-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathcompute_hv.py
362 lines (318 loc) · 11.2 KB
/
compute_hv.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
import torch
from botorch.utils.multi_objective.box_decompositions.dominated import (
DominatedPartitioning,
)
from os import listdir
from os.path import isfile, join
import numpy as np
import pandas as pd
import argparse
import pathlib
qnehvi_success_citeseq = ['upbeat-sweep-100',
'ethereal-sweep-97',
'cool-sweep-94',
'pleasant-sweep-93',
'trim-sweep-90',
'grateful-sweep-89',
'youthful-sweep-85',
'genial-sweep-83',
'vocal-sweep-78',
'apricot-sweep-75',
'quiet-sweep-74',
'skilled-sweep-73',
'avid-sweep-67',
'dry-sweep-66',
'devoted-sweep-63',
'playful-sweep-56',
'dulcet-sweep-52',
'cool-sweep-49',
'sandy-sweep-48',
'ruby-sweep-47',
'still-sweep-46',
'kind-sweep-43',
'vivid-sweep-41',
'lucky-sweep-37',
'devout-sweep-35',
'magic-sweep-27',
'lively-sweep-26',
'peachy-sweep-21',
'youthful-sweep-20',
'pious-sweep-18',
'tough-sweep-16',
'expert-sweep-11',
'true-sweep-4',
'crimson-sweep-5',
'giddy-sweep-3',
'pretty-sweep-1']
qparego_success_citeseq = ['different-sweep-98',
'bumbling-sweep-93',
'true-sweep-92',
'comfy-sweep-90',
'hopeful-sweep-89',
'eternal-sweep-86',
'dauntless-sweep-84',
'gentle-sweep-81',
'proud-sweep-74',
'jumping-sweep-73',
'glowing-sweep-67',
'pleasant-sweep-66',
'major-sweep-65',
'stellar-sweep-57',
'wild-sweep-52',
'frosty-sweep-49',
'crisp-sweep-48',
'divine-sweep-45',
'royal-sweep-42',
'rich-sweep-41',
'golden-sweep-40',
'dry-sweep-38',
'scarlet-sweep-37',
'likely-sweep-35',
'sweet-sweep-30',
'cool-sweep-29',
'peachy-sweep-26',
'glad-sweep-23',
'sunny-sweep-20',
'leafy-sweep-15',
'silvery-sweep-16',
'laced-sweep-12',
'radiant-sweep-10',
'clear-sweep-9',
'rural-sweep-7',
'still-sweep-4',
'fearless-sweep-3',
'true-sweep-1']
qnehvi_success_imc = ['brisk-sweep-96',
'pretty-sweep-95',
'avid-sweep-94',
'winter-sweep-93',
'radiant-sweep-87',
'usual-sweep-86',
'apricot-sweep-85',
'jumping-sweep-84',
'silver-sweep-83',
'trim-sweep-81',
'hopeful-sweep-78',
'spring-sweep-77',
'graceful-sweep-76',
'wild-sweep-72',
'apricot-sweep-71',
'decent-sweep-69',
'jolly-sweep-68',
'morning-sweep-65',
'stellar-sweep-63',
'fearless-sweep-62',
'twilight-sweep-61',
'cosmic-sweep-60',
'eager-sweep-56',
'brisk-sweep-53',
'jolly-sweep-48',
'decent-sweep-46',
'classic-sweep-45',
'dry-sweep-41',
'zesty-sweep-39',
'serene-sweep-37',
'warm-sweep-36',
'dulcet-sweep-34',
'dauntless-sweep-33',
'chocolate-sweep-27',
'brisk-sweep-26',
'scarlet-sweep-25',
'celestial-sweep-24',
'trim-sweep-22',
'radiant-sweep-21',
'rose-sweep-19',
'zany-sweep-17',
'light-sweep-14',
'genial-sweep-13',
'dry-sweep-12',
'fiery-sweep-11',
'polished-sweep-7',
'polished-sweep-6',
'helpful-sweep-4',
'deep-sweep-1']
qparego_success_imc = ['clear-sweep-95',
'effortless-sweep-88',
'comic-sweep-87',
'lucky-sweep-84',
'serene-sweep-81',
'graceful-sweep-76',
'good-sweep-72',
'leafy-sweep-71',
'vocal-sweep-70',
'flowing-sweep-61',
'dauntless-sweep-55',
'dutiful-sweep-50',
'confused-sweep-39',
'electric-sweep-37',
'charmed-sweep-26',
'denim-sweep-17',
'comic-sweep-11',
'zesty-sweep-8']
tkwargs = {
"dtype": torch.float32,
"device": torch.device("cuda" if torch.cuda.is_available() else "cpu"),
}
def make_hv_df(array, method):
hvs_df = pd.DataFrame(array.cpu())
hvs_df['run ID'] = range(array.shape[0])
hvs_df['method'] = method
hvs_df = pd.melt(hvs_df, id_vars=['run ID', 'method'])
hvs_df.columns = ['run ID', 'method', 'Acquisition step', 'Hypervolume']
return hvs_df
def compute_hv(acquisitions, num_init_points, ref_point):
hv_array = torch.zeros((acquisitions.shape[0], acquisitions.shape[1] - num_init_points), **tkwargs)
for run in range(acquisitions.shape[0]):
for i in range(acquisitions.shape[1] - num_init_points):
# compute hypervolume
if type(acquisitions) is torch.Tensor:
bd = DominatedPartitioning(ref_point=ref_point,
Y=acquisitions[run,:num_init_points+i+1,:])
else:
bd = DominatedPartitioning(ref_point=ref_point,
Y=torch.tensor(acquisitions, dtype=torch.float32)[run,:num_init_points+i+1,:])
bd.to(device=tkwargs["device"])
volume = bd.compute_hypervolume().item()
hv_array[run, i] = volume
if (run % 10) == 0:
print(f"Run {run}")
return hv_array
def export_acquisitions(path, saved_files, total_iter, num_tasks, method, experiment):
if experiment == 'citeseq':
qnehvi_success = qnehvi_success_citeseq
qparego_success = qparego_success_citeseq
elif experiment == 'imc':
qnehvi_success = qnehvi_success_imc
qparego_success = qparego_success_imc
else:
print("Invalid experiment")
acquisitions = torch.zeros((len(saved_files), total_iter, num_tasks), **tkwargs)
unsaved_runs = []
if method == "botorch":
acquisitions = torch.zeros((len(qnehvi_success), total_iter, num_tasks), **tkwargs)
for i,run in enumerate(qnehvi_success):
if run+'_output_dict.pt' in saved_files:
tmp = torch.load(path+"/"+run+'_output_dict.pt')
acquisitions[i,:,:] = tmp['train_y_original/'+method]
else:
print(f"{run} is not saved")
assert False
elif method == "qparego":
acquisitions = torch.zeros((len(qparego_success), total_iter, num_tasks), **tkwargs)
for i,run in enumerate(qparego_success):
if run+'_output_dict.pt' in saved_files:
tmp = torch.load(path+'/'+run+'_output_dict.pt')
acquisitions[i,:,:] = tmp['train_y_original/'+method]
else:
print(f"{run} is not saved")
unsaved_runs.append(run)
assert len(unsaved_runs) == 1, "Warning! Deleting runs will be wrong"
unsaved_ind = [i for i in range(len(qparego_success)) if qparego_success[i] == unsaved_runs[0]][0]
acquisitions = torch.cat((acquisitions[:unsaved_ind,:,:], acquisitions[unsaved_ind+1:,:,:]), axis=0)
else:
for i,run in enumerate(saved_files):
tmp = torch.load(path+'/'+run)
acquisitions[i,:,:] = tmp['train_y_original/'+method]
return acquisitions
def scale_data(data, y_min, y_max):
data = (data - y_min) / (y_max - y_min)
return data
def scale_acquisitions(array, mins, maxes):
if type(array) is torch.Tensor:
scaled_array = torch.zeros_like(array, **tkwargs)
else:
scaled_array = torch.zeros_like(torch.tensor(array), **tkwargs)
for obj in range(array.shape[2]):
if type(array) is torch.Tensor:
scaled_array[:,:,obj] = scale_data(array[:,:,obj], mins[obj], maxes[obj])
else:
scaled_array[:,:,obj] = torch.tensor(scale_data(array[:,:,obj], mins[obj], maxes[obj]))
return scaled_array
def define_settings(experiment):
citeseq_paths = ["manatee-run-dicts/citeseq-random-state-sweeps/p5q4rjku",
"manatee-run-dicts/citeseq-random-state-sweeps/gkqg9hiv_x9a5qwif",
"manatee-run-dicts/citeseq-random-state-sweeps/55nq05yl",
"manatee-run-dicts/citeseq-random-state-sweeps/v2qh0wsn_87fm2qv4",
"manatee-new/pipeline/citeseq_usemo-w9iqo4ar",
"manatee-run-dicts/citeseq-random-state-sweeps/73eys3zm",
"manatee-run-dicts/citeseq-random-state-sweeps/4t7umvjt"]
imc_paths = ["manatee-new/pipeline/final-imc-sweep/msa-ra-rs",
"manatee-new/pipeline/final-imc-sweep/msa-ra-rs",
"manatee-new/pipeline/final-imc-sweep/msa-ra-rs",
"manatee-new/pipeline/final-imc-sweep/mas",
"manatee-new/pipeline/imc_usemo-dew50r9d",
"manatee-run-dicts/final-imc-sweep/qnehvi",
"manatee-run-dicts/final-imc-sweep/qparego"]
citeseq_num_tasks = 9
citeseq_total_iter = 5+36
imc_num_tasks = 7
imc_total_iter = 5+35
db_lower_bound = -18.588226318359375
cal_upper_bound = 304.9784240722656
citeseq_ref_point = torch.tensor([-1., # sil
0., # cal
db_lower_bound, #db
-1.,
-1.,
-1.,
-1.,
-1.,
-1.], **tkwargs)
imc_ref_point = torch.tensor([-1.], **tkwargs).repeat(imc_num_tasks)
citeseq_mins = [-1., 0., db_lower_bound, -1., -1., -1., -1., -1., -1.]
citeseq_maxes = [1., cal_upper_bound, 0., 1., 1., 1., 1., 1., 1.]
imc_mins = [-1., -1., -1., -1., -1., -1., -1.]
imc_maxes = [1., 1., 1., 1., 1., 1., 1.]
if experiment == "citeseq":
paths = citeseq_paths
total_iter = citeseq_total_iter
num_tasks = citeseq_num_tasks
ref_point = citeseq_ref_point
mins = citeseq_mins
maxes = citeseq_maxes
elif experiment == 'imc':
paths = imc_paths
total_iter = imc_total_iter
num_tasks = imc_num_tasks
ref_point = imc_ref_point
mins = imc_mins
maxes = imc_maxes
else:
assert False
settings = {}
settings["paths"] = paths
settings["total_iter"] = total_iter
settings["num_tasks"] = num_tasks
settings["ref_point"] = ref_point
settings["mins"] = mins
settings["maxes"] = maxes
return settings
if __name__ == "__main__":
parser = argparse.ArgumentParser(description='Compute HV for experiment of choice.')
parser.add_argument('--experiment', type=str,
help='Type of experiment. Options: imc, citeseq')
args = parser.parse_args()
print(f"Using {tkwargs['device']}")
global_path = "/home/campbell/aselega/Projects/"
methods = ["manatee", "random prob", "random loc", "manatee", "usemo", "botorch", "qparego"]
method_labels = ["M-SA", "RS", "RA", "M-AS", "USeMO", "qNEHVI", "qNParEGO"]
num_init_points = 5
settings = define_settings(args.experiment)
for i, method in enumerate(methods):
mypath = global_path + settings["paths"][i]
print(f"Currently in directory {settings['paths'][i]}")
files = [f for f in listdir(mypath) if isfile(join(mypath, f)) and f.endswith(".pt")]
# Export acquisitions from .pt file
acq_array = export_acquisitions(mypath, files, settings["total_iter"], settings["num_tasks"], method, args.experiment)
# Scale acquisitions to [0,1]
scaled_array = scale_acquisitions(acq_array, settings["mins"], settings["maxes"])
# Compute HV
hvs_array = compute_hv(scaled_array, num_init_points, settings["ref_point"])
print(f"Processed {method_labels[i]}.")
hvs_df = make_hv_df(hvs_array, method_labels[i])
dir_path = pathlib.Path(f"{global_path}/manatee-run-dicts/hypervolumes/scaled-{args.experiment}/")
dir_path.mkdir(parents=True, exist_ok=True)
filename = method_labels[i] + "_hvs.csv"
filepath = dir_path / f"{filename}"
hvs_df.to_csv(filepath)
print(f"Written {filename}.")