Skip to content

fix: set gauge on all parameters#28

Merged
nbereux merged 1 commit intodevelopfrom
fix_gauge_potts
Mar 12, 2025
Merged

fix: set gauge on all parameters#28
nbereux merged 1 commit intodevelopfrom
fix_gauge_potts

Conversation

@nbereux
Copy link
Contributor

@nbereux nbereux commented Mar 12, 2025

See #27

@nbereux nbereux merged commit 2e7a155 into develop Mar 12, 2025
4 checks passed
@nbereux nbereux deleted the fix_gauge_potts branch March 12, 2025 13:49
@nbereux nbereux linked an issue Mar 12, 2025 that may be closed by this pull request
nbereux pushed a commit that referenced this pull request Mar 17, 2025
* Potts variables (#14)

* change ordering in alphabets

* change default binarize from True to False in the load_dataset function

* Sample state new chain (#17)

* sample_state now does not modify in place the chains given as an input

* new object for modifications in sample_state

* Type hint (#18)

* add return type to sample visibles and hiddens and change the docstring accordingly

* changed some Chain in methods docstring to dict[str, Tensor]

* UI (#19)

* add print of hyperparameters at the beginning of the training

* add overwrite option

* check for file existence  when training

* change softlink to hardlink in save model rbm (#22)

* add float16 as a possible dtype in scripts (#23)

* Map model rework (#24)

* fix concatenation wrong dim aats

* add centered option fit_batch_pcd

* rework map_model to ease using the code with new RBMs

* rework map_model to ease using the code with new RBMs

* update test for overwrite args

* add map_model pt_sampling + match_args_dtype_function

* update version number

* fix: set gauge on all parameters (#28)

* v0.2 (#29)

* update installation instructions

* break bash line on multiple lines

* replace ref to epochs with ref to update

* change TorchRBM to rbms in the doc

* update rbms training tuto

* update dataset tuto

* remove debug print command

* create EBM parent class

* add sample_state method for EBM

* add beta to sample_state method

* change type hint map_model dict

* change RBM to EBM for input/output to files

* replace RBM with EBM + change pbar description to the name of the model

* adapt PCD training for the EBM

* fix name of args in call to sample_state

* adapt PT sampling for the EBM

* remove unused import

* Update AIS for EBM

* Move sample state gibbs implementation to RBM class. Kept it for compatibility.

* add exact computation for generic EBM and RBM as special case

* fix overwrite key in the tests

* update version number

* update tests workflow to work for develop

* update doc

* what's new page

* link to whats_new.rst on the index page

* update doc for EBM

* use hidden magnetizations for the computation of the negative part of the gradient

* move plot_trajectories to analysis repo and remove seaborn dependency

* remove scikit learn dependency and use torch for one hot encoding

* use hidden magnetization for the chains in PBRBM

* fix potts initialization vbias

* remove test of train without weights for proteins

* ensure gauge on params init PBRBM
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.

Worse performance on Potts Bernoulli RBM w.r.t. Alfonso's version

1 participant