forked from flyIchtus/score_crawl
-
Notifications
You must be signed in to change notification settings - Fork 0
/
evaluation_frontend.py
521 lines (315 loc) · 16.9 KB
/
evaluation_frontend.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
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Thu Jun 30 11:04:19 2022
@author: brochetc
Metrics computation automation
"""
import pickle
from glob import glob
import numpy as np
from multiprocessing import Pool
from collections import defaultdict
from score_crawl.configurate import Experiment
import score_crawl.evaluation_backend as backend
import metrics4arome as metrics
########### standard parameters #####
num_proc = backend.num_proc
var_dict = backend.var_dict
data_dir = backend.data_dir_0
#####################################
class EnsembleMetricsCalculator(Experiment) :
def __init__(self, expe_config, add_name) :
super().__init__(expe_config)
self.add_name = add_name
def __print__(self):
super().__print__()
###########################################################################
######################### Main class method ###############################
###########################################################################
def estimation(self, metrics_list, program, parallel=False, standalone=False, real=False) :
"""
estimate all metrics contained in metrics_list on training runs
using specific strategies
-> parallel or sequential estimation
-> distance metrics or standalone metrics
-> on real samples only (if distance metrics)
Inputs :
metrics_list : list, the list of metrics to be computed
program : dict of shape {int : (int, int)}
contains all the informations about sample numbers and number of repeats
#### WARNING : in this case, each sample is supposed to represent a given ensemble
keys index the repeats
values[0] index the type of dataset manipulation
(either dividing the same dataset into parts, or selecting only one portion)
values[1] indicate the number of samples to use in the computation
Note : -> for tests on training dynamics, only 1 repeat is assumed
(at the moment)
-> for tests on self-distances on real datasets,
many repeats are possible (to account for test variance
or test different sampling sizes)
parallel, standalone, real : bool, the flags defining the estimation
strategy
Returns :
None
dumps the results in a pickle file
"""
########### sanity checks
if standalone and not parallel :
raise(ValueError, 'Estimation for standalone metric should be done in parallel')
if standalone :
assert set(metrics_list) <= metrics.standalone_metrics
else :
assert set(metrics_list) <= metrics.distance_metrics
for metric in metrics_list :
assert hasattr(metrics, metric)
########################
self.program = program
if parallel :
if standalone :
name='_standalone_metrics_'
if real :
func = lambda m_list : self.parallelEstimation_standAlone(m_list, option='real')
else :
func = self.parallelEstimation_standAlone
else :
name='_distance_metrics_'
if real :
func = self.parallelEstimation_realVreal
else :
func = self.parallelEstimation_realVSfake
else :
name='_distance_metrics_'
if real :
func = self.sequentialEstimation_realVSreal
else :
func = self.sequentialEstimation_realVSfake
results=func(metrics_list)
N_samples_set = [self.program[i][1] for i in range(len(program))]
N_samples_name = '_'.join([str(n) for n in N_samples_set])
if real :
temp_log_dir = self.log_dir
self.log_dir = backend.data_dir
dumpfile = self.log_dir + self.add_name+name + str(N_samples_name)+'.p'
if real :
self.log_dir = temp_log_dir
pickle.dump(results, open(dumpfile, 'wb'))
###########################################################################
############################ Estimation strategies ######################
###########################################################################
def parallelEstimation_realVSfake(self, metrics_list):
"""
makes a list of datasets with each item of self.steps
and use multiple processes to evaluate the metric in parallel on each
item.
The metric must be a distance metric and the data should be real / fake
Inputs :
metric : str, the metric to evaluate
Returns :
res : ndarray, the results array (precise shape defined by the metric)
"""
RES = {}
for i0 in self.program.keys() :
data_list=[]
for step in self.steps:
#getting first (and only) item of the random real dataset program
dataset_r = backend.build_datasets(data_dir, self.program)[i0]
N_samples = self.program[i0][1]
#getting files to analyze from fake dataset
files = glob(self.data_dir_f+"_Fsample_"+str(step)+'_*.npy')
data_list.append((metrics_list, {'real':dataset_r,'fake': files},\
N_samples, N_samples,\
self.VI, self.VI_f, self.CI, step, data_dir))
with Pool(num_proc) as p :
res = p.map(backend.eval_distance_metrics, data_list)
## some cuisine to produce a rightly formatted dictionary
ind_list=[]
d_res = defaultdict(list)
for res_index in res :
index = res_index[1]
res0 = res_index[0]
for k, v in res0.items():
d_res[k].append(v)
ind_list.append(index)
for k in d_res.keys():
d_res[k]= [x for _,x in sorted(zip(ind_list, d_res[k]))]
res = { k : np.concatenate([np.expand_dims(v[i], axis=0) \
for i in range(len(self.steps))], axis=0).squeeze()
for k,v in d_res.items()}
RES[i0] = res
if i0==1 :
return res
else :
return RES
def sequentialEstimation_realVSfake(self, metrics_list):
"""
Iterates the evaluation of the metric on each item of self.steps
The metric must be a distance metric and the data should be real / fake
Inputs :
metric : str, the metric to evaluate
Returns :
N_samples : int, the number of samples used in evaluation
res : ndarray, the results array (precise shape defined by the metric)
"""
RES = {}
for i0 in self.program.keys():
res = []
for step in self.steps:
#getting first (and only) item of the random real dataset program
dataset_r = backend.build_datasets(data_dir, self.program)[i0]
N_samples=self.program[i0][1]
#getting files to analyze from fake dataset
files = glob(self.data_dir_f+"_Fsample_"+str(step)+'_*.npy')
data = (metrics_list, {'real':dataset_r,'fake': files},\
N_samples, N_samples,
self.VI, self.VI_f, self.CI, step, data_dir)
res.append(backend.eval_distance_metrics(data))
## some cuisine to produce a rightly formatted dictionary
d_res = defaultdict(list)
for res_index in res :
res0 = res_index[0]
for k, v in res0.items():
d_res[k].append(v)
res = { k : np.concatenate([np.expand_dims(v[i], axis=0) \
for i in range(len(self.steps))], axis=0).squeeze()
for k,v in d_res.items()}
RES[i0] = res
if i0==1 :
return res
else :
return RES
def parallelEstimation_realVSreal(self, metric):
"""
makes a list of datasets with each pair of real datasets contained
in self.program.
Use multiple processes to evaluate the metric in parallel on each
item.
The metric must be a distance metric and the data should be real / real
Inputs :
metric : str, the metric to evaluate
Returns :
N_samples : int, the number of samples used in evaluation
res : ndarray, the results array (precise shape defined by the metric)
"""
datasets = backend.build_datasets(data_dir, self.program)
data_list = []
#getting the two random datasets programs
for i in range(len(datasets)):
N_samples = self.program[i][1]
data_list.append((metric,
{'real0':datasets[i][0],'real1': datasets[i][1]},
N_samples, N_samples,
self.VI, self.VI, self.CI,i, data_dir))
with Pool(num_proc) as p :
res = p.map(backend.eval_distance_metrics, data_list)
## some cuisine to produce a rightly formatted dictionary
ind_list=[]
d_res = defaultdict(list)
for res_index in res :
index = res_index[1]
res0 = res_index[0]
for k, v in res0.items():
d_res[k].append(v)
ind_list.append(index)
for k in d_res.keys():
d_res[k]= [x for _, x in sorted(zip(ind_list, d_res[k]))]
res = { k : np.concatenate([np.expand_dims(v[i], axis=0) \
for i in range(len(self.steps))], axis=0).squeeze()
for k,v in d_res.items()}
return res
def sequentialEstimation_realVSreal(self, metric):
"""
Iterates the evaluation of the metric on each item of pair of real datasets
defined in self.program.
The metric must be a distance metric and the data should be real / fake
Inputs :
metric : str, the metric to evaluate
Returns :
N_samples : int, the number of samples used in evaluation
res : ndarray, the results array (precise shape defined by the metric)
"""
#getting first (and only) item of the random real dataset program
datasets = backend.build_datasets(data_dir, self.program)
for i in range(len(datasets)):
N_samples = self.program[i][1]
data=(metric, {'real0':datasets[i][0],'real1': datasets[i][1]},
N_samples, N_samples,
self.VI, self.VI, self.CI, i, data_dir)
if i==0: res = [backend.eval_distance_metrics(data)]
else :
res.append(backend.eval_distance_metrics(data))
## some cuisine to produce a rightly formatted dictionary
d_res = defaultdict(list)
for res_index in res :
res0 = res_index[0]
for k, v in res0.items():
d_res[k].append(v)
res = { k : np.concatenate([np.expand_dims(v[i], axis=0) \
for i in range(len(self.steps))], axis=0).squeeze()
for k,v in d_res.items()}
return res
def parallelEstimation_standAlone(self, metrics_list, option='fake'):
"""
makes a list of datasets with each dataset contained
in self.program (case option =real) or directly from data files
(case option =fake)
Use multiple processes to evaluate the metric in parallel on each
item.
The metric must be a standalone metric.
Inputs :
metric : str, the metric to evaluate
Returns :
res : ndarray, the results array (precise shape defined by the metric)
"""
if option=='real':
self.steps =[0]
dataset_r = backend.build_datasets(data_dir, self.program)
data_list = [(metrics_list, dataset_r, self.program[i][1],
self.VI, self.VI, self.CI, i, option, data_dir) \
for i, dataset in enumerate(dataset_r)]
with Pool(num_proc) as p :
res = p.map(backend.global_dataset_eval, data_list)
ind_list=[]
d_res = defaultdict(list)
for res_index in res :
index = res_index[1]
res0 = res_index[0]
for k, v in res0.items():
d_res[k].append(v)
ind_list.append(index)
for k in d_res.keys():
d_res[k]= [x for _, x in sorted(zip(ind_list, d_res[k]))]
res = { k : np.concatenate([np.expand_dims(v[i], axis=0) \
for i in range(len(self.steps))], axis=0).squeeze()
for k,v in d_res.items()}
return res
elif option=='fake' :
RES = {}
for i0 in self.program.keys():
N_samples = self.program[i0][1]
data_list = []
for j,step in enumerate(self.steps):
#getting files to analyze from fake dataset
if option=='fake' :
files = glob(self.data_dir_f+"_Fsample_"+str(step)+'_*.npy')
data_list.append((metrics_list, files, N_samples,
self.VI, self.VI_f, self.CI, step, option, data_dir))
with Pool(num_proc) as p :
res = p.map(backend.global_dataset_eval, data_list)
ind_list=[]
d_res = defaultdict(list)
for res_index in res :
index = res_index[1]
res0 = res_index[0]
for k, v in res0.items():
d_res[k].append(v)
ind_list.append(index)
for k in d_res.keys():
d_res[k]= [x for _, x in sorted(zip(ind_list, d_res[k]))]
res = { k : np.concatenate([np.expand_dims(v[i], axis=0) \
for i in range(len(self.steps))], axis=0).squeeze()
for k,v in d_res.items()}
RES[i0] = res
if i0==1 :
return res
return RES