ST-SiameseNet/main.py at master · huiminren/ST-SiameseNet · GitHub

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
#!/usr/bin/env python
# coding: utf-8

# In[ ]:


from utils import *
from models import *
import os
import argparse

def create_parser():
    """
    Creates a parser for command-line arguments.
    """
    parser = argparse.ArgumentParser(description='ST-SiameseNet')
    # model hyper-parameters
    parser.add_argument('--with_speed', action='store_true',
                        help='input trajs with speed')
    parser.add_argument('--with_profile', action='store_true',
                        help='input trajs with profile features')
    parser.add_argument('--input_type', type=str, default='all',
                        help='input type, seek, serve or all')
    parser.add_argument('--num_train_plates', type=int, default=500,
                        help='number of training plates (default: 500)')
    parser.add_argument('--num_days', type=int, default=5,
                        help='number of training days (default: 5)')

    # training hyper-parameters
    parser.add_argument('--iteration', type=int, default=1000000,
                        help='number of iterations (default: 1000000)')

    # saving and loading directoreis
    parser.add_argument('--data_path', type=str, default='./dataset/')
    parser.add_argument('--log_path', type=str, default='./log/')
    parser.add_argument('--model_path', type=str, default='./models/')
    parser.add_argument('--log_step', type=int , default=1000)
    parser.add_argument('--checkpoint_every', type=int , default=50000)

    return parser

def main(opts):
    """
    Loads the data, creates checkpoint and sample directories, and starts the training loop.
    """
    create_dir(opts.log_path)
    create_dir(opts.model_path)
    model_path = opts.model_path

    # prepare logging file
    tag = str(opts.num_train_plates)+'plates_'+'days'+str(opts.num_days)+\
    '_inputs_'+str(opts.input_type)+'_speed'+str(opts.with_speed)+'_profile'+str(opts.with_profile)
    print(tag)
    logging.basicConfig(level=logging.DEBUG,
                        format='%(asctime)s %(levelname)s %(message)s',
                        filename=opts.log_path+tag+'.log',
                        filemode='a')

    # load data
    all_plates = load_data(opts.data_path,'plates.pkl')
    all_plates.remove('d1329')
    train_plates = all_plates[:opts.num_train_plates]
    test_plates = all_plates[2000:] # 197 plates which are unseen

    if opts.input_type == 'all':
        input_type = ''
    else:
        input_type = opts.input_type

    if opts.with_speed:
        raw_trajs = load_data(opts.data_path,'trajs_with_speed500.pkl')
    else:
        raw_trajs = load_data(opts.data_path,'trajs_without_speed500.pkl')

    if opts.with_profile:
        profile_data = load_data(opts.data_path,'profile_features500.pkl')
    else:
        profile_data = []

    # prepare model and input data
    if (input_type == '') and (opts.with_speed == False) and (opts.with_profile==True):
        # profile is True and speed is False --> profile only (no xyt)
        siamese_net = build_model_profileonly()
        get_pairs = get_pairs_s_and_d
        raw_trajs = []
    elif input_type == '':
        # both speed and profile are True --> xyt+v+profile
        # both speed and profile are False --> xyt
        # speed is True and profie is False --> xyt+v
        siamese_net = build_model_best(opts.with_speed,opts.with_profile)
        get_pairs = get_pairs_s_and_d
    elif input_type != '':
        # when input_type != '' --> seek or drive
        siamese_net = build_model_seekserve(opts.with_speed,opts.with_profile)
        get_pairs = get_pairs_s_or_d

    # start training
    loss_500 = []
    train_acc_500 = []
    val_acc_500 = []
    test_acc_500 = []

    iteration = opts.iteration
    # save best model
    current_train_acc = 0.7
    current_val_acc = 0.7
    current_test_acc = 0.7
    t0 = datetime.now()
    t1 = datetime.now()

    # prepare evaluation dataset
    pairs,labels = [],[]
    val_pairs,val_labels = [],[]
    test_pairs,test_labels = [],[]
    for _ in range(1000):
        pv,lv = get_pairs(raw_trajs, profile_data, train_plates, input_type, opts.num_days,new_days = True)
        val_pairs.append(pv)
        val_labels.append(lv)
        pt,lt = get_pairs(raw_trajs, profile_data, test_plates, input_type, opts.num_days,new_days = True)
        test_pairs.append(pt)
        test_labels.append(lt)

    for ite in range(iteration):
        pair,label = get_pairs(raw_trajs, profile_data, train_plates, input_type, opts.num_days)
        loss = siamese_net.train_on_batch(pair,label)
        pairs.append(pair)
        labels.append(label)
        # save log
        if ite % opts.log_step == 0 and ite != 0:
            t1 = datetime.now()
            print(ite)
            train_acc,train_prob_list = acc(siamese_net, pairs, labels)
            pairs = []
            labels = []
            val_acc, val_prob_list = acc(siamese_net, val_pairs, val_labels) # old plates new days
            test_acc,test_prob_list = acc(siamese_net, test_pairs, test_labels) # test with new plates and new days.
            loss_500.append(loss)
            train_acc_500.append(train_acc)
            val_acc_500.append(val_acc)
            test_acc_500.append(test_acc)
            logging.info('******iteration: '+str(ite)+'; loss: '+str(loss)+ '; train acc: '+str(train_acc)+'; validation acc: '+ str(val_acc)+'; test acc: '+str(test_acc))
            if train_acc > current_train_acc:
                save_model(siamese_net, model_path, tag = tag+'_best_train')
                current_train_acc = train_acc
                logging.info('best train model updated: ' + str(train_acc))
            if val_acc > current_val_acc:
                save_model(siamese_net, model_path, tag = tag+'_best_val')
                current_val_acc = val_acc
                logging.info('best validation model updated: ' + str(val_acc))
            if test_acc > current_test_acc:
                save_model(siamese_net, model_path, tag = tag+'_best_test')
                current_test_acc = test_acc
                logging.info('best test model updated: ' + str(test_acc))
        if ite % opts.checkpoint_every == 0:
            save_model(siamese_net, model_path, tag = tag + '_mid')
            pickle.dump(loss_500, open(model_path+'loss_{0}.pkl'.format(tag),'wb'))
            pickle.dump(train_acc_500, open(model_path+'train_acc_{0}.pkl'.format(tag),'wb'))
            pickle.dump(val_acc_500, open(model_path+'val_acc_{0}.pkl'.format(tag),'wb'))
            pickle.dump(test_acc_500, open(model_path+'test_acc_{0}.pkl'.format(tag),'wb'))
    logging.info('total running time: '+ str(datetime.now()-t0))
    save_model(siamese_net, model_path, tag = tag + '_iter'+str(iteration))
    pickle.dump(loss_500, open(model_path+'loss_{0}.pkl'.format(tag),'wb'))
    pickle.dump(train_acc_500, open(model_path+'train_acc_{0}.pkl'.format(tag),'wb'))
    pickle.dump(val_acc_500, open(model_path+'val_acc_{0}.pkl'.format(tag),'wb'))
    pickle.dump(test_acc_500, open(model_path+'test_acc_{0}.pkl'.format(tag),'wb'))

if __name__ == '__main__':
    parser = create_parser()
    opts = parser.parse_args()

    print(opts)
    main(opts)