-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathearly_stop.py
229 lines (205 loc) · 8.48 KB
/
early_stop.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
from torch.utils.data import DataLoader
from utils import load_data
from layers import Agent
from tqdm import tqdm
import numpy as np
from data_utils import *
from conf import *
# load dataset
train_samples, test_samples = load_data(train_path), load_data(test_path)
test_size = len(test_samples)
# construct symptom & disease vocabulary
sv = SymptomVocab(samples=train_samples, add_special_sxs=True, min_sx_freq=min_sx_freq, max_voc_size=max_voc_size)
dv = DiseaseVocab(samples=train_samples)
num_sxs, num_dis = sv.num_sxs, dv.num_dis
test_ds = SymptomDataset(test_samples, sv, dv, keep_unk, add_tgt_start=True, add_tgt_end=True)
test_ds_loader = DataLoader(test_ds, batch_size=1, num_workers=num_workers, shuffle=False, collate_fn=pg_collater)
# compute disease-symptom co-occurrence matrix
dscom = compute_dscom(train_samples, sv, dv)
# global median_list
sscom, mid_list, avg_list = compute_sscom(train_samples, sv, normalize=False)
# 将median_list中nan值替换为无穷大
for i in range(len(mid_list)):
if np.isnan(mid_list[i]):
mid_list[i] = float('inf')
if np.isnan(avg_list[i]):
avg_list[i] = float('inf')
# init reward distributor
rd = RewardDistributor(sv, dv, dscom)
# init patient simulator
ps = PatientSimulator(sv)
# init agent
model = Agent(num_sxs, num_dis).to(device)
max_turn = num_turns
# dense_one
model_name = 'metric_model_{}.pt'.format('40_1')
# model_name = 'acc_model_{}.pt'.format('27_4')
metric_model_path = 'saved/{}/dense_all/{}'.format(train_dataset, model_name)
# metric_model_path = 'saved/all/best_pt_model.pt'
model.load(metric_model_path)
print('load model from {}......'.format(metric_model_path))
flag = f"===========================pt_model:{pt_path},data_set:{train_dataset},model_name:{model_name},is_position:{dec_add_pos},is_shuffle:{is_shuffle},is_config:{is_config}==========================="
# 当症状的不确定性很大，确定性很小
# 疾病的确定性很大，确定性很大
# eps = [0.5, 0.6, 0.7, 0.8, 0.850,0.860,0.870,0.88,0.89,0.90,0.950, 0.955, 0.960, 0.965, 0.970, 0.975, 0.980, 0.985, 0.990, 0.995, 1.0]
eps = [0.1, 0.2, 0.3, 0.4, 0.5, 0.51, 0.52, 0.53, 0.54, 0.55, 0.56, 0.57, 0.6, 0.62, 0.65, 0.66, 0.7, 0.8, 0.82, 0.83,
0.84, 0.850, 0.860, 0.870, 0.88, 0.89,
0.90, 0.91, 0.915, 0.92,
0.925, 0.93, 0.935, 0.94, 0.950, 0.955, 0.960, 0.965, 0.970, 0.975, 0.980, 0.985, 0.990, 0.995, 1.0]
# eps = [0.85,0.86,0.87,0.875,0.88,0.885,0.89]
# eps = [0.970, 0.975, 0.980, 0.985, 0.990, 0.995, 1.0]
# eps = [0.5, 0.6, 0.65, 0.7, 0.73, 0.74, 0.75, 0.76, 0.8, 0.85, 0.9, 0.95, 1.0]
# recall = [0.5, 0.6, 0.65, 0.7, 0.73, 0.74, 0.75, 0.76, 0.8, 0.85, 0.9, 0.95, 1.0]
recall = [0.1, 0.2, 0.3, 0.4, 0.5, 0.51, 0.52, 0.53, 0.54, 0.55, 0.56, 0.57, 0.6, 0.62, 0.65, 0.66, 0.7, 0.8, 0.82,
0.83, 0.84, 0.850, 0.860, 0.870, 0.88, 0.89,
0.90, 0.91,
0.915, 0.92,
0.925, 0.93, 0.935, 0.94, 0.950, 0.955, 0.960, 0.965, 0.970, 0.975, 0.980, 0.985, 0.990, 0.995, 1.0]
miss_num = [12, 12, 11, 11, 10, 10, 9, 9, 8, 8, 7, 7, 6, 6, 5, 5, 3, 3, 2, 2, 5, 5, 5, 5]
# imp_num = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
# e = 0
# imp_num = [11,14,17]
turns = []
accs = []
rc = 0
with open('early_log.txt', 'a') as f:
f.write(str(flag) + '\n')
idx = 0
def early_s(e, rc, r, miss_num=6, si_num=12):
test_num_hits = 0
actual_turns = []
max_probs = []
exp_end_nums = 0
imp_rcs = 0
model.eval()
with torch.no_grad():
for batch in tqdm(test_ds_loader):
# if rand_num:
# random_num = random.randint(3, 12)
# else:
# random_num = 10
# print(random_num)
actual_turn, is_success, max_prob, exp_end, imp_recall = model.execute(batch, ps, sv, max_turn, e,
rc, r, miss_num, si_num)
test_num_hits += is_success
actual_turns.append(actual_turn)
max_probs.append(np.round(max_prob, digits))
exp_end_nums += exp_end
imp_rcs += imp_recall
test_acc = test_num_hits / test_size
avg_turn = np.mean(actual_turns)
turns.append(avg_turn)
accs.append(test_acc)
if imp_rcs != 0:
imp_recalls = imp_rcs / (test_size - exp_end_nums)
imp_rc_real = imp_rcs / test_size
else:
imp_recalls = 0
imp_rc_real = 0
result = 'eps: {}, avg turn: {}, acc: {}, recall: {},recall_r:{},exp_end_nums:{}, recall_threshold:{}.'.format(
e, np.round(np.mean(actual_turns), digits), np.round(test_acc, digits), np.round(imp_recalls, digits),
np.round(imp_rc_real, digits),
exp_end_nums, rc)
# 将result保存到log.txt中
with open('early_log.txt', 'a') as f:
f.write(str(result) + '\n')
print(result)
return test_acc, imp_rc_real
# # 随机值测试模拟的recall
#
# for e, rc in tqdm(zip(eps, recall)):
#
# early_s(e=e, rc=rc, r=2,rand_num=True)
# 固定值测试模拟的recall
# for i in range(15, 10,-1):
# for i in range(10, 5, -2):
# for e, rc in tqdm(zip(eps, recall)):
# early_s(e=0, rc=rc, r=2, si_num=15, miss_num=6)
# for e, rc in tqdm(zip(eps, recall)):
# early_s(e=e, rc=rc, r=2)
# 对于使用miss_num、si_num的转换准则，需要单独测试
def si_test():
# 基于rc+si_num+miss_num
for e, rc in tqdm(zip(eps, recall)):
early_s(e=0, rc=rc, r=2, si_num=15, miss_num=10)
# 基于e+miss_num
for e in tqdm(eps):
early_s(e=e, rc=0, r=1, miss_num=8)
def ruler_run():
# r1、r2、r3、r4
for r in range(4):
repeat = 0
last_acc = 0
last_rc = 0
with open('early_log.txt', 'a') as f:
f.write(f'----------------------{r + 1}-------------------------' + '\n')
if r == 0:
# r1 完全基于症状回归率的转换准则
for rc in tqdm(recall):
acc, rc = early_s(e=0, rc=rc, r=r)
if acc == last_acc and rc == last_rc:
repeat+=1
last_acc = acc
last_rc = rc
if repeat>=4:
break
if r == 1:
# r2 基于模拟症状回归率的转换准则
for rc in tqdm(recall):
acc, rc = early_s(e=0, rc=rc, r=r, miss_num=6, si_num=18) # dxybest 6/15
if acc == last_acc and rc == last_rc:
repeat+=1
last_acc = acc
last_rc = rc
if repeat>=4:
break
elif r == 2:
# r3 改进的DxFormer中的转换准则
for e in tqdm(eps):
early_s(e=e, rc=0, r=r, miss_num=8) # dxybest 6
elif r == 3:
# r4 DxFormer中的转换准则
for e in tqdm(eps):
early_s(e=e, rc=0, r=r)
# si_test()
ruler_run()
# miss_num、si_num
# dxy 6/15 mz4 mz10
# for e in tqdm(eps + recall):
#
# if idx > 13:
# rc = e
# e = 0
# for recall_threshold in recall:
# eps: 1.0, avg turn: 10.0, acc: 0.7536.
# eps: 0.99, avg turn: 8.0696, acc: 0.742.
# eps: 1.0, avg turn: 20.0, acc: 0.6942.
# eps: 0.99, avg turn: 14.966, acc: 0.6869.
# eps: 0.95, avg turn: 10.3131, acc: 0.6408.
# r2 基于模拟症状回归率的转换准则
# 将eps和recall组合，同时传入
# for rc in tqdm(recall):
# acc, rc = early_s(e=0, rc=rc, r=1, miss_num=6, si_num=12) # dxybest 6/15
# if acc == last_acc and rc == last_rc:
# break
# last_acc = acc
# last_rc = rc
# dec/enc eps: 1.0/0.0, avg turn: 20.0, acc: 0.6796, avg dec/enc entropy: 0.7219/0.173
# dec/enc eps: 1.0/0.005, avg turn: 16.6748, acc: 0.6772, avg dec/enc entropy: 0.6711/0.173
# dec/enc eps: 1.0/0.01, avg turn: 14.1092, acc: 0.6699, avg dec/enc entropy: 0.6711/0.173
# dec/enc eps: 1.0/0.02, avg turn: 10.6942, acc: 0.6456, avg dec/enc entropy: 0.6711/0.173
# dec/enc eps: 1.0/0.03, avg turn: 8.0825, acc: 0.6165, avg dec/enc entropy: 0.6294/0.173
# dec/enc eps: 1.0/0.04, avg turn: 6.9442, acc: 0.6044, avg dec/enc entropy: 0.6294/0.173
# dec/enc eps: 1.0/0.05, avg turn: 5.8374, acc: 0.585, avg dec/enc entropy: 0.6294/0.1667
# dec/enc eps: 1.0/0.08, avg turn: 2.7718, acc: 0.5534, avg dec/enc entropy: 0.5923/0.1667
# dec/enc eps: 1.0/0.1, avg turn: 1.4417, acc: 0.5291, avg dec/enc entropy: 0.5923/0.1667
# entropys = []
# model.eval()
# with torch.no_grad():
# for batch in tqdm(test_ds_loader):
# entropys.append(model.execute(batch, ps, sv, max_turn))
#
# ess = []
# for i in range(max_turn):
# es = sum([entropy[i] for entropy in entropys])
# ess.append(es)