Skip to content

Conversation

@nmheim
Copy link
Member

@nmheim nmheim commented Oct 15, 2019

This will need the following packages on master:

  [be33ccc6] CUDAnative v2.4.0 #master (https://github.com/JuliaGPU/CUDAnative.jl.git)
  [c5f51814] CUDAdrv v3.1.0 #master (https://github.com/JuliaGPU/CUDAdrv.jl.git)
  [3a865a2d] CuArrays v1.3.0 #master (https://github.com/JuliaGPU/CuArrays.jl.git)
  [587475ba] Flux v0.9.0 #master (https://github.com/FluxML/Flux.jl.git)
  [e88e6eb3] Zygote v0.3.4 #master (https://github.com/FluxML/Zygote.jl.git)
  [700de1a5] ZygoteRules v0.2.0 #master (https://github.com/FluxML/ZygoteRules.jl.git)
  [7869d1d1] IRTools v0.2.3 #master (https://github.com/MikeInnes/IRTools.jl.git)

With Zygote we will loose the ability to specify trainable parameters of a model through TrackedArrays. The replacement for this is Flux.trainable which we would have to overload for each type

TODO

  • update doc strings
  • the @nograd fix might break some stuff in Distributions.jl. only apply it to the specific functions we need?
  • fix GAN losses and uncomment corresponding tests
  • clean up dependecies
  • use StaticArray in UnitVar CGaussian NoGradArray
  • enable GPU tests for Rodent
  • enable all VAE/GAN tests
  • rename pdf files

@nmheim nmheim self-assigned this Oct 15, 2019
@nmheim
Copy link
Member Author

nmheim commented Oct 15, 2019

@nmheim nmheim changed the title Move to Julia 1.3 and Zygote WIP: Move to Julia 1.3 and Zygote Oct 15, 2019
@codecov-io
Copy link

codecov-io commented Oct 16, 2019

Codecov Report

Merging #36 into master will decrease coverage by 11.71%.
The diff coverage is 95.23%.

Impacted file tree graph

@@             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
Impacted Files Coverage Δ
src/GenerativeModels.jl 100% <ø> (ø) ⬆️
src/utils/utils.jl 15% <ø> (-40%) ⬇️
src/models/vae.jl 100% <100%> (+5%) ⬆️
src/models/rodent.jl 100% <100%> (+5.88%) ⬆️
src/pdfs/abstract_pdfs.jl 100% <100%> (ø) ⬆️
src/models/gan.jl 27.77% <100%> (-27.23%) ⬇️
src/pdfs/gaussian.jl 100% <100%> (ø) ⬆️
src/pdfs/abstract_cgaussian.jl 100% <100%> (ø) ⬆️
src/utils/saveload.jl 100% <100%> (ø) ⬆️
src/utils/ode_decoder_1dobs.jl 100% <100%> (ø)
... and 9 more

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 4239c6d...cb4ecdd. Read the comment docs.

@nmheim nmheim changed the base branch from master to layers October 17, 2019 06:06
nmheim and others added 7 commits October 18, 2019 15:16
* 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.
This was referenced Oct 23, 2019
@nmheim nmheim mentioned this pull request Oct 23, 2019
@nmheim nmheim changed the base branch from layers to master October 24, 2019 19:55
@nmheim nmheim merged commit 1637342 into master Oct 24, 2019
@nmheim nmheim deleted the julia-1.3 branch October 24, 2019 20:05
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants