rbms 0.2.0__tar.gz → 0.3.0__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.
Files changed (106) hide show
  1. {rbms-0.2.0 → rbms-0.3.0}/PKG-INFO +1 -1
  2. {rbms-0.2.0 → rbms-0.3.0}/pyproject.toml +1 -1
  3. {rbms-0.2.0 → rbms-0.3.0}/rbms/bernoulli_bernoulli/implement.py +9 -3
  4. {rbms-0.2.0 → rbms-0.3.0}/rbms/classes.py +1 -0
  5. {rbms-0.2.0 → rbms-0.3.0}/rbms/dataset/__init__.py +3 -2
  6. {rbms-0.2.0 → rbms-0.3.0}/rbms/dataset/dataset_class.py +54 -4
  7. {rbms-0.2.0 → rbms-0.3.0}/rbms/dataset/load_fasta.py +1 -3
  8. {rbms-0.2.0 → rbms-0.3.0}/rbms/dataset/load_h5.py +8 -0
  9. {rbms-0.2.0 → rbms-0.3.0}/rbms/dataset/parser.py +6 -0
  10. {rbms-0.2.0 → rbms-0.3.0}/rbms/io.py +4 -1
  11. {rbms-0.2.0 → rbms-0.3.0}/rbms/parser.py +41 -11
  12. {rbms-0.2.0 → rbms-0.3.0}/rbms/potts_bernoulli/classes.py +0 -3
  13. {rbms-0.2.0 → rbms-0.3.0}/rbms/potts_bernoulli/implement.py +15 -8
  14. {rbms-0.2.0 → rbms-0.3.0}/rbms/scripts/train_rbm.py +9 -7
  15. {rbms-0.2.0 → rbms-0.3.0}/rbms/training/pcd.py +22 -10
  16. {rbms-0.2.0 → rbms-0.3.0}/rbms/training/utils.py +27 -4
  17. {rbms-0.2.0 → rbms-0.3.0}/rbms/utils.py +5 -5
  18. {rbms-0.2.0 → rbms-0.3.0}/tests/conftest.py +5 -1
  19. {rbms-0.2.0 → rbms-0.3.0}/tests/unit_test/bernoulli_bernoulli/test_bernoulli_utils_bbrbm.py +6 -0
  20. {rbms-0.2.0 → rbms-0.3.0}/tests/unit_test/potts_bernoulli/test_utils_pbrbm.py +6 -0
  21. {rbms-0.2.0 → rbms-0.3.0}/tests/unit_test/training/test_pcd.py +0 -3
  22. {rbms-0.2.0 → rbms-0.3.0}/tests/use_cases/test_bbrbm.py +1 -0
  23. {rbms-0.2.0 → rbms-0.3.0}/tests/use_cases/test_pbrbm.py +2 -0
  24. rbms-0.2.0/.github/workflows/docs.yaml +0 -49
  25. rbms-0.2.0/docs/Makefile +0 -20
  26. rbms-0.2.0/docs/make.bat +0 -35
  27. rbms-0.2.0/docs/source/api.md +0 -151
  28. rbms-0.2.0/docs/source/auto_examples/images/sphx_glr_plot_generated_samples_001.png +0 -0
  29. rbms-0.2.0/docs/source/auto_examples/images/sphx_glr_plot_generated_samples_002.png +0 -0
  30. rbms-0.2.0/docs/source/auto_examples/images/sphx_glr_plot_generated_samples_003.png +0 -0
  31. rbms-0.2.0/docs/source/auto_examples/images/sphx_glr_plot_generated_samples_004.png +0 -0
  32. rbms-0.2.0/docs/source/auto_examples/images/thumb/sphx_glr_plot_generated_samples_thumb.png +0 -0
  33. rbms-0.2.0/docs/source/auto_examples/index.rst +0 -63
  34. rbms-0.2.0/docs/source/auto_examples/plot_generated_samples.codeobj.json +0 -989
  35. rbms-0.2.0/docs/source/auto_examples/plot_generated_samples.ipynb +0 -187
  36. rbms-0.2.0/docs/source/auto_examples/plot_generated_samples.py +0 -127
  37. rbms-0.2.0/docs/source/auto_examples/plot_generated_samples.py.md5 +0 -1
  38. rbms-0.2.0/docs/source/auto_examples/plot_generated_samples.rst +0 -343
  39. rbms-0.2.0/docs/source/auto_examples/sg_execution_times.rst +0 -37
  40. rbms-0.2.0/docs/source/conf.py +0 -66
  41. rbms-0.2.0/docs/source/examples.md +0 -0
  42. rbms-0.2.0/docs/source/index.md +0 -39
  43. rbms-0.2.0/docs/source/myst.yml +0 -4
  44. rbms-0.2.0/docs/source/rbm.md +0 -131
  45. rbms-0.2.0/docs/source/references.bib +0 -275
  46. rbms-0.2.0/docs/source/sg_execution_times.rst +0 -37
  47. rbms-0.2.0/docs/source/tuto/dataset.md +0 -39
  48. rbms-0.2.0/docs/source/tuto/parallel_tempering.md +0 -1
  49. rbms-0.2.0/docs/source/tuto/train_rbm.md +0 -51
  50. rbms-0.2.0/docs/source/tutorials.md +0 -12
  51. rbms-0.2.0/docs/source/whats_new.rst +0 -0
  52. rbms-0.2.0/examples/GALLERY_HEADER.rst +0 -4
  53. rbms-0.2.0/examples/plot_generated_samples.py +0 -127
  54. {rbms-0.2.0 → rbms-0.3.0}/.github/workflows/codecov.yaml +0 -0
  55. {rbms-0.2.0 → rbms-0.3.0}/.github/workflows/test.yaml +0 -0
  56. {rbms-0.2.0 → rbms-0.3.0}/.github/workflows/whats_new.yaml +0 -0
  57. {rbms-0.2.0 → rbms-0.3.0}/.gitignore +0 -0
  58. {rbms-0.2.0 → rbms-0.3.0}/LICENSE +0 -0
  59. {rbms-0.2.0 → rbms-0.3.0}/MANIFEST.in +0 -0
  60. {rbms-0.2.0 → rbms-0.3.0}/README.md +0 -0
  61. {rbms-0.2.0 → rbms-0.3.0}/rbms/__init__.py +0 -0
  62. {rbms-0.2.0 → rbms-0.3.0}/rbms/bernoulli_bernoulli/__init__.py +0 -0
  63. {rbms-0.2.0 → rbms-0.3.0}/rbms/bernoulli_bernoulli/classes.py +0 -0
  64. {rbms-0.2.0 → rbms-0.3.0}/rbms/bernoulli_bernoulli/functional.py +0 -0
  65. {rbms-0.2.0 → rbms-0.3.0}/rbms/const.py +0 -0
  66. {rbms-0.2.0 → rbms-0.3.0}/rbms/custom_fn.py +0 -0
  67. {rbms-0.2.0 → rbms-0.3.0}/rbms/dataset/fasta_utils.py +0 -0
  68. {rbms-0.2.0 → rbms-0.3.0}/rbms/dataset/utils.py +0 -0
  69. {rbms-0.2.0 → rbms-0.3.0}/rbms/map_model.py +0 -0
  70. {rbms-0.2.0 → rbms-0.3.0}/rbms/metrics/__init__.py +0 -0
  71. {rbms-0.2.0 → rbms-0.3.0}/rbms/metrics/aats.py +0 -0
  72. {rbms-0.2.0 → rbms-0.3.0}/rbms/partition_function/ais.py +0 -0
  73. {rbms-0.2.0 → rbms-0.3.0}/rbms/partition_function/exact.py +0 -0
  74. {rbms-0.2.0 → rbms-0.3.0}/rbms/plot.py +0 -0
  75. {rbms-0.2.0 → rbms-0.3.0}/rbms/potts_bernoulli/__init__.py +0 -0
  76. {rbms-0.2.0 → rbms-0.3.0}/rbms/potts_bernoulli/functional.py +0 -0
  77. {rbms-0.2.0 → rbms-0.3.0}/rbms/potts_bernoulli/tools.py +0 -0
  78. {rbms-0.2.0 → rbms-0.3.0}/rbms/potts_bernoulli/utils.py +0 -0
  79. {rbms-0.2.0 → rbms-0.3.0}/rbms/sampling/gibbs.py +0 -0
  80. {rbms-0.2.0 → rbms-0.3.0}/rbms/sampling/pt.py +0 -0
  81. {rbms-0.2.0 → rbms-0.3.0}/rbms/scripts/__init__.py +0 -0
  82. {rbms-0.2.0 → rbms-0.3.0}/rbms/scripts/entrypoint.py +0 -0
  83. {rbms-0.2.0 → rbms-0.3.0}/rbms/scripts/pt_sampling.py +0 -0
  84. {rbms-0.2.0 → rbms-0.3.0}/tests/unit_test/bernoulli_bernoulli/implement/test_compute_energy_bbrbm.py +0 -0
  85. {rbms-0.2.0 → rbms-0.3.0}/tests/unit_test/bernoulli_bernoulli/implement/test_compute_energy_hiddens_bbrbm.py +0 -0
  86. {rbms-0.2.0 → rbms-0.3.0}/tests/unit_test/bernoulli_bernoulli/implement/test_compute_energy_visibles_bbrbm.py +0 -0
  87. {rbms-0.2.0 → rbms-0.3.0}/tests/unit_test/bernoulli_bernoulli/implement/test_compute_gradient_bbrbm.py +0 -0
  88. {rbms-0.2.0 → rbms-0.3.0}/tests/unit_test/bernoulli_bernoulli/implement/test_init_chains_bbrbm.py +0 -0
  89. {rbms-0.2.0 → rbms-0.3.0}/tests/unit_test/bernoulli_bernoulli/implement/test_init_parameters_bbrbm.py +0 -0
  90. {rbms-0.2.0 → rbms-0.3.0}/tests/unit_test/bernoulli_bernoulli/implement/test_sample_hiddens_bbrbm.py +0 -0
  91. {rbms-0.2.0 → rbms-0.3.0}/tests/unit_test/bernoulli_bernoulli/implement/test_sample_visibles_bbrbm.py +0 -0
  92. {rbms-0.2.0 → rbms-0.3.0}/tests/unit_test/bernoulli_bernoulli/test_functional_bbrbm.py +0 -0
  93. {rbms-0.2.0 → rbms-0.3.0}/tests/unit_test/bernoulli_bernoulli/test_methods_bbrbm.py +0 -0
  94. {rbms-0.2.0 → rbms-0.3.0}/tests/unit_test/metrics/test_aats.py +0 -0
  95. {rbms-0.2.0 → rbms-0.3.0}/tests/unit_test/potts_bernoulli/implement/test_compute_energy_hiddens_pbrbm.py +0 -0
  96. {rbms-0.2.0 → rbms-0.3.0}/tests/unit_test/potts_bernoulli/implement/test_compute_energy_pbrbm.py +0 -0
  97. {rbms-0.2.0 → rbms-0.3.0}/tests/unit_test/potts_bernoulli/implement/test_compute_energy_visibles_pbrbm.py +0 -0
  98. {rbms-0.2.0 → rbms-0.3.0}/tests/unit_test/potts_bernoulli/implement/test_compute_gradient_pbrbm.py +0 -0
  99. {rbms-0.2.0 → rbms-0.3.0}/tests/unit_test/potts_bernoulli/implement/test_init_chains_pbrbm.py +0 -0
  100. {rbms-0.2.0 → rbms-0.3.0}/tests/unit_test/potts_bernoulli/implement/test_init_parameters_pbrbm.py +0 -0
  101. {rbms-0.2.0 → rbms-0.3.0}/tests/unit_test/potts_bernoulli/implement/test_sample_hiddens_pbrbm.py +0 -0
  102. {rbms-0.2.0 → rbms-0.3.0}/tests/unit_test/potts_bernoulli/implement/test_sample_visibles_pbrbm.py +0 -0
  103. {rbms-0.2.0 → rbms-0.3.0}/tests/unit_test/potts_bernoulli/test_functional_pbrbm.py +0 -0
  104. {rbms-0.2.0 → rbms-0.3.0}/tests/unit_test/potts_bernoulli/test_methods_pbrbm.py +0 -0
  105. {rbms-0.2.0 → rbms-0.3.0}/tests/unit_test/test_custom_fn.py +0 -0
  106. {rbms-0.2.0 → rbms-0.3.0}/tests/unit_test/test_utils.py +0 -0
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: rbms
3
- Version: 0.2.0
3
+ Version: 0.3.0
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
@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
4
4
 
5
5
  [project]
6
6
  name = "rbms"
7
- version = "0.2.0"
7
+ version = "0.3.0"
8
8
  authors = [
9
9
  {name="Nicolas Béreux", email="nicolas.bereux@gmail.com"},
10
10
  {name="Aurélien Decelle"},
@@ -47,7 +47,9 @@ def _compute_energy_visibles(
47
47
  ) -> Tensor:
48
48
  field = v @ vbias
49
49
  exponent = hbias + (v @ weight_matrix)
50
- log_term = torch.where(exponent < 10, torch.log(1.0 + torch.exp(exponent)), exponent)
50
+ log_term = torch.where(
51
+ exponent < 10, torch.log(1.0 + torch.exp(exponent)), exponent
52
+ )
51
53
  return -field - log_term.sum(1)
52
54
 
53
55
 
@@ -57,7 +59,9 @@ def _compute_energy_hiddens(
57
59
  ) -> Tensor:
58
60
  field = h @ hbias
59
61
  exponent = vbias + (h @ weight_matrix.T)
60
- log_term = torch.where(exponent < 10, torch.log(1.0 + torch.exp(exponent)), exponent)
62
+ log_term = torch.where(
63
+ exponent < 10, torch.log(1.0 + torch.exp(exponent)), exponent
64
+ )
61
65
  return -field - log_term.sum(1)
62
66
 
63
67
 
@@ -140,7 +144,9 @@ def _init_chains(
140
144
 
141
145
  if start_v is None:
142
146
  # Dummy mean visible
143
- mv = torch.ones(size=(num_samples, num_visibles), device=device, dtype=dtype) / 2
147
+ mv = (
148
+ torch.ones(size=(num_samples, num_visibles), device=device, dtype=dtype) / 2
149
+ )
144
150
  v = torch.bernoulli(mv)
145
151
  else:
146
152
  # Dummy mean visible
@@ -249,4 +249,5 @@ class RBM(EBM):
249
249
  for _ in range(n_steps):
250
250
  new_chains = self.sample_hiddens(chains=new_chains, beta=beta)
251
251
  new_chains = self.sample_visibles(chains=new_chains, beta=beta)
252
+ new_chains = self.sample_hiddens(chains=new_chains, beta=beta)
252
253
  return new_chains
@@ -42,8 +42,6 @@ def load_dataset(
42
42
  alphabet=alphabet,
43
43
  device=device,
44
44
  )
45
- if not binarize:
46
- is_binary = False
47
45
  case _:
48
46
  raise ValueError(
49
47
  """
@@ -53,6 +51,9 @@ def load_dataset(
53
51
  - '.fasta'
54
52
  """
55
53
  )
54
+ if not binarize:
55
+ is_binary = False
56
+
56
57
  # Select subset of dataset w.r.t. labels
57
58
  if subset_labels is not None and labels is not None:
58
59
  data, labels = get_subset_labels(data, labels, subset_labels)
@@ -1,6 +1,6 @@
1
1
  import gzip
2
2
  import textwrap
3
- from typing import Dict, Union
3
+ from typing import Dict, Union, Self, Tuple, Optional
4
4
 
5
5
  import numpy as np
6
6
  import torch
@@ -116,9 +116,59 @@ class RBMDataset(Dataset):
116
116
  for i in pbar:
117
117
  en[i] = len(
118
118
  gzip.compress(
119
- (self.data[torch.randperm(self.data.shape[0])[:num_samples]]).astype(
120
- int
121
- )
119
+ (
120
+ self.data[torch.randperm(self.data.shape[0])[:num_samples]]
121
+ ).astype(int)
122
122
  )
123
123
  )
124
124
  return np.mean(en)
125
+
126
+ def split_train_test(
127
+ self,
128
+ rng: np.random.Generator,
129
+ train_size: float,
130
+ test_size: Optional[float] = None,
131
+ ) -> Tuple[Self, Self | None]:
132
+ num_samples = self.data.shape[0]
133
+ # Shuffle dataset
134
+ permutation_index = rng.permutation(num_samples)
135
+ train_size = int(train_size * num_samples)
136
+ if test_size is not None:
137
+ test_size = int(test_size * num_samples)
138
+ else:
139
+ test_size = num_samples - train_size
140
+ train_dataset = RBMDataset(
141
+ data=self.data[permutation_index[:train_size]].cpu().numpy(),
142
+ labels=self.labels[permutation_index[:train_size]].cpu().numpy(),
143
+ weights=self.weights[permutation_index[:train_size]].cpu().numpy(),
144
+ names=self.names[permutation_index[:train_size]],
145
+ dataset_name=self.dataset_name,
146
+ is_binary=self.is_binary,
147
+ device=self.device,
148
+ dtype=self.dtype,
149
+ )
150
+ test_dataset = None
151
+ if test_size > 0:
152
+ test_dataset = RBMDataset(
153
+ data=self.data[permutation_index[train_size : train_size + test_size]]
154
+ .cpu()
155
+ .numpy(),
156
+ labels=self.labels[
157
+ permutation_index[train_size : train_size + test_size]
158
+ ]
159
+ .cpu()
160
+ .numpy(),
161
+ weights=self.weights[
162
+ permutation_index[train_size : train_size + test_size]
163
+ ]
164
+ .cpu()
165
+ .numpy(),
166
+ names=self.names[
167
+ permutation_index[train_size : train_size + test_size]
168
+ ],
169
+ dataset_name=self.dataset_name,
170
+ is_binary=self.is_binary,
171
+ device=self.device,
172
+ dtype=self.dtype,
173
+ )
174
+ return train_dataset, test_dataset
@@ -52,11 +52,9 @@ def load_FASTA(
52
52
  weights = weights.squeeze(-1)
53
53
  if binarize:
54
54
  num_categories = len(np.unique(dataset))
55
- dataset_oh = (
55
+ dataset = (
56
56
  one_hot(torch.from_numpy(dataset).int(), num_classes=num_categories)
57
57
  .view(dataset.shape[0], -1)
58
58
  .numpy()
59
59
  )
60
- print("ici")
61
- assert np.allclose(dataset, dataset_oh)
62
60
  return dataset, weights, names
@@ -31,6 +31,14 @@ def load_HDF5(
31
31
  f"Ignoring labels since its dimension ({labels.shape[0]}) does not match the number of samples ({dataset.shape[0]})."
32
32
  )
33
33
  labels = None
34
+ # From BEG to Potts
35
+ if not binarize:
36
+ X = (dataset[:, :] + 1) / 2
37
+ for i in range(X.shape[0]):
38
+ for j in range(X.shape[1]):
39
+ if X[i, j] == 0.5:
40
+ X[i, j] = 2
41
+ return X, labels
34
42
  if "cont" not in str(filename.resolve()):
35
43
  unique_values = np.unique(dataset)
36
44
  is_ising = np.all(unique_values == np.array([-1, 1]))
@@ -51,4 +51,10 @@ def add_args_dataset(parser: argparse.ArgumentParser) -> argparse.ArgumentParser
51
51
  action="store_true",
52
52
  help="(Defaults to False). Binarize the dataset.",
53
53
  )
54
+ dataset_args.add_argument(
55
+ "--seed",
56
+ default=None,
57
+ type=int,
58
+ help="(Defaults to None). Seed to split the dataset.",
59
+ )
54
60
  return parser
@@ -130,7 +130,10 @@ def load_model(
130
130
  hyperparameters["learning_rate"] = float(
131
131
  f["hyperparameters"]["learning_rate"][()]
132
132
  )
133
-
133
+ if "seed" in f["hyperparameters"].keys():
134
+ hyperparameters["seed"] = int(f["hyperparameters"]["seed"][()])
135
+ if "train_size" in f["hyperparameters"].keys():
136
+ hyperparameters["train_size"] = int(f["hyperparameters"]["train_size"][()])
134
137
  params = load_params(
135
138
  filename=filename, index=index, device=device, dtype=dtype, map_model=map_model
136
139
  )
@@ -2,6 +2,7 @@ import argparse
2
2
  from typing import Any
3
3
 
4
4
  import torch
5
+ import numpy as np
5
6
 
6
7
 
7
8
  def add_args_pytorch(parser: argparse.ArgumentParser) -> argparse.ArgumentParser:
@@ -38,7 +39,7 @@ def add_args_saves(parser: argparse.ArgumentParser) -> argparse.ArgumentParser:
38
39
  "-o",
39
40
  "--filename",
40
41
  type=str,
41
- default="RBM.h5",
42
+ default=None,
42
43
  help="(Defaults to RBM.h5). Path to the file where to save the model or load if training is restored.",
43
44
  )
44
45
  save_args.add_argument(
@@ -50,14 +51,14 @@ def add_args_saves(parser: argparse.ArgumentParser) -> argparse.ArgumentParser:
50
51
  save_args.add_argument(
51
52
  "--acc_ptt",
52
53
  type=float,
53
- default=0.25,
54
+ default=None,
54
55
  help="(Defaults to 0.25). Minimum PTT acceptance to save configurations for ptt file.",
55
56
  )
56
57
  save_args.add_argument(
57
58
  "--acc_ll",
58
59
  type=float,
59
- default=0.7,
60
- help="(Defaults to 0.75). Minimum PTT acceptance to save configurations for ll file.",
60
+ default=None,
61
+ help="(Defaults to 0.7). Minimum PTT acceptance to save configurations for ll file.",
61
62
  )
62
63
  save_args.add_argument(
63
64
  "--spacing",
@@ -88,42 +89,42 @@ def add_args_rbm(parser: argparse.ArgumentParser) -> argparse.ArgumentParser:
88
89
  rbm_args.add_argument(
89
90
  "--num_hiddens",
90
91
  type=int,
91
- default=100,
92
+ default=None,
92
93
  help="(Defaults to 100). Number of hidden units.",
93
94
  )
94
95
  rbm_args.add_argument(
95
96
  "--batch_size",
96
97
  type=int,
97
- default=2000,
98
+ default=None,
98
99
  help="(Defaults to 2000). Minibatch size.",
99
100
  )
100
101
  rbm_args.add_argument(
101
102
  "--gibbs_steps",
102
103
  type=int,
103
- default=100,
104
+ default=None,
104
105
  help="(Defaults to 100). Number of gibbs steps to perform for each gradient update.",
105
106
  )
106
107
  rbm_args.add_argument(
107
108
  "--learning_rate",
108
109
  type=float,
109
- default=0.01,
110
+ default=None,
110
111
  help="(Defaults to 0.01). Learning rate.",
111
112
  )
112
113
  rbm_args.add_argument(
113
114
  "--num_chains",
114
115
  type=int,
115
- default=2000,
116
+ default=None,
116
117
  help="(Defaults to 2000). Number of parallel chains.",
117
118
  )
118
119
  rbm_args.add_argument(
119
120
  "--num_updates",
120
- default=10_000,
121
+ default=None,
121
122
  type=int,
122
123
  help="(Defaults to 10 000). Number of gradient updates to perform.",
123
124
  )
124
125
  rbm_args.add_argument(
125
126
  "--beta",
126
- default=1.0,
127
+ default=None,
127
128
  type=float,
128
129
  help="(Defaults to 1.0). The inverse temperature of the RBM",
129
130
  )
@@ -166,3 +167,32 @@ def match_args_dtype(args: dict[str, Any]) -> dict[str, Any]:
166
167
  case "double":
167
168
  args["dtype"] = torch.float64
168
169
  return args
170
+
171
+
172
+ default_args: dict[str, Any] = {
173
+ "filename": "RBM.h5",
174
+ "n_save": 50,
175
+ "acc_ptt": 0.25,
176
+ "acc_ll": 0.7,
177
+ "spacing": "exp",
178
+ "log": True,
179
+ "overwrite": True,
180
+ "num_hiddens": 100,
181
+ "batch_size": 2000,
182
+ "gibbs_steps": 100,
183
+ "learning_rate": 0.01,
184
+ "num_chains": 2000,
185
+ "num_updates": 10000,
186
+ "beta": 1.0,
187
+ "restore": False,
188
+ "seed": np.random.randint(0, 1000000000000),
189
+ }
190
+
191
+
192
+ def set_args_default(
193
+ args: dict[str, Any], default_args: dict[str, Any]
194
+ ) -> dict[str, Any]:
195
+ for k, v in args.items():
196
+ if v is None and k in default_args.keys():
197
+ args[k] = default_args[k]
198
+ return args
@@ -159,9 +159,6 @@ class PBRBM(RBM):
159
159
  var_init=var_init,
160
160
  )
161
161
  params = PBRBM(weight_matrix=weight_matrix, vbias=vbias, hbias=hbias)
162
- from rbms.potts_bernoulli.utils import ensure_zero_sum_gauge
163
-
164
- ensure_zero_sum_gauge(params)
165
162
  return params
166
163
 
167
164
  def named_parameters(self):
@@ -158,15 +158,21 @@ def _compute_gradient(
158
158
  h_gen_centered = h_chain
159
159
 
160
160
  # Gradient
161
- grad_weight_matrix = torch.tensordot(
162
- (v_data_one_hot * w_data),
163
- mh_data,
164
- dims=[[0], [0]],
165
- ) / w_data_norm - torch.tensordot(
166
- (v_gen_one_hot * chain_weights),
167
- h_chain,
168
- dims=[[0], [0]],
161
+ grad_weight_matrix = (
162
+ torch.tensordot(
163
+ v_data_centered,
164
+ h_data_centered,
165
+ dims=[[0], [0]],
166
+ )
167
+ / v_data.shape[0]
168
+ - torch.tensordot(
169
+ v_gen_centered,
170
+ h_gen_centered,
171
+ dims=[[0], [0]],
172
+ )
173
+ / v_chain.shape[0]
169
174
  )
175
+
170
176
  grad_vbias = v_data_mean - v_gen_mean
171
177
  grad_hbias = h_data_mean - h_gen_mean
172
178
  weight_matrix.grad.set_(grad_weight_matrix)
@@ -229,4 +235,5 @@ def _init_parameters(
229
235
  )
230
236
  * var_init
231
237
  )
238
+ # print(torch.svd(weight_matrix.reshape(-1, weight_matrix.shape[-1])).S)
232
239
  return vbias, hbias, weight_matrix
@@ -11,6 +11,7 @@ from rbms.parser import (
11
11
  add_args_saves,
12
12
  match_args_dtype,
13
13
  remove_argument,
14
+ default_args,
14
15
  )
15
16
  from rbms.training.pcd import train
16
17
  from rbms.training.utils import get_checkpoints
@@ -30,30 +31,31 @@ def train_rbm(args: dict):
30
31
  checkpoints = get_checkpoints(
31
32
  num_updates=args["num_updates"], n_save=args["n_save"], spacing=args["spacing"]
32
33
  )
33
- train_dataset, test_dataset = load_dataset(
34
+ dataset, _ = load_dataset(
34
35
  dataset_name=args["data"],
35
36
  subset_labels=args["subset_labels"],
36
37
  use_weights=args["use_weights"],
37
38
  alphabet=args["alphabet"],
38
39
  binarize=args["binarize"],
39
- train_size=args["train_size"],
40
- test_size=args["test_size"],
40
+ train_size=1.0,
41
+ test_size=None,
41
42
  device=args["device"],
42
43
  dtype=args["dtype"],
44
+ seed=args["seed"],
43
45
  )
44
- print(train_dataset)
45
- if train_dataset.is_binary:
46
+ print(dataset)
47
+ if dataset.is_binary:
46
48
  model_type = "BBRBM"
47
49
  else:
48
50
  model_type = "PBRBM"
49
51
  train(
50
- dataset=train_dataset,
51
- test_dataset=test_dataset,
52
+ dataset=dataset,
52
53
  model_type=model_type,
53
54
  args=args,
54
55
  dtype=args["dtype"],
55
56
  checkpoints=checkpoints,
56
57
  map_model=map_model,
58
+ default_args=default_args,
57
59
  )
58
60
 
59
61
 
@@ -14,6 +14,8 @@ from rbms.potts_bernoulli.classes import PBRBM
14
14
  from rbms.potts_bernoulli.utils import ensure_zero_sum_gauge
15
15
  from rbms.training.utils import create_machine, setup_training
16
16
  from rbms.utils import check_file_existence, log_to_csv
17
+ from rbms.parser import default_args, set_args_default
18
+ from rbms.potts_bernoulli.utils import ensure_zero_sum_gauge
17
19
 
18
20
 
19
21
  def fit_batch_pcd(
@@ -24,12 +26,12 @@ def fit_batch_pcd(
24
26
  beta: float,
25
27
  centered: bool = True,
26
28
  ) -> Tuple[dict[str, Tensor], dict]:
27
- """Sample the RBM and compute the gradient.
29
+ """Sample the EBM and compute the gradient.
28
30
 
29
31
  Args:
30
32
  batch (Tuple[Tensor, Tensor]): Dataset samples and associated weights.
31
33
  parallel_chains (dict[str, Tensor]): Parallel chains used for gradient computation.
32
- params (RBM): Parameters of the RBM.
34
+ params (EBM): Parameters of the EBM.
33
35
  gibbs_steps (int): Number of Gibbs steps to perform.
34
36
  beta (float): Inverse temperature.
35
37
 
@@ -54,12 +56,12 @@ def fit_batch_pcd(
54
56
 
55
57
  def train(
56
58
  dataset: RBMDataset,
57
- test_dataset: RBMDataset,
58
59
  model_type: str,
59
60
  args: dict,
60
61
  dtype: torch.dtype,
61
62
  checkpoints: np.ndarray,
62
63
  map_model: dict[str, EBM] = map_model,
64
+ default_args: dict = default_args,
63
65
  ) -> None:
64
66
  """Train an EBM.
65
67
 
@@ -71,20 +73,28 @@ def train(
71
73
  dtype (torch.dtype): The data type for the parameters.
72
74
  checkpoints (np.ndarray): An array of checkpoints for saving model states.
73
75
  """
76
+
74
77
  filename = args["filename"]
75
78
  if not (args["overwrite"]):
76
79
  check_file_existence(filename)
77
80
 
78
81
  num_visibles = dataset.get_num_visibles()
79
-
80
82
  # Create a first archive with the initialized model
81
83
  if not (args["restore"]):
84
+ args = set_args_default(args=args, default_args=default_args)
85
+ rng = np.random.default_rng(args["seed"])
86
+ train_dataset, test_dataset = dataset.split_train_test(
87
+ rng, args["train_size"], args["test_size"]
88
+ )
82
89
  params = map_model[model_type].init_parameters(
83
90
  num_hiddens=args["num_hiddens"],
84
- dataset=dataset,
91
+ dataset=train_dataset,
85
92
  device=args["device"],
86
93
  dtype=dtype,
87
94
  )
95
+
96
+ if isinstance(params, PBRBM):
97
+ ensure_zero_sum_gauge(params)
88
98
  create_machine(
89
99
  filename=filename,
90
100
  params=params,
@@ -94,23 +104,25 @@ def train(
94
104
  batch_size=args["batch_size"],
95
105
  gibbs_steps=args["gibbs_steps"],
96
106
  learning_rate=args["learning_rate"],
107
+ train_size=args["train_size"],
97
108
  log=args["log"],
98
109
  flags=["checkpoint"],
110
+ seed=args["seed"],
99
111
  )
100
-
101
112
  (
102
113
  params,
103
114
  parallel_chains,
104
115
  args,
105
- learning_rate,
106
116
  num_updates,
107
117
  start,
108
118
  elapsed_time,
109
119
  log_filename,
110
120
  pbar,
111
- ) = setup_training(args, map_model=map_model)
112
-
113
- optimizer = SGD(params.parameters(), lr=learning_rate, maximize=True)
121
+ train_dataset,
122
+ test_dataset,
123
+ ) = setup_training(args, map_model=map_model, dataset=dataset)
124
+ args = set_args_default(args=args, default_args=default_args)
125
+ optimizer = SGD(params.parameters(), lr=args["learning_rate"], maximize=True)
114
126
 
115
127
  for k, v in args.items():
116
128
  print(f"{k} : {v}")
@@ -13,13 +13,25 @@ from rbms.const import LOG_FILE_HEADER
13
13
  from rbms.io import load_model, save_model
14
14
  from rbms.map_model import map_model
15
15
  from rbms.utils import get_saved_updates
16
+ from rbms.dataset.dataset_class import RBMDataset
16
17
 
17
18
 
18
19
  def setup_training(
19
20
  args: dict,
21
+ dataset: RBMDataset,
20
22
  map_model: dict[str, EBM] = map_model,
21
23
  ) -> Tuple[
22
- EBM, dict[str, Tensor], dict[str, Any], float, int, float, float, pathlib.Path, tqdm
24
+ EBM,
25
+ dict[str, Tensor],
26
+ dict[str, Any],
27
+ float,
28
+ int,
29
+ float,
30
+ float,
31
+ pathlib.Path,
32
+ tqdm,
33
+ RBMDataset,
34
+ RBMDataset,
23
35
  ]:
24
36
  # Retrieve the the number of training updates already performed on the model
25
37
  updates = get_saved_updates(filename=args["filename"])
@@ -40,8 +52,14 @@ def setup_training(
40
52
 
41
53
  # Hyperparameters
42
54
  for k, v in hyperparameters.items():
43
- args[k] = v
44
- learning_rate = args["learning_rate"]
55
+ if v is not None:
56
+ args[k] = v
57
+
58
+ train_dataset, test_dataset = dataset.split_train_test(
59
+ rng=np.random.default_rng(args["seed"]),
60
+ train_size=args["train_size"],
61
+ test_size=args["test_size"],
62
+ )
45
63
 
46
64
  # Open the log file if it exists
47
65
  log_filename = pathlib.Path(args["filename"]).parent / pathlib.Path(
@@ -70,12 +88,13 @@ def setup_training(
70
88
  params,
71
89
  parallel_chains,
72
90
  args,
73
- learning_rate,
74
91
  num_updates,
75
92
  start,
76
93
  elapsed_time,
77
94
  log_filename,
78
95
  pbar,
96
+ train_dataset,
97
+ test_dataset,
79
98
  )
80
99
 
81
100
 
@@ -88,8 +107,10 @@ def create_machine(
88
107
  batch_size: int,
89
108
  gibbs_steps: int,
90
109
  learning_rate: float,
110
+ train_size: float,
91
111
  log: bool,
92
112
  flags: List[str],
113
+ seed: int,
93
114
  ) -> None:
94
115
  """Create a RBM and save it to a new file.
95
116
 
@@ -116,6 +137,8 @@ def create_machine(
116
137
  hyperparameters["gibbs_steps"] = gibbs_steps
117
138
  hyperparameters["filename"] = str(filename)
118
139
  hyperparameters["learning_rate"] = learning_rate
140
+ hyperparameters["train_size"] = True
141
+ hyperparameters["seed"] = seed
119
142
 
120
143
  save_model(
121
144
  filename=filename,
@@ -8,7 +8,7 @@ import numpy as np
8
8
  import torch
9
9
  from torch import Tensor
10
10
 
11
- from rbms.classes import RBM
11
+ from rbms.classes import EBM
12
12
  from rbms.const import LOG_FILE_HEADER
13
13
 
14
14
 
@@ -28,7 +28,7 @@ def get_eigenvalues_history(filename: str):
28
28
  gradient_updates = []
29
29
  eigenvalues = []
30
30
  for key in f.keys():
31
- if "update" in key:
31
+ if "update_" in key:
32
32
  weight_matrix = f[key]["params"]["weight_matrix"][()]
33
33
  weight_matrix = weight_matrix.reshape(-1, weight_matrix.shape[-1])
34
34
  eig = np.linalg.svd(weight_matrix, compute_uv=False)
@@ -56,7 +56,7 @@ def get_saved_updates(filename: str) -> np.ndarray:
56
56
  updates = []
57
57
  with h5py.File(filename, "r") as f:
58
58
  for key in f.keys():
59
- if "update" in key:
59
+ if "update_" in key:
60
60
  update = int(key.replace("update_", ""))
61
61
  updates.append(update)
62
62
  return np.sort(np.array(updates))
@@ -202,7 +202,7 @@ def log_to_csv(logs: dict[str, float], log_file: str) -> None:
202
202
 
203
203
 
204
204
  def compute_log_likelihood(
205
- v_data: Tensor, w_data: Tensor, params: RBM, log_z: float
205
+ v_data: Tensor, w_data: Tensor, params: EBM, log_z: float
206
206
  ) -> float:
207
207
  """Compute the log likelihood of the RBM on the data, given its log partition function.
208
208
 
@@ -295,7 +295,7 @@ def get_flagged_updates(filename: str, flag: str) -> np.ndarray:
295
295
  flagged_updates = []
296
296
  with h5py.File(filename, "r") as f:
297
297
  for key in f.keys():
298
- if "update" in key:
298
+ if "update_" in key:
299
299
  update = int(key.replace("update_", ""))
300
300
  if flag in f[key]["flags"]:
301
301
  if f[key]["flags"][flag][()]:
@@ -13,7 +13,8 @@ pytest.NUM_SAMPLES = 11
13
13
  pytest.NUM_CHAINS = 13
14
14
  pytest.BATCH_SIZE = 17
15
15
  pytest.GIBBS_STEPS = 19
16
-
16
+ pytest.SEED = 42
17
+ pytest.TRAIN_SIZE = 0.6
17
18
 
18
19
  pytest.DEVICE_CUDA = "meta"
19
20
  pytest.LEARNING_RATE = 0.03
@@ -148,6 +149,9 @@ def sample_args(tmp_path):
148
149
  "filename": filename,
149
150
  "beta": 1.0,
150
151
  "overwrite": True,
152
+ "seed": pytest.SEED,
153
+ "train_size": 0.6,
154
+ "test_size": None,
151
155
  }
152
156
 
153
157