-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmain.py
137 lines (98 loc) · 5.04 KB
/
main.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
import torch
import itertools
from configs.opts import *
from src.train import *
from src.test import *
from models.networks import get_network
import random
import copy
import numpy as np
def safe_print(x):
try:
print(x)
except:
return
# INSTANTIATE TRAINING
NUM_DOMS=1
for i in DOMAINS:
NUM_DOMS*=len(i)
# LOAD NETWORK
net = get_network(CLASSES, NUM_DOMS, residual=RESIDUAL)
net = net.to(DEVICE)
meta_vectors = torch.FloatTensor(NUM_DOMS,NUM_META).fill_(0)
edge_vals=torch.FloatTensor(NUM_DOMS,NUM_DOMS).fill_(0)
edge_vals_no_self=torch.FloatTensor(NUM_DOMS,NUM_DOMS).fill_(0)
full_list=[]
for meta in itertools.product(*DOMAINS):
print(meta)
full_list.append(meta)
meta_vectors[domain_converter(meta)]=get_meta_vector(meta)
for i,vector in enumerate(meta_vectors):
edge_vals[i,:]=compute_edge(vector,meta_vectors,i,1.)
edge_vals_no_self[i,:]=compute_edge(vector,meta_vectors,i,0.)
EXP=NUM_DOMS*(NUM_DOMS-1)
res_source=[]
res_refined=[]
res_upperbound=[]
res_upperbound_ref=[]
res_adagraph=[]
res_adagraph_refinement=[]
upperbound_loader=init_loader(BATCH_SIZE, domains=full_list, auxiliar= True, size=SIZE, std=STD)
for meta_source in itertools.product(*DOMAINS):
source_domain=meta_source
net_std=copy.deepcopy(net).to(DEVICE)
source_loader = init_loader(BATCH_SIZE, domains=[source_domain], shuffle=True, auxiliar=False, size=SIZE, std=STD)
idx_source=domain_converter(source_domain)
net_std.reset_edges()
training_loop(net_std, source_loader, idx_source, epochs=EPOCHS, training_group=SOURCE_GROUP, store=None, auxiliar=False)
net_std.copy_source(idx_source)
net_upperbound=copy.deepcopy(net_std)
net_upperbound.init_edges(edge_vals)
training_loop(net_upperbound,upperbound_loader, idx_source,epochs=1, training_group=TRAINING_GROUP, store=None, auxiliar=True)
for meta_target in itertools.product(*DOMAINS):
target_domain=meta_target
idx_target=domain_converter(meta_target)
if idx_target == idx_source or skip_rule(meta_source,meta_target):
continue
safe_print(str(meta_source) + ' vs ' + str(meta_target))
current_edges=copy.deepcopy(edge_vals)
current_edges[:,idx_target]=0.0
current_edges=current_edges/current_edges.sum(-1).unsqueeze(-1)
current_edges[idx_target,:]=0.0
available_domains=full_list[:]
available_domains.remove(target_domain)
auxiliar_loader=init_loader(BATCH_SIZE, domains=available_domains, auxiliar=True, size=SIZE, std=STD)
net_adagraph=copy.deepcopy(net_std)
net_adagraph.init_edges(current_edges)
training_loop(net_adagraph,auxiliar_loader, idx_source, target_idx=idx_target, epochs=1, training_group=TRAINING_GROUP, store=None, auxiliar=True)
target_loader = init_loader(BATCH_SIZE, domains=[target_domain], shuffle=True, auxiliar=False, size=SIZE, std=STD)
test_loader = init_loader(TEST_BATCH_SIZE, domains=[target_domain], shuffle=False, auxiliar=False, size=SIZE, std=STD)
current_res_source = test(net_std, test_loader, idx_source)
net_adagraph.set_bn_from_edges(idx_target, ew=edge_vals_no_self[idx_target,:])
net_adagraph.init_edges(edge_vals)
net_refined = copy.deepcopy(net_std)
current_res_adagraph = test(net_adagraph, test_loader, idx_target)
current_res_refined = online_test(net_refined,idx_target,target_loader, training_group=TRAINING_GROUP, device=DEVICE, bs=BATCH_SIZE)
current_res_ag_refinement = online_test(net_adagraph,idx_target,target_loader, training_group=TRAINING_GROUP, device=DEVICE, bs=BATCH_SIZE)
current_res_upperbound = test(net_upperbound, test_loader, idx_target)
current_res_upperbound_refined = online_test(net_upperbound,idx_target,target_loader, training_group=TRAINING_GROUP, device=DEVICE, bs=BATCH_SIZE)
res_source.append(current_res_source)
res_refined.append(current_res_refined)
res_adagraph.append(current_res_adagraph)
res_adagraph_refinement.append(current_res_ag_refinement)
res_upperbound.append(current_res_upperbound)
res_upperbound_ref.append(current_res_upperbound_refined)
safe_print('-------------------------res after ' + str(len(res_source))+'--------------------------')
safe_print('BASELINE '+str(np.mean(np.array(res_source))))
safe_print('BASELINE + REFINEMENT '+str(np.mean(np.array(res_refined))))
safe_print('ADAGRAPH '+str(np.mean(np.array(res_adagraph))))
safe_print('ADAGRAPH + REFINEMENT '+str(np.mean(np.array(res_adagraph_refinement))))
safe_print('UPPER BOUND '+str(np.mean(np.array(res_upperbound))))
safe_print('UPPER BOUND + REFINEMENT '+str(np.mean(np.array(res_upperbound_ref))))
safe_print('')
np.save('./results/source'+SUFFIX+'.npy', np.array(res_source))
np.save('./results/refined'+SUFFIX+'.npy', np.array(res_refined))
np.save('./results/adagraph'+SUFFIX+'.npy', np.array(res_adagraph))
np.save('./results/adagraph_refined'+SUFFIX+'.npy', np.array(res_adagraph_refinement))
np.save('./results/upper_bound'+SUFFIX+'.npy', np.array(res_upperbound))
np.save('./results/upper_bound_refined'+SUFFIX+'.npy', np.array(res_upperbound_ref))