質問編集履歴
1
該当コードを追加しました。
test
CHANGED
|
File without changes
|
test
CHANGED
|
@@ -66,6 +66,324 @@
|
|
|
66
66
|
|
|
67
67
|
|
|
68
68
|
|
|
69
|
+
以下長いですが、コードになります。
|
|
70
|
+
|
|
71
|
+
```datasets.py
|
|
72
|
+
|
|
73
|
+
def create_dataset(dataset_name,
|
|
74
|
+
|
|
75
|
+
compute_node_feature_stats=True,
|
|
76
|
+
|
|
77
|
+
node_feature_stats_filename=None,
|
|
78
|
+
|
|
79
|
+
**dataset_params):
|
|
80
|
+
|
|
81
|
+
|
|
82
|
+
|
|
83
|
+
if (dataset_name == 'shrec_16'):
|
|
84
|
+
|
|
85
|
+
dataset = Shrec2016DualPrimal(**dataset_params)
|
|
86
|
+
|
|
87
|
+
elif (dataset_name == 'cubes'):
|
|
88
|
+
|
|
89
|
+
dataset = CubesDualPrimal(**dataset_params)
|
|
90
|
+
|
|
91
|
+
elif (dataset_name == 'coseg'):
|
|
92
|
+
|
|
93
|
+
dataset = CosegDualPrimal(**dataset_params)
|
|
94
|
+
|
|
95
|
+
elif (dataset_name == 'human_seg'):
|
|
96
|
+
|
|
97
|
+
dataset = HumanSegDualPrimal(**dataset_params)
|
|
98
|
+
|
|
99
|
+
elif(dataset_name == 'vessel'):
|
|
100
|
+
|
|
101
|
+
dataset = []
|
|
102
|
+
|
|
103
|
+
else:
|
|
104
|
+
|
|
105
|
+
raise KeyError(
|
|
106
|
+
|
|
107
|
+
f"No known dataset can be generated with the name '{dataset_name}'."
|
|
108
|
+
|
|
109
|
+
)
|
|
110
|
+
|
|
111
|
+
|
|
112
|
+
|
|
113
|
+
node_statistics = None
|
|
114
|
+
|
|
115
|
+
print(dataset)
|
|
116
|
+
|
|
117
|
+
if (compute_node_feature_stats):
|
|
118
|
+
|
|
119
|
+
dataset_params = dataset.input_parameters
|
|
120
|
+
|
|
121
|
+
(primal_graph_mean, primal_graph_std, dual_graph_mean,
|
|
122
|
+
|
|
123
|
+
dual_graph_std) = compute_mean_and_std(
|
|
124
|
+
|
|
125
|
+
dataset=dataset,
|
|
126
|
+
|
|
127
|
+
dataset_params=dataset_params,
|
|
128
|
+
|
|
129
|
+
filename=node_feature_stats_filename)
|
|
130
|
+
|
|
131
|
+
node_statistics = (primal_graph_mean, primal_graph_std, dual_graph_mean,
|
|
132
|
+
|
|
133
|
+
dual_graph_std)
|
|
134
|
+
|
|
135
|
+
return dataset, node_statistics
|
|
136
|
+
|
|
137
|
+
|
|
138
|
+
|
|
139
|
+
|
|
140
|
+
|
|
141
|
+
def compute_mean_and_std(dataset=None, dataset_params=None, filename=None):
|
|
142
|
+
|
|
143
|
+
|
|
144
|
+
|
|
145
|
+
if (dataset_params is not None):
|
|
146
|
+
|
|
147
|
+
for param_keyword in ['mean', 'std']:
|
|
148
|
+
|
|
149
|
+
for graph_keyword in ['primal', 'dual']:
|
|
150
|
+
|
|
151
|
+
keyword = f"{graph_keyword}_{param_keyword}"
|
|
152
|
+
|
|
153
|
+
if (keyword in dataset_params):
|
|
154
|
+
|
|
155
|
+
raise KeyError(
|
|
156
|
+
|
|
157
|
+
f"The parameters of the input dataset already contain "
|
|
158
|
+
|
|
159
|
+
f"an entry '{keyword}'. Exiting.")
|
|
160
|
+
|
|
161
|
+
file_exists = False
|
|
162
|
+
|
|
163
|
+
if (filename is not None):
|
|
164
|
+
|
|
165
|
+
# Load the data from disk, if the file exists.
|
|
166
|
+
|
|
167
|
+
if (os.path.exists(filename)):
|
|
168
|
+
|
|
169
|
+
file_exists = True
|
|
170
|
+
|
|
171
|
+
if (file_exists):
|
|
172
|
+
|
|
173
|
+
assert (dataset_params is not None)
|
|
174
|
+
|
|
175
|
+
assert (isinstance(dataset_params, dict))
|
|
176
|
+
|
|
177
|
+
try:
|
|
178
|
+
|
|
179
|
+
with open(filename, "rb") as f:
|
|
180
|
+
|
|
181
|
+
data_from_disk = pkl.load(f)
|
|
182
|
+
|
|
183
|
+
except IOError:
|
|
184
|
+
|
|
185
|
+
raise IOError(f"Error loading cache mean-std file '{filename}'. "
|
|
186
|
+
|
|
187
|
+
"Exiting.")
|
|
188
|
+
|
|
189
|
+
# Check that the file contains the mean and standard deviation.
|
|
190
|
+
|
|
191
|
+
for keyword in ['primal', 'dual']:
|
|
192
|
+
|
|
193
|
+
if (f'{keyword}_mean' not in data_from_disk):
|
|
194
|
+
|
|
195
|
+
raise KeyError(
|
|
196
|
+
|
|
197
|
+
f"Cached file '{filename}' does not contain the mean of "
|
|
198
|
+
|
|
199
|
+
f"the {keyword}-graph node features. Exiting.")
|
|
200
|
+
|
|
201
|
+
if (f'{keyword}_std' not in data_from_disk):
|
|
202
|
+
|
|
203
|
+
raise KeyError(
|
|
204
|
+
|
|
205
|
+
f"Cached file '{filename}' does not contain the standard "
|
|
206
|
+
|
|
207
|
+
f"deviation of the {keyword}-graph node features. Exiting.")
|
|
208
|
+
|
|
209
|
+
# Check that the size of the dataset is compatible.
|
|
210
|
+
|
|
211
|
+
try:
|
|
212
|
+
|
|
213
|
+
size_dataset_of_file = data_from_disk['dataset_size']
|
|
214
|
+
|
|
215
|
+
except KeyError:
|
|
216
|
+
|
|
217
|
+
raise KeyError(
|
|
218
|
+
|
|
219
|
+
f"Cached file '{filename}' does not contain the dataset size. "
|
|
220
|
+
|
|
221
|
+
f"Exiting.")
|
|
222
|
+
|
|
223
|
+
current_dataset_size = len(dataset)
|
|
224
|
+
|
|
225
|
+
if (size_dataset_of_file != current_dataset_size):
|
|
226
|
+
|
|
227
|
+
warnings.warn("Please note that the current dataset has size "
|
|
228
|
+
|
|
229
|
+
f"{current_dataset_size}, whereas the cached file ("
|
|
230
|
+
|
|
231
|
+
f"'{filename}') was generated from a dataset of size "
|
|
232
|
+
|
|
233
|
+
f"{size_dataset_of_file}.")
|
|
234
|
+
|
|
235
|
+
|
|
236
|
+
|
|
237
|
+
# Check that the parameters match.
|
|
238
|
+
|
|
239
|
+
for param_name, param_value in dataset_params.items():
|
|
240
|
+
|
|
241
|
+
if (param_name not in data_from_disk):
|
|
242
|
+
|
|
243
|
+
raise KeyError(
|
|
244
|
+
|
|
245
|
+
f"Could not find dataset parameter {param_name} in the "
|
|
246
|
+
|
|
247
|
+
f"cached file '{filename}'. Please provide a different "
|
|
248
|
+
|
|
249
|
+
"filename.")
|
|
250
|
+
|
|
251
|
+
else:
|
|
252
|
+
|
|
253
|
+
if (data_from_disk[param_name] != param_value):
|
|
254
|
+
|
|
255
|
+
raise ValueError(
|
|
256
|
+
|
|
257
|
+
f"Cached file '{filename}' is incompatible with "
|
|
258
|
+
|
|
259
|
+
f"current dataset. Expected parameter {param_name} to "
|
|
260
|
+
|
|
261
|
+
f"be {param_value}, found "
|
|
262
|
+
|
|
263
|
+
f"{data_from_disk[param_name]}. Please provide a "
|
|
264
|
+
|
|
265
|
+
"different filename.")
|
|
266
|
+
|
|
267
|
+
for cached_param_name in dataset_params.keys():
|
|
268
|
+
|
|
269
|
+
if (cached_param_name in [
|
|
270
|
+
|
|
271
|
+
'primal_mean', 'primal_std', 'dual_mean', 'dual_std'
|
|
272
|
+
|
|
273
|
+
]):
|
|
274
|
+
|
|
275
|
+
continue
|
|
276
|
+
|
|
277
|
+
if (cached_param_name not in dataset_params):
|
|
278
|
+
|
|
279
|
+
raise KeyError(
|
|
280
|
+
|
|
281
|
+
f"Cached file '{filename}' is incompatible with "
|
|
282
|
+
|
|
283
|
+
"current dataset, as it contains parameter "
|
|
284
|
+
|
|
285
|
+
f"{cached_param_name}, which is missing in the input "
|
|
286
|
+
|
|
287
|
+
"dataset. Please provide a different filename.")
|
|
288
|
+
|
|
289
|
+
# Return the cached data.
|
|
290
|
+
|
|
291
|
+
primal_graph_mean = data_from_disk['primal_mean']
|
|
292
|
+
|
|
293
|
+
primal_graph_std = data_from_disk['primal_std']
|
|
294
|
+
|
|
295
|
+
dual_graph_mean = data_from_disk['dual_mean']
|
|
296
|
+
|
|
297
|
+
dual_graph_std = data_from_disk['dual_std']
|
|
298
|
+
|
|
299
|
+
else:
|
|
300
|
+
|
|
301
|
+
# Compute the mean and standard deviation of the node features from
|
|
302
|
+
|
|
303
|
+
# scratch.
|
|
304
|
+
|
|
305
|
+
primal_graph_xs = torch.empty([0, dataset[0][0].x.shape[1]])
|
|
306
|
+
|
|
307
|
+
print('len',primal_graph_xs.size())
|
|
308
|
+
|
|
309
|
+
dual_graph_xs = torch.empty([0, dataset[0][1].x.shape[1]])
|
|
310
|
+
|
|
311
|
+
for sample_idx, (primal_graph, dual_graph, _, _) in enumerate(dataset):
|
|
312
|
+
|
|
313
|
+
primal_graph_xs = torch.cat([primal_graph_xs, primal_graph.x])
|
|
314
|
+
|
|
315
|
+
dual_graph_xs = torch.cat([dual_graph_xs, dual_graph.x])
|
|
316
|
+
|
|
317
|
+
assert (len(dataset) == sample_idx + 1)
|
|
318
|
+
|
|
319
|
+
primal_graph_mean = primal_graph_xs.mean(axis=0).numpy()
|
|
320
|
+
|
|
321
|
+
primal_graph_std = primal_graph_xs.std(axis=0).numpy()
|
|
322
|
+
|
|
323
|
+
dual_graph_mean = dual_graph_xs.mean(axis=0).numpy()
|
|
324
|
+
|
|
325
|
+
dual_graph_std = dual_graph_xs.std(axis=0).numpy()
|
|
326
|
+
|
|
327
|
+
assert (np.all(
|
|
328
|
+
|
|
329
|
+
primal_graph_std > 10 * np.finfo(primal_graph_std.dtype).eps))
|
|
330
|
+
|
|
331
|
+
assert (np.all(
|
|
332
|
+
|
|
333
|
+
dual_graph_std > 10 * np.finfo(dual_graph_std.dtype).eps))
|
|
334
|
+
|
|
335
|
+
|
|
336
|
+
|
|
337
|
+
if (filename is not None):
|
|
338
|
+
|
|
339
|
+
# Save the values to file, together with the dataset parameters and
|
|
340
|
+
|
|
341
|
+
# the dataset size, if required.
|
|
342
|
+
|
|
343
|
+
if (dataset_params is None):
|
|
344
|
+
|
|
345
|
+
dataset_params = {}
|
|
346
|
+
|
|
347
|
+
output_values = {
|
|
348
|
+
|
|
349
|
+
**dataset_params, 'primal_mean': primal_graph_mean,
|
|
350
|
+
|
|
351
|
+
'primal_std': primal_graph_std,
|
|
352
|
+
|
|
353
|
+
'dual_mean': dual_graph_mean,
|
|
354
|
+
|
|
355
|
+
'dual_std': dual_graph_std,
|
|
356
|
+
|
|
357
|
+
'dataset_size': sample_idx + 1
|
|
358
|
+
|
|
359
|
+
}
|
|
360
|
+
|
|
361
|
+
try:
|
|
362
|
+
|
|
363
|
+
with open(filename, 'wb') as f:
|
|
364
|
+
|
|
365
|
+
pkl.dump(output_values, f)
|
|
366
|
+
|
|
367
|
+
except IOError:
|
|
368
|
+
|
|
369
|
+
raise IOError(
|
|
370
|
+
|
|
371
|
+
"Unable to save mean-std data to file at location "
|
|
372
|
+
|
|
373
|
+
f"{filename}.")
|
|
374
|
+
|
|
375
|
+
|
|
376
|
+
|
|
377
|
+
return (primal_graph_mean, primal_graph_std, dual_graph_mean,
|
|
378
|
+
|
|
379
|
+
dual_graph_std)
|
|
380
|
+
|
|
381
|
+
|
|
382
|
+
|
|
383
|
+
```
|
|
384
|
+
|
|
385
|
+
|
|
386
|
+
|
|
69
387
|
分かる方がいらっしゃいましたら回答いただけますと幸いです。
|
|
70
388
|
|
|
71
389
|
よろしくお願い致します。
|