Complex numbers alla Flux 1776 #47
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
mcabbott
commented
Jan 30, 2022
mcabbott
force-pushed
the
complex
branch
2 times, most recently
from
72fe5e3
to
5edf5a6
Compare
mcabbott
commented
Jan 30, 2022
mcabbott
force-pushed
the
complex
branch
2 times, most recently
from
284714f
to
309e606
Compare
|
What's the loss function you are using in the plot? |
|
It's the one from the tests, here https://github.com/FluxML/Optimisers.jl/pull/47/files#diff-9f183063573ccf8f01dd29629bad9986f3faa3ce2fdeeb3f0855eaa1f6450715R214 and the plot commands are in comments there. The same loss is used in the Flux PR's tests. |
mcabbott
commented
Feb 2, 2022
Comment on lines
+207
to
+185
| # The Flux PR had 1e-2 for all. But ADADelta(ρ) needs ρ≈0.9 not small. And it helps to make ε not too small too: | ||
| ADAM(1e-2), RMSProp(1e-2), RADAM(1e-2), OADAM(1e-2), ADAGrad(1e-2), ADADelta(0.9, 1e-5), NADAM(1e-2), AdaBelief(1e-2), |
|
Good to go? |
ToucheSir
reviewed
Feb 5, 2022
test/rules.jl
Outdated
Comment on lines
22
to
28
| LOG = Dict() | ||
|
|
||
| loggradient(o) = (f, xs...) -> begin | ||
| y, dxs = Zygote.withgradient(f, xs...) | ||
| push!(get!(() -> Float32[], LOG, name(o)), y) | ||
| dxs # save the loss, return the gradient | ||
| end |
There was a problem hiding this comment.
This (LOG) doesn't seem to be in use anywhere, is it still necessary?
There was a problem hiding this comment.
It just makes debugging easier if you can plot things form the tests you just ran. It's not strictly necessary but also doesn't really get in the way, I think.
There was a problem hiding this comment.
Fair enough. A comment with what you just described would help then.
ToucheSir
approved these changes
Feb 5, 2022
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This copies FluxML/Flux.jl#1776
but doesn't work. Why?Xref #38
Edit:
Now this is on top of #48.The original tests pass, but ADADelta only just passes, only the last few digits of the loss decrease.Not sure ifThat's true in Flux too.Might be a bug somewhere?Because the parameters chosen are strange ones.