-
Notifications
You must be signed in to change notification settings - Fork 3
WIP: Move to Julia 1.3 and Zygote #36
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
|
the trainable issue could be solved like this: https://github.com/nmheim/GenerativeModels.jl/blob/ce9e6ecdcc0e7199e34211a1b44dd7a6f123613e/src/pdfs/abstract_pdfs.jl#L126-L129 @smidl @vitskvara @honzabim what do you think? |
Codecov Report
@@ Coverage Diff @@
## master #36 +/- ##
===========================================
- Coverage 86.95% 75.24% -11.72%
===========================================
Files 11 13 +2
Lines 230 206 -24
===========================================
- Hits 200 155 -45
- Misses 30 51 +21
Continue to review full report at Codecov.
|
* use only first element of state as `observation` of state
- this fixes jacobian/@adjoint but does not generalize to
n-dimensional time series
* rename ode_decoder.jl -> ode_decoder_1dobs.jl
also turn of tests for detection of variance as this functionality will be removed soon.
and re-enable other vae tests
This will need the following packages on master:
With Zygote we will loose the ability to specify trainable parameters of a model through
TrackedArrays. The replacement for this isFlux.trainablewhich we would have to overload for each typeTODO
@nogradfix might break some stuff inDistributions.jl. only apply it to the specific functions we need?useNoGradArrayStaticArrayin UnitVar CGaussian