat(J, dx) |
There was a problem hiding this comment.
doing a mapreduce would likely perform the hcat in a smarter way IIRC. This won't allocate as much as before, but it will still allocate quite a bit.
|
I think central differencing is broken? |
|
No it wasn't broken, I used wrong argument. How does this look like now @ChrisRackauckas ? |
|
Can you test this locally with OrdinaryDiffEq.jl before we tag and merge? Last time it seems like we hit something that was missing from the FiniteDiff tests |
|
Yes, I have started one |
| end | ||
|
|
||
| if jac_prototype isa Nothing | ||
| if jac_prototype isa Nothing && sparsity isa Nothing |
There was a problem hiding this comment.
is there a specific file in OrdinaryDiffEq.jl that we should include as a downstream test here?
There was a problem hiding this comment.
There was a problem hiding this comment.
yes that's a good one. Can you setup an environment which does a downstream test to that? That should hopefully catch any outrageous issues.
|
I will merge but let's not tag until we have a bit of downstream testing. |
2 participants
Let's see how test turns out, and move to sparse ones next