-
Notifications
You must be signed in to change notification settings - Fork 0
/
explore.py
179 lines (157 loc) · 7.46 KB
/
explore.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
# explore optim hyparameters
from time import strftime, localtime
import random, sys
from data_utils import *
from train import set_logger, Instructor
from config import DEFAULT_OPTION
def parameter_explore(opt, par_vals, datasets=None, semi_sup_compare=False):
# par_vals:{parameter_name:[parameter_value,...],...}
def is_valid_option(opt):
is_valid_opt = True
logger.info('comparing……'.center(30, '='))
# valid_ratios = [0, 0.25, 0.5, 0.75]
ratio_name = 'train_len'
ratios = [None, 1500, 1000, 500]
# ratios = [None, 1500]
# only these factors influence supervised
valid_opt_name = '{lr}_{l2}_{window_weight}_{drop_lab}'.format(lr=opt.lr, l2=opt.l2,
window_weight=opt.window_weight,
drop_lab=opt.drop_lab)
# {valid_opt_name:{ratio:acc,...}}
if not semi_valid_ratio_scores.get(valid_opt_name): # init
semi_valid_ratio_scores[valid_opt_name] = {x: None for x in ratios}
semi_scores = semi_valid_ratio_scores.get(valid_opt_name) # {ratio:acc}
for ratio in ratios:
# the difference between sup/sems is only the toggle of semi_supervised
# keep other settings same:drop_lab/drop_unlab/mask_ratio
# sup
_opt = opt.set({ratio_name: ratio}, name='compare_' + opt.name)
if not semi_scores.get(ratio):
_ins_sup = Instructor(_opt.set({'semi_supervised': False}), logout=False)
res_sup = _ins_sup.run()
semi_scores[ratio] = res_sup
else:
res_sup = semi_scores.get(ratio) # fetch sup results directly
# semi
_ins_semi = Instructor(_opt.set({'semi_supervised': True}), logout=False)
res_semi = _ins_semi.run()
logger.info(
'ratio: {} semi[ acc:{} f1:{} ] super_acc[ acc:{} f1:{} ]'.format(ratio, res_semi['acc'],
res_semi['f1'],
res_sup['acc'], res_sup['f1']))
is_valid_opt = res_semi['acc'] > res_sup['acc']
if not is_valid_opt:
logger.info('option:{} is dropped!'.format(opt.name).center(30, '='))
break
else:
logger.info('option:{} is valid!'.format(opt.name).center(30, '='))
return is_valid_opt
datasets = ['laptop'] if datasets is None else datasets
# logger
logger_fname = 'p'
for p in par_vals.keys():
logger_fname += '_{}{}'.format(p, len(par_vals[p])) # p_{parameter_name}{search_len}
logger = set_logger(name='parameter_explore',
file='{}_{}.log'.format(logger_fname, strftime("%m%d-%H%M%S", localtime())))
# dynamic build search_options
search_options = []
if len(par_vals) == 0:
search_options.append(opt)
elif len(par_vals) == 1:
# {p:[v1,]}
for p, values in par_vals.items():
for v in values:
tmp_opt = opt.set({
p: v,
}, 'p_' + '{name}[{value}]'.format(name=p, value=v))
search_options.append(tmp_opt)
else:
# random search
max_len = 50
hy_params_his = set() # history
for i in range(max_len):
opt_name = 'p_'
hy_params = {}
for p, vals in par_vals.items():
v = random.sample(vals, 1)[0]
hy_params[p] = v
opt_name += '{name}[{value}]_'.format(name=p, value=v)
if opt_name not in hy_params_his: # check if has created before
hy_params_his.add(opt_name)
search_options.append(opt.set(hy_params, name=opt_name))
# filter invalid options,when drop_unlab <= drop_lab
search_options = filter(lambda option: option.drop_lab <= option.drop_unlab, search_options)
if semi_sup_compare:
semi_valid_ratio_scores = {} # saved different scores with same options
logger.warning('have opened the semi_sup_compare!!!')
search_results = {d: [] for d in datasets}
# search
best_result = 0
best_params = None
for dataset in datasets:
results = []
for search_option in search_options:
if semi_sup_compare and not is_valid_option(search_option.set({'dataset': dataset})):
continue
_ins = Instructor(search_option.set({'dataset': dataset}))
res = _ins.run()
results.append(res)
vr = '[dataset]:{dataset} [{option}] [result]:{result}'.format(dataset=dataset,
option=search_option.name,
result=res)
search_results[dataset].append(vr)
logger.info(vr)
acc = res['acc']
if acc > best_result:
best_result = acc
best_params = vr
# show results
logger.info('final results'.center(30, '*'))
logger.info('sys:{}'.format(sys.platform, ))
for d, res in search_results.items():
# d:dataset res:List[Str]
res = sorted(res) if len(par_vals) > 1 else res # random search ,then sort the results
for r in res: logger.info(r)
logger.info('*' * 30)
logger.info('best params：{}'.format(best_params))
if __name__ == '__main__':
opt = DEFAULT_OPTION
ps = {
# 'batch_size': [32, 64],
# 'seed': [random.randint(100,10000) for _ in range(3)], # repeat just ,brings no influence
# 'lr': [1e-2, 1e-3, 1e-4],
# 'l2': [5e-3, 1e-3, 5e-4,1e-4,5e-5,1e-5],
# 'encoder_hidden_size':[300,512,768,1024]
# 'window_weight': range(0,12,2),
# 'drop_lab': [x / 10 for x in rasnge(0, 5)],
# 'drop_unlab': [x / 10 for x in range(3, 9)],
# 'unlabel_len': [5000, 10000,15000,20000],
# 'train_len': [500], # 500, 1000, 2000
# 'semi_supervised': [True, False],
# 'use_weight': [False, True],
}
# datasets = opt.datasets.keys()
# datasets = ['laptop']
datasets = ['restaurant']
# seed
# seed = random.randint(100, 1234)
# print('random seed:', seed)
# opt = opt.set({'seed': None})
# clear model
# opt = opt.set({'clear_model': False})
if datasets == ['laptop']:
sup_opt = opt.set({'drop_lab': 0, 'window_weight': 8})
semi_opt = opt.set({"semi_supervised": True, 'drop_lab': 0, 'drop_unlab': 0.7,
'window_weight': 8})
elif datasets == ['restaurant']:
sup_opt = opt.set({'drop_lab': 0, 'window_weight': 0})
semi_opt = opt.set({"semi_supervised": True, 'drop_lab': 0, 'drop_unlab': 0.6,
'window_weight': 0})
parameter_explore(sup_opt, ps, datasets=datasets)
# parameter_explore(semi_opt, ps, datasets=datasets)
# parameter_explore(semi_opt, ps,
# semi_sup_compare=True,
# datasets=datasets) # semi default laptop restaurant
# parameter_explore(opt.set({"semi_supervised": True}), ps) # semi default lap
# parameter_explore(opt.set({"semi_supervised": True,}), ps,datasets=['restaurant']) # semi default res
# parameter_explore(opt.set({"semi_supervised": True}), ps,datasets=datasets) # semi all#