Skip to content
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.

Sign up for GitHub

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

Jump to bottom

Add joint (discrete + continuous) VAE #9

Merged
merged 55 commits into from
Feb 26, 2021
Merged

Add joint (discrete + continuous) VAE #9

merged 55 commits into from
Feb 26, 2021

Conversation

ziatdinovmax
Copy link
Collaborator

... plus bug fixes in other VAE modules

ziatdinovmax and others added 30 commits December 30, 2020 04:33
@ziatdinovmax
Fix bug that led to incorrect rVAE decoding on non-square data
26f4b32
@ziatdinovmax
Add options for KL annealing and setting up custom encoder/decoder
c86f3ac
@ziatdinovmax
Add (r)VAE models for joint continous and discrete representations
e79e8ca
@ziatdinovmax
Update imports
4c901e1
@ziatdinovmax
Fix bug in discrete variable reparameterization classmethod
2a36759
@ziatdinovmax
Add torchvision to install requirements
0991c98
@ziatdinovmax
Fix issues with decoder's dimensions error for discrete distribution(s)
6f295c3
@ziatdinovmax
Fix manifold_traversal for d != disc_dim
d499e52
@ziatdinovmax
Create a coordinate grid at VAE initialization...
b253544 
... instead of when calling self.fit()
@ziatdinovmax
Automatically add sigmoid "layer" in self.decode)_ when trained w/ "ce"
8d74f9e
@ziatdinovmax
Allow for visualization of 2d slices for joint cont+disc representations
db1a84a
@ziatdinovmax
Clean code
70f89ba
@ziatdinovmax
Add reset options to base vitrainer
fac0f9e
@ziatdinovmax
Update optimizer (re-)setting and saving options
94d06c6
@ziatdinovmax
Fix bug in model loaders that always loaded only base VAE
d01a2db
@ziatdinovmax
Fix dimensionality bugs in encoding methods
84b37a6 
Also, move fit and forward_compute_elbo for VAE out of BaseVAE
@ziatdinovmax
Add additional test for loading atomai's models
1abc3e1
@ziatdinovmax
Add tests for variational autoencoders
c8f9f2f
@ziatdinovmax
Add a provision for loading older models without saved num_iter
513a43a
@ziatdinovmax
Add option to exclude KL rotation term from information capacity limit
5111453
@ziatdinovmax
Fix typo in kwarg for rotation prior
59a0bd9
@ziatdinovmax
Add temperature annealing option for joint rVAE
8c54f41
@ziatdinovmax
Account for padding when displaying non-square traversal manifold grid
9fbcfac
@ziatdinovmax
Update docs
7dc4afd
@ziatdinovmax
Add optional softplus to VAE's encoders
20583bc 
Also, add (optional) auxillary dimensions to VAE's encoders
@ziatdinovmax
Add auxillary dimension to VAE initialization and fix phi_prior kwarg
aeff81d
@ziatdinovmax
Move loss functions for VAE to a spearate module
153f4a7
@ziatdinovmax
Update .gitpod.yml
6de5896
@ziatdinovmax
Update requirements.txt
d18c118
@ziatdinovmax
Update imports
e1cd548
ziatdinovmax and others added 25 commits February 1, 2021 04:16
@ziatdinovmax
Merge branch 'maxdev'
e8d1770
@ziatdinovmax
Add option to choose BCE with/without logits
a752bfb
@ziatdinovmax
New subdir for dgm* modules plus a first shot at auxillary VAE
a292e5a 
*dgm = deep generative model
@ziatdinovmax
Use summation instead of "manually" multipying by the number of pixels
dedfaff
@ziatdinovmax
Pass trainable parameters of auxillary models (if any) to compiler
812455e
@ziatdinovmax
Fix classifier for ssVAE
efd3381
@ziatdinovmax
Make sure that all the auxiliary models are on the proper devices
07fa47a
@ziatdinovmax
Use leaky relu as default activation in ssVAE classifier
7a95981
@ziatdinovmax
Move standard KLD functions to viBaseTrainer
ab8ae92
@ziatdinovmax
Fix ordering of subplots for inputs-targets visualization
c1bc558
@ziatdinovmax
Remove aux_dim argument
c823cc3
@ziatdinovmax
Update matplotlib requirements
60bf504
@ziatdinovmax
Fix typo in install_requires
489099b
@ziatdinovmax
Split vae module into vae, tvae, jvae, and jrvae
d8420c1
@ziatdinovmax
Update docstring
f1a416e
@ziatdinovmax
Fix reconstruction method for VAEs
b4aef41
@ziatdinovmax
Simplify function for KL div calculation for discrete variable
8fcd77f
@ziatdinovmax
Move computation of channel capacity to a separate function
df4d84f
@ziatdinovmax
Clean __init__ args (and docstring) in jVAE and jrVAE
082a8e0
@ziatdinovmax
Update version
bcbd936
@ziatdinovmax
Update docs
28f118a
@ziatdinovmax
Put the nb_classes argument back to __init__
062942c 
Removing it breaks  the j(r)VAE model loading
Will have to remove it in the future anyway
@ziatdinovmax
Fix type annotations
d095cb1
@ziatdinovmax
Merge pull request #9 from ziatdinovmax/maxdev
b820b03 
Joint (r)VAE and bug fixes for VAE modules
@ziatdinovmax
Remove auxillary-VAE
e2f9ad1 
- Keep only VAE models with one encoder and one decoder
- More complex VAE/VED models will be a part of a separate package
@ziatdinovmax ziatdinovmax merged commit 7fb6dcd into pycroscopy:master Feb 26, 2021
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
1 participant
@ziatdinovmax