-
Notifications
You must be signed in to change notification settings - Fork 0
/
data.py
443 lines (363 loc) · 16.4 KB
/
data.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
import os
import os.path as osp
import json
import numpy as np
import pandas as pd
from functools import reduce
from tqdm import tqdm
import dgl
from dgl.data.utils import load_graphs, save_graphs, Subset
import torch
# Collate function for ordinary graph classification
def collate_dgl(samples):
samples = list(map(list, zip(*samples)))
if len(samples) == 3:
graphs, energies, has_dft = samples
batched_graph = dgl.batch(graphs)
return batched_graph, torch.stack(energies), torch.stack(has_dft)
elif len(samples) == 4:
graphs, ep_energies, ep_indicators, has_dft = samples
batched_graph = dgl.batch(graphs)
return batched_graph, torch.stack(ep_energies), torch.stack(ep_indicators), torch.stack(has_dft)
else:
raise RuntimeError('Wrong batch with len {}.'.format(len(samples)))
class PreTrainRegDataset(object):
def __init__(self, cache_dir, potentials=['sw'], cut_per_atom_energy=200, graph_type='mg'):
self.potentials = ['energy_'+p for p in potentials]
self.cache_dir = cache_dir
self.graph_type = graph_type
self.cut_per_atom_energy = cut_per_atom_energy
self._load_graphs()
def get_config_index(self, configs):
indices = list()
for c in configs:
if c in self.config2id:
indices.append(self.config2id[c])
indices = torch.LongTensor(indices)
return indices
def _load_graphs(self):
print('Loading...')
self.graphs, properties = load_graphs(
osp.join(self.cache_dir, 'graphs')
)
label_df = pd.read_csv(
osp.join(self.cache_dir, 'labels.csv'),
dtype={'configs': str}
)
self.config_names = label_df['configs']
self.config2id = {config:i for i, config in enumerate(self.config_names)}
print('Number of graphs {}.'.format(len(self.graphs)))
if self.graph_type == 'hmg':
self.num_atoms = [g.nodes['atom'].data['atom_is_contributing'].sum() for g in self.graphs]
elif self.graph_type == 'mg':
self.num_atoms = [g.ndata['is_contributing'].sum() for g in self.graphs]
else:
raise ValueError('Graph type could only be hmg or mg.')
# filter ep energy values
for pname in self.potentials:
for i in range(len(self.graphs)):
per_atom_energy = label_df[pname].values[i] / self.num_atoms[i]
if per_atom_energy > 200:
print('Dropping ep {} on config {} with per atom energy {:.4f} and total energy {:.4f}.'.format(
pname, self.config_names[i], per_atom_energy, label_df[pname].values[i]))
# if per atom energy is larger than 200, drop
label_df[pname].values[i] = 200001
self.ep_energies = torch.FloatTensor(label_df[self.potentials].values)
def __getitem__(self, idx):
'''Get datapoint with index'''
if isinstance(idx, int):
return self.graphs[idx], self.ep_energies[idx]
elif torch.is_tensor(idx) and idx.dtype == torch.long:
if idx.dim() == 0:
return self.graphs[idx], self.ep_energies[idx]
elif idx.dim() == 1:
return Subset(self, idx.cpu())
raise IndexError(
'Only integers and long are valid '
'indices (got {}).'.format(type(idx).__name__))
def __len__(self):
'''Length of the dataset
Returns
-------
int
Length of Dataset
'''
return len(self.graphs)
class PreTrainClsDataset(object):
def __init__(self, cache_dir, potentials=['sw'], cut_per_atom_energy=1, graph_type='mg'):
self.potentials = ['energy_'+p for p in potentials]
self.cache_dir = cache_dir
self.graph_type = graph_type
self.cut_per_atom_energy = cut_per_atom_energy
self._load_graphs()
def get_config_index(self, configs):
indices = list()
for c in configs:
if c in self.config2id:
indices.append(self.config2id[c])
indices = torch.LongTensor(indices)
return indices
def get_configs(self, indices):
configs = [self.config_names[idx] for idx in indices.tolist()]
return configs
def _load_graphs(self):
print('Loading...')
self.graphs, _ = load_graphs(
osp.join(self.cache_dir, 'graphs')
)
label_df = pd.read_csv(
osp.join(self.cache_dir, 'labels.csv'),
dtype={'configs': str}
)
self.config_names = label_df['configs']
self.config2id = {config:i for i, config in enumerate(self.config_names)}
print('Number of graphs {}.'.format(len(self.graphs)))
if self.graph_type == 'hmg':
self.num_atoms = [g.nodes['atom'].data['atom_is_contributing'].sum() for g in self.graphs]
elif self.graph_type == 'mg':
self.num_atoms = [g.ndata['is_contributing'].sum() for g in self.graphs]
else:
raise ValueError('Graph type could only be hmg or mg.')
self.dft = label_df['energy_reference'].values
self.ep = label_df[self.potentials].values
num_invalid_ep_configs = 0
best_potentials = np.abs(self.ep - self.dft[:, None]).argmin(axis=1)
for i, pid in enumerate(best_potentials):
# test if the instance can be correctly predicted by an ep
if np.isnan(self.dft[i]):
best_potentials[i] = len(self.potentials)
elif np.abs(self.ep[i, pid] - self.dft[i]) / self.num_atoms[i] > self.cut_per_atom_energy:
best_potentials[i] = len(self.potentials)
num_invalid_ep_configs += 1
print('Found {}/{} invalid configs without good ep predictions.'.format(num_invalid_ep_configs, len(best_potentials)))
self.ep_indices = torch.LongTensor(best_potentials)
def get_best_ep_preds(self, config_indices, ep_logits):
configs = self.get_configs(config_indices)
ep_indices = ep_logits.argmax(axis=1)
valid_configs = list()
valid_ep_preds = list()
valid_dfts = list()
for c, pid in zip(configs, ep_indices):
if pid != len(self.potentials):
valid_configs.append(c)
valid_ep_preds.append(self.ep[self.config2id[c], pid])
valid_dfts.append(self.dft[self.config2id[c]])
df = pd.DataFrame({
'configs': valid_configs,
'dft_preds': valid_ep_preds,
'dft': valid_dfts
})
print('Found {} valid configurations.'.format(len(valid_configs)))
return df
def __getitem__(self, idx):
'''Get datapoint with index'''
if isinstance(idx, int):
return self.graphs[idx], self.ep_indices[idx]
elif torch.is_tensor(idx) and idx.dtype == torch.long:
if idx.dim() == 0:
return self.graphs[idx], self.ep_indices[idx]
elif idx.dim() == 1:
return Subset(self, idx.cpu())
raise IndexError(
'Only integers and long are valid '
'indices (got {}).'.format(type(idx).__name__))
def __len__(self):
'''Length of the dataset
Returns
-------
int
Length of Dataset
'''
return len(self.graphs)
class PreTrainDataset(object):
def __init__(self, cache_dir, potentials=['sw'], cut_per_atom_energy=1, graph_type='mg'):
self.potentials = ['energy_'+p for p in potentials]
self.cache_dir = cache_dir
self.graph_type = graph_type
self.cut_per_atom_energy = cut_per_atom_energy
self._load_graphs()
def get_config_index(self, configs):
indices = list()
for c in configs:
if c in self.config2id:
indices.append(self.config2id[c])
indices = torch.LongTensor(indices)
return indices
def get_configs(self, indices):
configs = [self.config_names[idx] for idx in indices.tolist()]
return configs
def _load_graphs(self):
print('Loading...')
self.graphs, _ = load_graphs(
osp.join(self.cache_dir, 'graphs')
)
label_df = pd.read_csv(
osp.join(self.cache_dir, 'labels.csv'),
dtype={'configs': str}
)
self.config_names = label_df['configs']
self.config2id = {config:i for i, config in enumerate(self.config_names)}
print('Number of graphs {}.'.format(len(self.graphs)))
if self.graph_type == 'hmg':
self.num_atoms = [g.nodes['atom'].data['atom_is_contributing'].sum() for g in self.graphs]
elif self.graph_type == 'mg':
self.num_atoms = [g.ndata['is_contributing'].sum() for g in self.graphs]
else:
raise ValueError('Graph type could only be hmg or mg.')
# filter ep energy values
for pname in self.potentials:
for i in range(len(self.graphs)):
per_atom_energy = label_df[pname].values[i] / self.num_atoms[i]
if per_atom_energy > 200:
print('Dropping ep {} on config {} with per atom energy {:.4f} and total energy {:.4f}.'.format(
pname, self.config_names[i], per_atom_energy, label_df[pname].values[i]))
# if per atom energy is larger than 200, drop
label_df[pname].values[i] = 200001
self.dft = label_df['energy_reference'].values
self.ep = label_df[self.potentials].values
num_invalid_ep_configs = 0
best_potentials = np.abs(self.ep - self.dft[:, None]).argmin(axis=1)
for i, pid in enumerate(best_potentials):
# test if the instance can be correctly predicted by an ep
if np.isnan(self.dft[i]):
best_potentials[i] = len(self.potentials)
elif np.abs(self.ep[i, pid] - self.dft[i]) / self.num_atoms[i] > self.cut_per_atom_energy:
best_potentials[i] = len(self.potentials)
num_invalid_ep_configs += 1
print('Found {}/{} invalid configs without good ep predictions.'.format(num_invalid_ep_configs, len(best_potentials)))
self.ep_indices = torch.LongTensor(best_potentials)
self.ep_energies = torch.FloatTensor(label_df[self.potentials].values)
self.has_dft = torch.ones_like(self.ep_indices, dtype=torch.bool)
def update_pretrain_configs(self, pretrain_configs):
indices = self.get_config_index(pretrain_configs)
self.has_dft[indices] = False
def get_best_ep_preds(self, config_indices, ep_logits):
configs = self.get_configs(config_indices)
ep_indices = ep_logits.argmax(axis=1)
valid_configs = list()
valid_ep = list()
valid_ep_preds = list()
valid_dfts = list()
for c, pid in zip(configs, ep_indices):
if pid != len(self.potentials):
valid_configs.append(c)
valid_ep_preds.append(self.ep[self.config2id[c], pid])
valid_dfts.append(self.dft[self.config2id[c]])
valid_ep.append(self.potentials[pid])
df = pd.DataFrame({
'configs': valid_configs,
'dft_preds': valid_ep_preds,
'dft': valid_dfts,
'ep': valid_ep
})
print('Found {} valid configurations.'.format(len(valid_configs)))
return df
def __getitem__(self, idx):
'''Get datapoint with index'''
if isinstance(idx, int):
return self.graphs[idx], self.ep_energies[idx], self.ep_indices[idx], self.has_dft[idx]
elif torch.is_tensor(idx) and idx.dtype == torch.long:
if idx.dim() == 0:
return self.graphs[idx], self.ep_energies[idx], self.ep_indices[idx], self.has_dft[idx]
elif idx.dim() == 1:
return Subset(self, idx.cpu())
raise IndexError(
'Only integers and long are valid '
'indices (got {}).'.format(type(idx).__name__))
def __len__(self):
'''Length of the dataset
Returns
-------
int
Length of Dataset
'''
return len(self.graphs)
class FineTuneDataset(object):
def __init__(self, cache_dir):
self.cache_dir = cache_dir
self._load_graphs()
def get_config_index(self, configs):
indices = list()
for c in configs:
if c in self.config2id:
indices.append(self.config2id[c])
indices = torch.LongTensor(indices)
return indices
def get_configs(self, indices):
configs = [self.config_names[idx] for idx in indices.tolist()]
return configs
def get_statistics(self, config_ids):
mean = self.target[config_ids].mean()
std = self.target[config_ids].std()
return mean, std
def purterb_dft(self, configs, purterbed_labels, mean, std, mode='gaussian'):
"""Purterb the dft energies with gaussian noise
Args:
configs (_type_): _description_
mean (_type_): _description_
std (_type_): _description_
"""
purterb_indices = self.get_config_index(configs)
self.has_dft[purterb_indices] = False
if mode == 'ep_pred':
print('Purterbing {} configurations with predicted best performing potentials.'.format(len(configs)))
self.target[purterb_indices] = torch.FloatTensor(purterbed_labels)
elif mode == 'ep':
print('Purterbing {} configurations with best performing potentials.'.format(len(configs)))
label_df = pd.read_csv(
osp.join(self.cache_dir, 'labels.csv'),
dtype={'configs':str}
)
potentials = [p for p in list(label_df.columns) if 'energy' in p and p != 'energy_reference']
print('Available potentials: {}'.format(','.join(potentials)))
best_potentials = np.abs(label_df[potentials].values - label_df['energy_reference'].values[:, None]).argmin(axis=1)
best_potential_preds = label_df[potentials].values[np.arange(best_potentials.shape[0]), best_potentials]
print('Best potential prediction mae: {:.4f}'.format((best_potential_preds - label_df['energy_reference'].values).mean()))
best_potential_preds = torch.FloatTensor(best_potential_preds)
self.target[purterb_indices] = best_potential_preds[purterb_indices]
print('Purterbation mean: {:.4f}, std: {:.4f}'.format(best_potential_preds[purterb_indices].mean().item(), best_potential_preds[purterb_indices].std().item()))
elif mode == 'gaussian':
print('Purterbing {} configurations with mean {:.4f}, std {:.4f}.'.format(len(configs), mean, std))
self.target[purterb_indices] = self.target[purterb_indices] + \
torch.normal(mean, std, size=self.target[purterb_indices].size())
else:
print('Unknown purterb mode, not purterbing.')
def _load_graphs(self):
print('Loading...')
self.graphs, _ = load_graphs(
osp.join(self.cache_dir, 'graphs')
)
label_df = pd.read_csv(
osp.join(self.cache_dir, 'labels.csv'),
dtype={'configs':str}
)
self.config_names = label_df['configs'].values
self.config2id = {config:i for i, config in enumerate(self.config_names)}
print('Setting DFT as the target.')
self.target = torch.FloatTensor(label_df['energy_reference'].values)
self.has_dft = torch.ones_like(self.target, dtype=torch.bool)
def __getitem__(self, idx):
'''Get datapoint with index'''
if isinstance(idx, int):
return self.graphs[idx], self.target[idx], self.has_dft[idx]
elif torch.is_tensor(idx) and idx.dtype == torch.long:
if idx.dim() == 0:
return self.graphs[idx], self.target[idx], self.has_dft[idx]
elif idx.dim() == 1:
return Subset(self, idx.cpu())
raise IndexError(
'Only integers and long are valid '
'indices (got {}).'.format(type(idx).__name__))
def __len__(self):
'''Length of the dataset
Returns
-------
int
Length of Dataset
'''
return len(self.graphs)
if __name__ == '__main__':
dataset = FineTuneDataset(
cache_dir='CachedData/soapnet/Al'
)
print(collate_dgl([dataset[0], dataset[1], dataset[2]]))