rbms 0.1__tar.gz → 0.2.0__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {rbms-0.1 → rbms-0.2.0}/.github/workflows/codecov.yaml +4 -9
- {rbms-0.1 → rbms-0.2.0}/.github/workflows/test.yaml +4 -9
- rbms-0.2.0/.github/workflows/whats_new.yaml +18 -0
- {rbms-0.1 → rbms-0.2.0}/PKG-INFO +52 -23
- rbms-0.2.0/README.md +95 -0
- rbms-0.2.0/docs/source/auto_examples/images/sphx_glr_plot_generated_samples_001.png +0 -0
- rbms-0.2.0/docs/source/auto_examples/images/sphx_glr_plot_generated_samples_002.png +0 -0
- rbms-0.2.0/docs/source/auto_examples/images/sphx_glr_plot_generated_samples_003.png +0 -0
- rbms-0.2.0/docs/source/auto_examples/images/sphx_glr_plot_generated_samples_004.png +0 -0
- rbms-0.2.0/docs/source/auto_examples/images/thumb/sphx_glr_plot_generated_samples_thumb.png +0 -0
- {rbms-0.1 → rbms-0.2.0}/docs/source/auto_examples/plot_generated_samples.codeobj.json +14 -0
- {rbms-0.1 → rbms-0.2.0}/docs/source/auto_examples/plot_generated_samples.ipynb +1 -1
- rbms-0.2.0/docs/source/auto_examples/plot_generated_samples.py.md5 +1 -0
- {rbms-0.1 → rbms-0.2.0}/docs/source/auto_examples/plot_generated_samples.rst +4 -4
- {rbms-0.1 → rbms-0.2.0}/docs/source/auto_examples/sg_execution_times.rst +2 -2
- {rbms-0.1 → rbms-0.2.0}/docs/source/index.md +10 -14
- {rbms-0.1 → rbms-0.2.0}/docs/source/sg_execution_times.rst +2 -2
- {rbms-0.1 → rbms-0.2.0}/docs/source/tuto/dataset.md +2 -3
- rbms-0.2.0/docs/source/tuto/train_rbm.md +51 -0
- {rbms-0.1/docs/source/auto_examples → rbms-0.2.0/examples}/plot_generated_samples.py +2 -2
- {rbms-0.1 → rbms-0.2.0}/pyproject.toml +1 -3
- {rbms-0.1 → rbms-0.2.0}/rbms/bernoulli_bernoulli/classes.py +1 -1
- {rbms-0.1 → rbms-0.2.0}/rbms/classes.py +83 -47
- {rbms-0.1 → rbms-0.2.0}/rbms/dataset/__init__.py +1 -1
- {rbms-0.1 → rbms-0.2.0}/rbms/dataset/fasta_utils.py +3 -3
- {rbms-0.1 → rbms-0.2.0}/rbms/dataset/load_fasta.py +8 -9
- {rbms-0.1 → rbms-0.2.0}/rbms/io.py +17 -10
- {rbms-0.1 → rbms-0.2.0}/rbms/map_model.py +2 -2
- {rbms-0.1 → rbms-0.2.0}/rbms/metrics/aats.py +1 -2
- {rbms-0.1 → rbms-0.2.0}/rbms/parser.py +23 -2
- {rbms-0.1 → rbms-0.2.0}/rbms/partition_function/ais.py +12 -11
- {rbms-0.1 → rbms-0.2.0}/rbms/partition_function/exact.py +17 -3
- {rbms-0.1 → rbms-0.2.0}/rbms/plot.py +6 -44
- {rbms-0.1 → rbms-0.2.0}/rbms/potts_bernoulli/classes.py +6 -2
- {rbms-0.1 → rbms-0.2.0}/rbms/potts_bernoulli/implement.py +33 -20
- {rbms-0.1 → rbms-0.2.0}/rbms/potts_bernoulli/utils.py +4 -1
- {rbms-0.1 → rbms-0.2.0}/rbms/sampling/gibbs.py +3 -6
- {rbms-0.1 → rbms-0.2.0}/rbms/sampling/pt.py +16 -20
- rbms-0.2.0/rbms/scripts/__init__.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/rbms/scripts/entrypoint.py +2 -2
- {rbms-0.1 → rbms-0.2.0}/rbms/scripts/pt_sampling.py +11 -11
- {rbms-0.1 → rbms-0.2.0}/rbms/scripts/train_rbm.py +4 -7
- {rbms-0.1 → rbms-0.2.0}/rbms/training/pcd.py +15 -12
- {rbms-0.1 → rbms-0.2.0}/rbms/training/utils.py +12 -10
- {rbms-0.1 → rbms-0.2.0}/rbms/utils.py +6 -2
- {rbms-0.1 → rbms-0.2.0}/tests/conftest.py +1 -0
- {rbms-0.1 → rbms-0.2.0}/tests/unit_test/bernoulli_bernoulli/test_methods_bbrbm.py +3 -0
- {rbms-0.1 → rbms-0.2.0}/tests/unit_test/potts_bernoulli/test_methods_pbrbm.py +3 -0
- {rbms-0.1 → rbms-0.2.0}/tests/unit_test/training/test_pcd.py +19 -2
- {rbms-0.1 → rbms-0.2.0}/tests/use_cases/test_bbrbm.py +3 -2
- {rbms-0.1 → rbms-0.2.0}/tests/use_cases/test_pbrbm.py +7 -3
- rbms-0.1/README.md +0 -65
- rbms-0.1/docs/source/Makefile +0 -20
- rbms-0.1/docs/source/auto_examples/images/sphx_glr_plot_generated_samples_001.png +0 -0
- rbms-0.1/docs/source/auto_examples/images/sphx_glr_plot_generated_samples_002.png +0 -0
- rbms-0.1/docs/source/auto_examples/images/sphx_glr_plot_generated_samples_003.png +0 -0
- rbms-0.1/docs/source/auto_examples/images/sphx_glr_plot_generated_samples_004.png +0 -0
- rbms-0.1/docs/source/auto_examples/images/sphx_glr_plot_test_001.png +0 -0
- rbms-0.1/docs/source/auto_examples/images/thumb/sphx_glr_plot_generated_samples_thumb.png +0 -0
- rbms-0.1/docs/source/auto_examples/images/thumb/sphx_glr_plot_test_thumb.png +0 -0
- rbms-0.1/docs/source/auto_examples/plot_generated_samples.py.md5 +0 -1
- rbms-0.1/docs/source/auto_examples/plot_test.codeobj.json +0 -18
- rbms-0.1/docs/source/auto_examples/plot_test.ipynb +0 -43
- rbms-0.1/docs/source/auto_examples/plot_test.py +0 -10
- rbms-0.1/docs/source/auto_examples/plot_test.py.md5 +0 -1
- rbms-0.1/docs/source/auto_examples/plot_test.rst +0 -85
- rbms-0.1/docs/source/tuto/train_rbm.md +0 -38
- {rbms-0.1 → rbms-0.2.0}/.github/workflows/docs.yaml +0 -0
- {rbms-0.1 → rbms-0.2.0}/.gitignore +0 -0
- {rbms-0.1 → rbms-0.2.0}/LICENSE +0 -0
- {rbms-0.1 → rbms-0.2.0}/MANIFEST.in +0 -0
- {rbms-0.1 → rbms-0.2.0}/docs/Makefile +0 -0
- {rbms-0.1 → rbms-0.2.0}/docs/make.bat +0 -0
- {rbms-0.1 → rbms-0.2.0}/docs/source/api.md +0 -0
- {rbms-0.1 → rbms-0.2.0}/docs/source/auto_examples/index.rst +0 -0
- {rbms-0.1/examples → rbms-0.2.0/docs/source/auto_examples}/plot_generated_samples.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/docs/source/conf.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/docs/source/examples.md +0 -0
- {rbms-0.1 → rbms-0.2.0}/docs/source/myst.yml +0 -0
- {rbms-0.1 → rbms-0.2.0}/docs/source/rbm.md +0 -0
- {rbms-0.1 → rbms-0.2.0}/docs/source/references.bib +0 -0
- {rbms-0.1 → rbms-0.2.0}/docs/source/tuto/parallel_tempering.md +0 -0
- {rbms-0.1 → rbms-0.2.0}/docs/source/tutorials.md +0 -0
- /rbms-0.1/rbms/__init__.py → /rbms-0.2.0/docs/source/whats_new.rst +0 -0
- {rbms-0.1 → rbms-0.2.0}/examples/GALLERY_HEADER.rst +0 -0
- {rbms-0.1/rbms/metrics → rbms-0.2.0/rbms}/__init__.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/rbms/bernoulli_bernoulli/__init__.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/rbms/bernoulli_bernoulli/functional.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/rbms/bernoulli_bernoulli/implement.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/rbms/const.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/rbms/custom_fn.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/rbms/dataset/dataset_class.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/rbms/dataset/load_h5.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/rbms/dataset/parser.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/rbms/dataset/utils.py +0 -0
- {rbms-0.1/rbms/potts_bernoulli → rbms-0.2.0/rbms/metrics}/__init__.py +0 -0
- {rbms-0.1/rbms/scripts → rbms-0.2.0/rbms/potts_bernoulli}/__init__.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/rbms/potts_bernoulli/functional.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/rbms/potts_bernoulli/tools.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/tests/unit_test/bernoulli_bernoulli/implement/test_compute_energy_bbrbm.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/tests/unit_test/bernoulli_bernoulli/implement/test_compute_energy_hiddens_bbrbm.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/tests/unit_test/bernoulli_bernoulli/implement/test_compute_energy_visibles_bbrbm.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/tests/unit_test/bernoulli_bernoulli/implement/test_compute_gradient_bbrbm.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/tests/unit_test/bernoulli_bernoulli/implement/test_init_chains_bbrbm.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/tests/unit_test/bernoulli_bernoulli/implement/test_init_parameters_bbrbm.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/tests/unit_test/bernoulli_bernoulli/implement/test_sample_hiddens_bbrbm.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/tests/unit_test/bernoulli_bernoulli/implement/test_sample_visibles_bbrbm.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/tests/unit_test/bernoulli_bernoulli/test_bernoulli_utils_bbrbm.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/tests/unit_test/bernoulli_bernoulli/test_functional_bbrbm.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/tests/unit_test/metrics/test_aats.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/tests/unit_test/potts_bernoulli/implement/test_compute_energy_hiddens_pbrbm.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/tests/unit_test/potts_bernoulli/implement/test_compute_energy_pbrbm.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/tests/unit_test/potts_bernoulli/implement/test_compute_energy_visibles_pbrbm.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/tests/unit_test/potts_bernoulli/implement/test_compute_gradient_pbrbm.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/tests/unit_test/potts_bernoulli/implement/test_init_chains_pbrbm.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/tests/unit_test/potts_bernoulli/implement/test_init_parameters_pbrbm.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/tests/unit_test/potts_bernoulli/implement/test_sample_hiddens_pbrbm.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/tests/unit_test/potts_bernoulli/implement/test_sample_visibles_pbrbm.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/tests/unit_test/potts_bernoulli/test_functional_pbrbm.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/tests/unit_test/potts_bernoulli/test_utils_pbrbm.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/tests/unit_test/test_custom_fn.py +0 -0
- {rbms-0.1 → rbms-0.2.0}/tests/unit_test/test_utils.py +0 -0
|
@@ -1,11 +1,9 @@
|
|
|
1
1
|
name: Run tests and upload coverage
|
|
2
2
|
|
|
3
|
-
on:
|
|
4
|
-
push
|
|
3
|
+
on: [push, pull_request, workflow_dispatch]
|
|
5
4
|
|
|
6
5
|
jobs:
|
|
7
6
|
test:
|
|
8
|
-
if: github.ref == 'refs/heads/main'
|
|
9
7
|
name: Run tests on latest python version and collect coverage
|
|
10
8
|
runs-on: ubuntu-latest
|
|
11
9
|
steps:
|
|
@@ -26,16 +24,13 @@ jobs:
|
|
|
26
24
|
run: pip install -e .
|
|
27
25
|
|
|
28
26
|
- name: Get data for binary use case
|
|
29
|
-
run:
|
|
30
|
-
wget https://github.com/nbereux/data_test/raw/refs/heads/main/dummy.zip
|
|
27
|
+
run: wget https://github.com/nbereux/data_test/raw/refs/heads/main/dummy.zip
|
|
31
28
|
|
|
32
29
|
- name: Get data for potts use case
|
|
33
|
-
run:
|
|
34
|
-
wget https://github.com/nbereux/data_test/raw/refs/heads/main/dummy.fasta
|
|
30
|
+
run: wget https://github.com/nbereux/data_test/raw/refs/heads/main/dummy.fasta
|
|
35
31
|
|
|
36
32
|
- name: Extract binary data
|
|
37
|
-
run:
|
|
38
|
-
unzip ./dummy.zip
|
|
33
|
+
run: unzip ./dummy.zip
|
|
39
34
|
|
|
40
35
|
- name: Run tests
|
|
41
36
|
run: PYTORCH_JIT=0 pytest --cov=rbms tests/
|
|
@@ -1,11 +1,9 @@
|
|
|
1
1
|
name: Run tests and upload coverage
|
|
2
2
|
|
|
3
|
-
on:
|
|
4
|
-
push
|
|
3
|
+
on: [push, pull_request, workflow_dispatch]
|
|
5
4
|
|
|
6
5
|
jobs:
|
|
7
6
|
test:
|
|
8
|
-
if: github.ref == 'refs/heads/main'
|
|
9
7
|
name: Run tests on previous python versions
|
|
10
8
|
runs-on: ubuntu-latest
|
|
11
9
|
strategy:
|
|
@@ -29,16 +27,13 @@ jobs:
|
|
|
29
27
|
run: pip install .
|
|
30
28
|
|
|
31
29
|
- name: Get data for binary use case
|
|
32
|
-
run:
|
|
33
|
-
wget https://github.com/nbereux/data_test/raw/refs/heads/main/dummy.zip
|
|
30
|
+
run: wget https://github.com/nbereux/data_test/raw/refs/heads/main/dummy.zip
|
|
34
31
|
|
|
35
32
|
- name: Get data for potts use case
|
|
36
|
-
run:
|
|
37
|
-
wget https://github.com/nbereux/data_test/raw/refs/heads/main/dummy.fasta
|
|
33
|
+
run: wget https://github.com/nbereux/data_test/raw/refs/heads/main/dummy.fasta
|
|
38
34
|
|
|
39
35
|
- name: Extract binary data
|
|
40
|
-
run:
|
|
41
|
-
unzip dummy.zip
|
|
36
|
+
run: unzip dummy.zip
|
|
42
37
|
|
|
43
38
|
- name: Run tests
|
|
44
39
|
run: pytest tests/
|
|
@@ -0,0 +1,18 @@
|
|
|
1
|
+
name: Check What's News Changes
|
|
2
|
+
|
|
3
|
+
concurrency:
|
|
4
|
+
group: ${{ github.workflow }}-${{ github.event.number }}-${{ github.event.ref }}
|
|
5
|
+
cancel-in-progress: true
|
|
6
|
+
|
|
7
|
+
on:
|
|
8
|
+
pull_request:
|
|
9
|
+
types: [assigned, opened, synchronize, reopened, labeled, unlabeled]
|
|
10
|
+
branches:
|
|
11
|
+
- "*"
|
|
12
|
+
jobs:
|
|
13
|
+
check-whats-news:
|
|
14
|
+
runs-on: ubuntu-latest
|
|
15
|
+
steps:
|
|
16
|
+
- uses: tarides/changelog-check-action@v2
|
|
17
|
+
with:
|
|
18
|
+
changelog: docs/source/whats_new.rst
|
{rbms-0.1 → rbms-0.2.0}/PKG-INFO
RENAMED
|
@@ -1,12 +1,13 @@
|
|
|
1
|
-
Metadata-Version: 2.
|
|
1
|
+
Metadata-Version: 2.4
|
|
2
2
|
Name: rbms
|
|
3
|
-
Version: 0.
|
|
3
|
+
Version: 0.2.0
|
|
4
4
|
Summary: Training and analyzing Restricted Boltzmann Machines in PyTorch
|
|
5
5
|
Project-URL: Homepage, https://github.com/DsysDML/rbms
|
|
6
6
|
Project-URL: Issues, https://github.com/DsysDML/rbms/issues
|
|
7
7
|
Author: Aurélien Decelle, Cyril Furtlehner, Alfonso Navas, Lorenzo Rosset, Beatriz Seoane
|
|
8
8
|
Author-email: Nicolas Béreux <nicolas.bereux@gmail.com>
|
|
9
9
|
Maintainer-email: Nicolas Béreux <nicolas.bereux@gmail.com>
|
|
10
|
+
License-File: LICENSE
|
|
10
11
|
Keywords: Restricted Boltzmann Machine,rbm
|
|
11
12
|
Classifier: License :: OSI Approved :: MIT License
|
|
12
13
|
Classifier: Operating System :: OS Independent
|
|
@@ -15,8 +16,6 @@ Requires-Python: >=3.11
|
|
|
15
16
|
Requires-Dist: h5py>=3.12.0
|
|
16
17
|
Requires-Dist: matplotlib>=3.8.0
|
|
17
18
|
Requires-Dist: numpy<=2.1.2,>=2.0.0
|
|
18
|
-
Requires-Dist: scikit-learn==1.5.2
|
|
19
|
-
Requires-Dist: seaborn>=0.13.0
|
|
20
19
|
Requires-Dist: torch<=2.5.0,>=2.0.0
|
|
21
20
|
Requires-Dist: tqdm>=4.65.0
|
|
22
21
|
Description-Content-Type: text/markdown
|
|
@@ -25,7 +24,6 @@ Description-Content-Type: text/markdown
|
|
|
25
24
|
[](https://github.com/pre-commit/pre-commit)
|
|
26
25
|

|
|
27
26
|
|
|
28
|
-
|
|
29
27
|
# RBMs
|
|
30
28
|
|
|
31
29
|
## Overview
|
|
@@ -33,18 +31,43 @@ Description-Content-Type: text/markdown
|
|
|
33
31
|
`rbms` is a GPU-accelerated package designed to train and analyze Restricted Boltzmann Machines (RBMs). It is intended for students and researchers who need an efficient tool for working with RBMs.
|
|
34
32
|
Features:
|
|
35
33
|
|
|
36
|
-
|
|
37
|
-
|
|
38
|
-
|
|
34
|
+
- GPU Acceleration through PyTorch.
|
|
35
|
+
- Multiple RBM Types: Supports Bernoulli-Bernoulli RBM and Potts-Bernoulli RBM.
|
|
36
|
+
- Extensible Design: Provides an abstract class RBM with methods that can be implemented for new types of RBMs, minimizing the need to reimplement training algorithms, analysis methods, and sampling methods.
|
|
39
37
|
|
|
40
38
|
## Installation
|
|
41
39
|
|
|
42
|
-
|
|
40
|
+
### PyPI
|
|
41
|
+
|
|
42
|
+
`rbms` can be installed from PyPI:
|
|
43
|
+
|
|
44
|
+
```bash
|
|
45
|
+
pip install rbms
|
|
46
|
+
```
|
|
47
|
+
|
|
48
|
+
The last version on PyPI corresponds to the main branch of this repo. To update the installed package run
|
|
43
49
|
|
|
44
50
|
```bash
|
|
45
|
-
pip install
|
|
51
|
+
pip install rbms --upgrade
|
|
46
52
|
```
|
|
47
53
|
|
|
54
|
+
### Github
|
|
55
|
+
|
|
56
|
+
If you want a more up-to-date version or wish to contribute, you can install the package directly from this repository:
|
|
57
|
+
|
|
58
|
+
```bash
|
|
59
|
+
git clone git@github.com:DsysDML/rbms.git
|
|
60
|
+
cd rbms && pip install -e .
|
|
61
|
+
```
|
|
62
|
+
|
|
63
|
+
If you want to update the package run
|
|
64
|
+
|
|
65
|
+
```bash
|
|
66
|
+
git pull
|
|
67
|
+
```
|
|
68
|
+
|
|
69
|
+
from inside the cloned repo.
|
|
70
|
+
|
|
48
71
|
## Dependencies
|
|
49
72
|
|
|
50
73
|
The dependencies are included in the pyproject.toml file and should be compiled in the wheel distribution. If you have a GPU, make sure to install PyTorch with GPU support.
|
|
@@ -52,17 +75,22 @@ The dependencies are included in the pyproject.toml file and should be compiled
|
|
|
52
75
|
## Usage
|
|
53
76
|
|
|
54
77
|
Main Classes and Functions:
|
|
55
|
-
|
|
56
|
-
|
|
57
|
-
|
|
58
|
-
|
|
59
|
-
|
|
78
|
+
|
|
79
|
+
- `EBM`: An abstract class that provides the basic structure and methods for Energy-Based Models (EBMs).
|
|
80
|
+
- `RBM`: An abstract class that provides the basic structure and methods for RBMs.
|
|
81
|
+
- `BBRBM`: A concrete implementation of the Bernoulli-Bernoulli RBM.
|
|
82
|
+
- `PBRBM`: A concrete implementation of the Potts-Bernoulli RBM.
|
|
83
|
+
- `rbms.sampling`: Submodule for sampling methods.
|
|
84
|
+
- `rbms.training`: Submodule for training algorithms.
|
|
60
85
|
|
|
61
86
|
## Basic Example
|
|
62
87
|
|
|
63
88
|
Train a RBM on MNIST-01 with PCD-100 for 10 000 steps, using 200 hidden nodes and 2000 permanent chains:
|
|
89
|
+
|
|
64
90
|
```bash
|
|
65
|
-
rbms train -d ./data/MNIST.h5 --subset_labels 0 1
|
|
91
|
+
rbms train -d ./data/MNIST.h5 --subset_labels 0 1 \
|
|
92
|
+
--num_hiddens 200 --gibbs_steps 100 --num_chains 2000 \
|
|
93
|
+
--num_updates 10000 --filename ./RBM_MNIST01.h5
|
|
66
94
|
```
|
|
67
95
|
|
|
68
96
|
## Documentation
|
|
@@ -73,16 +101,17 @@ The documentation is available at [https://dsysdml.github.io/rbms/](https://dsys
|
|
|
73
101
|
|
|
74
102
|
We welcome contributions to the development of TorchRBM. Here's how you can contribute:
|
|
75
103
|
|
|
76
|
-
|
|
77
|
-
|
|
78
|
-
|
|
79
|
-
|
|
80
|
-
|
|
81
|
-
|
|
104
|
+
- **Fork the Repository**: Fork the main repository and make your changes.
|
|
105
|
+
- **Propose a Merge Request**: Once your changes are complete, propose a merge request.
|
|
106
|
+
- **Testing**: Ensure your code is tested using the pytest framework. Include your tests in the tests folder.
|
|
107
|
+
- **Code Style**: Follow the pre-commit configuration for code style.
|
|
108
|
+
- **Documentation**: Document your code using the Google docstring style and provide type hints .
|
|
109
|
+
- **Dependencies**: Avoid adding extra dependencies unless absolutely necessary.
|
|
82
110
|
|
|
83
111
|
## Support
|
|
84
112
|
|
|
85
113
|
If you encounter any issues, please open an issue on the main repository.
|
|
86
114
|
|
|
87
115
|
## License
|
|
88
|
-
|
|
116
|
+
|
|
117
|
+
`rbms` is released under the MIT License.
|
rbms-0.2.0/README.md
ADDED
|
@@ -0,0 +1,95 @@
|
|
|
1
|
+
[](https://codecov.io/gh/DsysDML/rbms)
|
|
2
|
+
[](https://github.com/pre-commit/pre-commit)
|
|
3
|
+

|
|
4
|
+
|
|
5
|
+
# RBMs
|
|
6
|
+
|
|
7
|
+
## Overview
|
|
8
|
+
|
|
9
|
+
`rbms` is a GPU-accelerated package designed to train and analyze Restricted Boltzmann Machines (RBMs). It is intended for students and researchers who need an efficient tool for working with RBMs.
|
|
10
|
+
Features:
|
|
11
|
+
|
|
12
|
+
- GPU Acceleration through PyTorch.
|
|
13
|
+
- Multiple RBM Types: Supports Bernoulli-Bernoulli RBM and Potts-Bernoulli RBM.
|
|
14
|
+
- Extensible Design: Provides an abstract class RBM with methods that can be implemented for new types of RBMs, minimizing the need to reimplement training algorithms, analysis methods, and sampling methods.
|
|
15
|
+
|
|
16
|
+
## Installation
|
|
17
|
+
|
|
18
|
+
### PyPI
|
|
19
|
+
|
|
20
|
+
`rbms` can be installed from PyPI:
|
|
21
|
+
|
|
22
|
+
```bash
|
|
23
|
+
pip install rbms
|
|
24
|
+
```
|
|
25
|
+
|
|
26
|
+
The last version on PyPI corresponds to the main branch of this repo. To update the installed package run
|
|
27
|
+
|
|
28
|
+
```bash
|
|
29
|
+
pip install rbms --upgrade
|
|
30
|
+
```
|
|
31
|
+
|
|
32
|
+
### Github
|
|
33
|
+
|
|
34
|
+
If you want a more up-to-date version or wish to contribute, you can install the package directly from this repository:
|
|
35
|
+
|
|
36
|
+
```bash
|
|
37
|
+
git clone git@github.com:DsysDML/rbms.git
|
|
38
|
+
cd rbms && pip install -e .
|
|
39
|
+
```
|
|
40
|
+
|
|
41
|
+
If you want to update the package run
|
|
42
|
+
|
|
43
|
+
```bash
|
|
44
|
+
git pull
|
|
45
|
+
```
|
|
46
|
+
|
|
47
|
+
from inside the cloned repo.
|
|
48
|
+
|
|
49
|
+
## Dependencies
|
|
50
|
+
|
|
51
|
+
The dependencies are included in the pyproject.toml file and should be compiled in the wheel distribution. If you have a GPU, make sure to install PyTorch with GPU support.
|
|
52
|
+
|
|
53
|
+
## Usage
|
|
54
|
+
|
|
55
|
+
Main Classes and Functions:
|
|
56
|
+
|
|
57
|
+
- `EBM`: An abstract class that provides the basic structure and methods for Energy-Based Models (EBMs).
|
|
58
|
+
- `RBM`: An abstract class that provides the basic structure and methods for RBMs.
|
|
59
|
+
- `BBRBM`: A concrete implementation of the Bernoulli-Bernoulli RBM.
|
|
60
|
+
- `PBRBM`: A concrete implementation of the Potts-Bernoulli RBM.
|
|
61
|
+
- `rbms.sampling`: Submodule for sampling methods.
|
|
62
|
+
- `rbms.training`: Submodule for training algorithms.
|
|
63
|
+
|
|
64
|
+
## Basic Example
|
|
65
|
+
|
|
66
|
+
Train a RBM on MNIST-01 with PCD-100 for 10 000 steps, using 200 hidden nodes and 2000 permanent chains:
|
|
67
|
+
|
|
68
|
+
```bash
|
|
69
|
+
rbms train -d ./data/MNIST.h5 --subset_labels 0 1 \
|
|
70
|
+
--num_hiddens 200 --gibbs_steps 100 --num_chains 2000 \
|
|
71
|
+
--num_updates 10000 --filename ./RBM_MNIST01.h5
|
|
72
|
+
```
|
|
73
|
+
|
|
74
|
+
## Documentation
|
|
75
|
+
|
|
76
|
+
The documentation is available at [https://dsysdml.github.io/rbms/](https://dsysdml.github.io/rbms/)
|
|
77
|
+
|
|
78
|
+
## Contributing
|
|
79
|
+
|
|
80
|
+
We welcome contributions to the development of TorchRBM. Here's how you can contribute:
|
|
81
|
+
|
|
82
|
+
- **Fork the Repository**: Fork the main repository and make your changes.
|
|
83
|
+
- **Propose a Merge Request**: Once your changes are complete, propose a merge request.
|
|
84
|
+
- **Testing**: Ensure your code is tested using the pytest framework. Include your tests in the tests folder.
|
|
85
|
+
- **Code Style**: Follow the pre-commit configuration for code style.
|
|
86
|
+
- **Documentation**: Document your code using the Google docstring style and provide type hints .
|
|
87
|
+
- **Dependencies**: Avoid adding extra dependencies unless absolutely necessary.
|
|
88
|
+
|
|
89
|
+
## Support
|
|
90
|
+
|
|
91
|
+
If you encounter any issues, please open an issue on the main repository.
|
|
92
|
+
|
|
93
|
+
## License
|
|
94
|
+
|
|
95
|
+
`rbms` is released under the MIT License.
|
|
Binary file
|
|
Binary file
|
|
Binary file
|
|
Binary file
|
|
Binary file
|
|
@@ -613,6 +613,20 @@
|
|
|
613
613
|
"module_short": "rbms",
|
|
614
614
|
"name": "RBM.init_chains"
|
|
615
615
|
},
|
|
616
|
+
{
|
|
617
|
+
"is_class": false,
|
|
618
|
+
"is_explicit": false,
|
|
619
|
+
"module": "rbms.classes",
|
|
620
|
+
"module_short": "rbms.classes",
|
|
621
|
+
"name": "EBM.init_chains"
|
|
622
|
+
},
|
|
623
|
+
{
|
|
624
|
+
"is_class": false,
|
|
625
|
+
"is_explicit": false,
|
|
626
|
+
"module": "rbms",
|
|
627
|
+
"module_short": "rbms",
|
|
628
|
+
"name": "EBM.init_chains"
|
|
629
|
+
},
|
|
616
630
|
{
|
|
617
631
|
"is_class": false,
|
|
618
632
|
"is_explicit": false,
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
21caaed3471ff131aeda19a34e85cc5d
|
|
@@ -19,7 +19,7 @@
|
|
|
19
19
|
|
|
20
20
|
|
|
21
21
|
Analyze a post-training Bernoulli-Bernoulli RBM
|
|
22
|
-
|
|
22
|
+
===========================
|
|
23
23
|
|
|
24
24
|
This script shows how to analyze the RBM after having trained it.
|
|
25
25
|
|
|
@@ -281,7 +281,7 @@ We can now sample those chains and compare again the distribution
|
|
|
281
281
|
.. GENERATED FROM PYTHON SOURCE LINES 115-118
|
|
282
282
|
|
|
283
283
|
Compute the AIS estimation of the log-likelihood.
|
|
284
|
-
|
|
284
|
+
------------------------------
|
|
285
285
|
For now, we only looked at a qualitative evaluation of the model
|
|
286
286
|
|
|
287
287
|
.. GENERATED FROM PYTHON SOURCE LINES 118-128
|
|
@@ -306,7 +306,7 @@ For now, we only looked at a qualitative evaluation of the model
|
|
|
306
306
|
|
|
307
307
|
.. code-block:: none
|
|
308
308
|
|
|
309
|
-
-
|
|
309
|
+
-387.64599609375
|
|
310
310
|
|
|
311
311
|
|
|
312
312
|
|
|
@@ -314,7 +314,7 @@ For now, we only looked at a qualitative evaluation of the model
|
|
|
314
314
|
|
|
315
315
|
.. rst-class:: sphx-glr-timing
|
|
316
316
|
|
|
317
|
-
**Total running time of the script:** (0 minutes
|
|
317
|
+
**Total running time of the script:** (0 minutes 8.271 seconds)
|
|
318
318
|
|
|
319
319
|
|
|
320
320
|
.. _sphx_glr_download_auto_examples_plot_generated_samples.py:
|
|
@@ -6,7 +6,7 @@
|
|
|
6
6
|
|
|
7
7
|
Computation times
|
|
8
8
|
=================
|
|
9
|
-
**00:
|
|
9
|
+
**00:08.271** total execution time for 1 file **from auto_examples**:
|
|
10
10
|
|
|
11
11
|
.. container::
|
|
12
12
|
|
|
@@ -33,5 +33,5 @@ Computation times
|
|
|
33
33
|
- Time
|
|
34
34
|
- Mem (MB)
|
|
35
35
|
* - :ref:`sphx_glr_auto_examples_plot_generated_samples.py` (``plot_generated_samples.py``)
|
|
36
|
-
- 00:
|
|
36
|
+
- 00:08.271
|
|
37
37
|
- 0.0
|
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
|
|
1
|
+
# Welcome to Restricted Boltzmann Machines (RBM) in PyTorch
|
|
2
2
|
|
|
3
3
|
`rbms` is a GPU-accelerated package designed to train and analyze Restricted Boltzmann Machines (RBMs). It is intended for students and researchers who need an efficient tool for working with RBMs.
|
|
4
4
|
|
|
@@ -8,23 +8,22 @@
|
|
|
8
8
|
- **Multiple RBM Types**: Supports Bernoulli-Bernoulli RBM and Potts-Bernoulli RBM.
|
|
9
9
|
- **Extensible Design**: Provides an abstract class `RBM` with methods that can be implemented for new types of RBMs, minimizing the need to reimplement training algorithms, analysis methods, and sampling methods.
|
|
10
10
|
|
|
11
|
-
|
|
12
11
|
## Installation
|
|
13
12
|
|
|
14
|
-
To install
|
|
13
|
+
To install `rbms`, you can use pip:
|
|
15
14
|
|
|
16
15
|
```bash
|
|
17
|
-
pip install
|
|
16
|
+
pip install rbms
|
|
18
17
|
```
|
|
19
18
|
|
|
20
|
-
## What's New
|
|
21
|
-
|
|
19
|
+
## [What's New](whats_new.rst)
|
|
20
|
+
|
|
21
|
+
### Version 0.2
|
|
22
|
+
|
|
23
|
+
- EBM abstract class for more generic EBMs.
|
|
24
|
+
-
|
|
22
25
|
|
|
23
|
-
|
|
24
|
-
- Support for Bernoulli-Bernoulli RBM and Potts-Bernoulli RBM.
|
|
25
|
-
- Abstract class RBM for easy extension to new RBM types.
|
|
26
|
-
- GPU acceleration for training and analysis.
|
|
27
|
-
- Comprehensive sampling and training submodules.
|
|
26
|
+
See [Releases](https://github.com/DsysDML/rbms/releases)
|
|
28
27
|
|
|
29
28
|
## [Restricted Boltzmann Machines](rbm.md)
|
|
30
29
|
|
|
@@ -34,10 +33,7 @@ pip install .
|
|
|
34
33
|
|
|
35
34
|
## [API](api.md)
|
|
36
35
|
|
|
37
|
-
|
|
38
|
-
|
|
39
36
|
```{toctree}
|
|
40
37
|
:caption: 'Contents:'
|
|
41
38
|
:maxdepth: 4
|
|
42
39
|
```
|
|
43
|
-
|
|
@@ -6,7 +6,7 @@
|
|
|
6
6
|
|
|
7
7
|
Computation times
|
|
8
8
|
=================
|
|
9
|
-
**00:
|
|
9
|
+
**00:08.271** total execution time for 1 file **from all galleries**:
|
|
10
10
|
|
|
11
11
|
.. container::
|
|
12
12
|
|
|
@@ -33,5 +33,5 @@ Computation times
|
|
|
33
33
|
- Time
|
|
34
34
|
- Mem (MB)
|
|
35
35
|
* - :ref:`sphx_glr_auto_examples_plot_generated_samples.py` (``../../examples/plot_generated_samples.py``)
|
|
36
|
-
- 00:
|
|
36
|
+
- 00:08.271
|
|
37
37
|
- 0.0
|
|
@@ -31,9 +31,8 @@ with h5py.File("my_dataset.h5", "w") as f:
|
|
|
31
31
|
## Command line arguments
|
|
32
32
|
|
|
33
33
|
When running a script requiring a dataset, you have several flags:
|
|
34
|
-
- `-d` or `--data` The path to the dataset
|
|
35
|
-
- `--
|
|
36
|
-
- `--subset_labels` For datasets with labels specified, it allows to select only a subset of the dataset matching the specified labels. For example setting `-d MNIST --subset_labels 0 1` will load the $0$ and $1$ digits of the MNIST dataset (what we refer to as `MNIST-01` dataset)
|
|
34
|
+
- `-d` or `--data` The path to the dataset (e.g.`-d ./data/dataset.h5`).
|
|
35
|
+
- `--subset_labels` For datasets with labels specified, it allows to select only a subset of the dataset matching the specified labels. For example setting `-d MNIST --subset_labels 0 1` will load the $0$ and $1$ digits of the MNIST dataset (what we refer to as `MNIST-01` dataset). If specified in a dataset without labels, the full dataset is selected.
|
|
37
36
|
- `--train_size` The proportion of the dataset to use as training dataset. It can go from $0$ to $1$ and the default is $0.6$.
|
|
38
37
|
- `--test_size` Same as above, but for the test dataset. Defaults to $1-$ train size
|
|
39
38
|
- `--use_weights` Compute the weights for protein sequences.
|
|
@@ -0,0 +1,51 @@
|
|
|
1
|
+
# Train a RBM
|
|
2
|
+
|
|
3
|
+
To train a RBM, you need to use the rbms train script.
|
|
4
|
+
|
|
5
|
+
## RBM hyperparameters
|
|
6
|
+
|
|
7
|
+
- `--num_hiddens` Number of hidden nodes for the RBM. Setting it to $20$ or less allows to recover the exact log-likelihood of the model by enumerating on all hidden configurations.
|
|
8
|
+
- `--batch_size` Batch size, defaults to $2000$. Changing the batch size has an impact on the noise in the estimation of the positive term of the gradient. Setting it to a low value can lead to a very bad estimation and a bad training, but setting it too high can lead to an exact gradient, losing the benefits of the SGD (and remain trapped in a local minima for example).
|
|
9
|
+
- `--num_chains` Number of parallel chains, defaults to $2000$. Setting it to a much higher value than the batch size does not provide benefits, since it only impacts the estimation of the negative term of the gradient.
|
|
10
|
+
- `--gibbs_steps` Number of sampling steps performed at each gradient update. The $k$ in PCD-$k$.
|
|
11
|
+
- `--learning_rate` Learning rate. Defaults to $0.01$, setting a larger learning rate often leads to instability.
|
|
12
|
+
- `--num_updates` The training time is indexed on the number of gradient updates performed and not the number of epochs.
|
|
13
|
+
- `--beta` The inverse temperature to use during training (Defaults to $1$ and should not be changed)
|
|
14
|
+
|
|
15
|
+
## Save options
|
|
16
|
+
|
|
17
|
+
- `--filename` The path to the hdf5 archive to save the RBM during training. It will overwrite previously existing file.
|
|
18
|
+
- `--n_save` The number of machines to save during the training.
|
|
19
|
+
- `--spacing` Can be `exp` or `linear`, defaults to `exp`. When `exp` is selected, the time between the save of two models will increase exponentially. (It will look good in log-scale). When `linear` is selected, the time between the save of two models will be constant.
|
|
20
|
+
Saving lots of models can quickly become the computational bottleneck, leading to long execution times.
|
|
21
|
+
- `--log` For now it is deprecated so you don't care about it.
|
|
22
|
+
- `--acc_ptt` Target acceptance rate. Defaults to $0.25$. Models will be saved when the acceptance rate between two consecutive models when sampling them using PTT drops below this threshold.
|
|
23
|
+
- `--acc_ll` Same as before but defaults to $0.75$. This allows to have two different schemes when saving models.
|
|
24
|
+
|
|
25
|
+
## PyTorch options
|
|
26
|
+
|
|
27
|
+
- `--device` The device on which to run the computations. Follows the PyTorch semantic so you can select which GPU to use with 'cuda:1' for example.
|
|
28
|
+
- `--dtype` The dtype of all the tensors. can be `int`, `double` or `float`. The default is `float` which corresponds to `torch.float32`.
|
|
29
|
+
|
|
30
|
+
## Example
|
|
31
|
+
|
|
32
|
+
The command I typically use to train a RBM on `MNIST-01` will be
|
|
33
|
+
|
|
34
|
+
```bash
|
|
35
|
+
rbms train -d ./path/to/MNIST.h5 --subset_labels 0 1 \
|
|
36
|
+
--filename output/rbm/MNIST01_from_scratch.h5 --num_updates 10000 \
|
|
37
|
+
--n_save 50 --spacing exp --num_hiddens 20 --batch_size 2000 --num_chains 2000 \
|
|
38
|
+
--learning_rate 0.01 --device cuda --dtype float
|
|
39
|
+
```
|
|
40
|
+
|
|
41
|
+
# Restore the training
|
|
42
|
+
|
|
43
|
+
If you want to continue the training of a RBM (be it one recovered from a RCM or a previously trained one), you can use the same scripts. The differences are that you should add the `--restore` flags. Also some arguments are not useful anymore and can be safely ignored:
|
|
44
|
+
|
|
45
|
+
- `--num_hiddens`
|
|
46
|
+
- `--batch_size`
|
|
47
|
+
- `--gibbs_steps`
|
|
48
|
+
- `--learning_rate`
|
|
49
|
+
- `--num_chains`
|
|
50
|
+
|
|
51
|
+
Finally the updates will be added to the same archive you provide as an input through `--filename`. If the `--restore` flag is set, then the file will **not** be overwritten.
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
"""
|
|
2
2
|
Analyze a post-training Bernoulli-Bernoulli RBM
|
|
3
|
-
|
|
3
|
+
===========================
|
|
4
4
|
|
|
5
5
|
This script shows how to analyze the RBM after having trained it.
|
|
6
6
|
"""
|
|
@@ -113,7 +113,7 @@ plot_PCA(
|
|
|
113
113
|
|
|
114
114
|
# %%
|
|
115
115
|
# Compute the AIS estimation of the log-likelihood.
|
|
116
|
-
#
|
|
116
|
+
# ------------------------------
|
|
117
117
|
# For now, we only looked at a qualitative evaluation of the model
|
|
118
118
|
from rbms.partition_function.ais import compute_partition_function_ais
|
|
119
119
|
from rbms.utils import compute_log_likelihood
|
|
@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
|
|
|
4
4
|
|
|
5
5
|
[project]
|
|
6
6
|
name = "rbms"
|
|
7
|
-
version = "0.
|
|
7
|
+
version = "0.2.0"
|
|
8
8
|
authors = [
|
|
9
9
|
{name="Nicolas Béreux", email="nicolas.bereux@gmail.com"},
|
|
10
10
|
{name="Aurélien Decelle"},
|
|
@@ -22,9 +22,7 @@ requires-python = ">=3.11"
|
|
|
22
22
|
dependencies = [
|
|
23
23
|
"h5py>=3.12.0",
|
|
24
24
|
"numpy>=2.0.0, <=2.1.2",
|
|
25
|
-
"seaborn>=0.13.0",
|
|
26
25
|
"matplotlib>=3.8.0",
|
|
27
|
-
"scikit-learn==1.5.2",
|
|
28
26
|
"torch>=2.0.0, <=2.5.0",
|
|
29
27
|
"tqdm>=4.65.0",
|
|
30
28
|
]
|