forked from Notlaughinggirl/modeling-attflow-on-graphs
-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
183 lines (154 loc) · 8.25 KB
/
train.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
from __future__ import absolute_import
from __future__ import print_function
from __future__ import division
import numpy as np
import tensorflow as tf
from utils import write_rows
TABLE = 'modeling_attflow_on_graphs'
FIELDS = ['model', 'dataset', 'gridworld_seed', 'splitting_seed', 'shuffling_seeed', 'model_seed', 'rank',
'h1r_valid', 'h1r_test', 'h1r_epoch', 'h5r_valid', 'h5r_test', 'h5r_epoch', 'h10r_valid', 'h10r_test', 'h10r_epoch',
'mr_r_valid', 'mr_r_test', 'mr_r_epoch', 'mrr_r_valid', 'mrr_r_test', 'mrr_r_epoch',
'h1f_valid', 'h1f_test', 'h1f_epoch', 'h5f_valid', 'h5f_test', 'h5f_epoch', 'h10f_valid', 'h10f_test', 'h10f_epoch',
'mr_f_valid', 'mr_f_test', 'mr_f_epoch', 'mrr_f_valid', 'mrr_f_test', 'mrr_f_epoch']
class Trainer(object):
def __init__(self, model, logger):
self.model = model
self.logger = logger
self.train_tracker = []
self.valid_evaluator = Evaluator(model, source='valid')
self.test_evaluator = Evaluator(model, source='test')
self.logger.info('\n=========================')
self.logger.info('model: %s, dataset: %s, gridword_seed: %d, splitting_seed: %d, shuffling_seed: %d, model_seed: %d'
% (self.model_name, self.dataset_name, self.gridworld_seed, self.splitting_seed, self.shuffling_seed, self.model_seed))
def __getattr__(self, name):
if hasattr(self.model, name):
return getattr(self.model, name)
else:
raise ValueError('`%s` is not defined.' % name)
def __call__(self):
with tf.Session(graph=self.tf_graph, config=self.tf_config) as sess:
if self.checkpoint is not None:
self.saver.restore(sess, self.checkpoint)
else:
sess.run(self.init_op)
n_epochs = 0
n_itrs = 0
while n_epochs < self.max_epochs:
n_epochs += 1
for bs, batch in self.get_train_batch(self.batch_size):
n_itrs += 1
_, loss, accuracy = \
sess.run([self.train_op, self.loss, self.accuracy],
feed_dict={self.inputs: batch, self.learning_rate: self.learning_rates[int((n_epochs-1)/10)]})
self.train_tracker.append((n_epochs, n_itrs, loss, accuracy))
metric_valid = self.valid_evaluator(sess, n_epochs)
metric_test = self.test_evaluator(sess, n_epochs)
self._write_log(n_epochs, metric_valid, metric_test)
self._summarize_and_write_sql()
def _write_log(self, n_epochs, metric_valid, metric_test):
self.logger.info('[EVAL] epoch: %d, h1_r: %.4f (%.4f), h5_r: %.4f (%.4f), h10_r: %.4f (%.4f), mr_r: %.4f (%.4f), mrr_r: %.4f (%.4f), '
'h1_f: %.4f (%.4f), h5_f: %.4f (%.4f), h10_f: %.4f (%.4f), mr_f: %.4f (%.4f), mrr_f: %.4f (%.4f)' %
(n_epochs, metric_valid[1], metric_test[1], metric_valid[2], metric_test[2],
metric_valid[3], metric_test[3], metric_valid[4], metric_test[4],
metric_valid[5], metric_test[5], metric_valid[6], metric_test[6],
metric_valid[7], metric_test[7], metric_valid[8], metric_test[8],
metric_valid[9], metric_test[9], metric_valid[10], metric_test[10],))
def _summarize_and_write_sql(self):
mtr_va = self.valid_evaluator.metrics
mtr_te = self.test_evaluator.metrics
mtr = []
for c in range(1, 11):
epoch_mv_mt = [(mtr_va[r][0], mtr_va[r][c], mtr_te[r][c]) for r in range(len(mtr_va))]
if c == 4 or c == 9:
epoch_mv_mt = sorted(epoch_mv_mt, cmp=lambda (e1, v1, t1), (e2, v2, t2): 0 if (e1, v1) == (e2, v2) else 1 if (v1, e1) > (v2, e2) else -1)
else:
epoch_mv_mt = sorted(epoch_mv_mt, cmp=lambda (e1, v1, t1), (e2, v2, t2): 0 if (e1, v1) == (e2, v2) else 1 if (-v1, e1) > (-v2, e2) else -1)
mtr.append(epoch_mv_mt)
rows = []
for r in range(len(mtr_va)):
rank = r + 1
summ = '[SUMM] %3d | h1r: %.4f (%.5f, %d) | h5r: %.4f (%.5f, %d) | h10r: %.4f (%.5f, %d) | mr_r: %.4f (%.5f, %d) | mrr_r: %.4f (%.5f, %d) | ' \
'h1f: %.4f (%.5f, %d) | h5f: %.4f (%.5f, %d) | h10f: %.4f (%.5f, %d) | mr_f: %.4f (%.5f, %d) | mrr_f: %.4f (%.5f, %d)' % \
(rank, mtr[0][r][1], mtr[0][r][2], mtr[0][r][0], mtr[1][r][1], mtr[1][r][2], mtr[1][r][0], mtr[2][r][1], mtr[2][r][2], mtr[2][r][0],
mtr[3][r][1], mtr[3][r][2], mtr[3][r][0], mtr[4][r][1], mtr[4][r][2], mtr[4][r][0], mtr[5][r][1], mtr[5][r][2], mtr[5][r][0],
mtr[6][r][1], mtr[6][r][2], mtr[6][r][0], mtr[7][r][1], mtr[7][r][2], mtr[7][r][0], mtr[8][r][1], mtr[8][r][2], mtr[8][r][0],
mtr[9][r][1], mtr[9][r][2], mtr[9][r][0])
row = (self.model_name, self.dataset_name, self.gridworld_seed, self.splitting_seed, self.shuffling_seed, self.model_seed, rank,
mtr[0][r][1], mtr[0][r][2], mtr[0][r][0], mtr[1][r][1], mtr[1][r][2], mtr[1][r][0], mtr[2][r][1], mtr[2][r][2], mtr[2][r][0],
mtr[3][r][1], mtr[3][r][2], mtr[3][r][0], mtr[4][r][1], mtr[4][r][2], mtr[4][r][0], mtr[5][r][1], mtr[5][r][2], mtr[5][r][0],
mtr[6][r][1], mtr[6][r][2], mtr[6][r][0], mtr[7][r][1], mtr[7][r][2], mtr[7][r][0], mtr[8][r][1], mtr[8][r][2], mtr[8][r][0],
mtr[9][r][1], mtr[9][r][2], mtr[9][r][0])
self.logger.info(summ)
rows.append(row)
write_rows(rows, TABLE, FIELDS)
class Evaluator(object):
def __init__(self, model, source='test'):
self.model = model
self.source = source
self.metrics = []
def __getattr__(self, name):
if hasattr(self.model, name):
return getattr(self.model, name)
else:
raise ValueError('`%s` is not defined.' % name)
def __call__(self, sess, n_epochs):
observed = []
predicted = []
for bs, batch in self.get_eval_batch(self.batch_size, source=self.source):
prediction = sess.run(self.prediction, feed_dict={self.inputs: batch})
pred_idx = np.argsort(-prediction)
predicted.append(pred_idx)
observed.append(batch)
observed = np.concatenate(observed)
predicted = np.concatenate(predicted)
h1_r, h5_r, h10_r, mr_r, mrr_r, h1_f, h5_f, h10_f, mr_f, mrr_f = \
self._calc_metrics(predicted, observed, self.observation_pool)
metric = (n_epochs, h1_r, h5_r, h10_r, mr_r, mrr_r, h1_f, h5_f, h10_f, mr_f, mrr_f)
self.metrics.append(metric)
return metric
def _calc_metrics(self, predicted, observed, filter_pool):
hit_1_raw, hit_5_raw, hit_10_raw = 0., 0., 0.
hit_1_flt, hit_5_flt, hit_10_flt = 0., 0., 0.
mr_raw, mrr_raw = 0., 0.
mr_flt, mrr_flt = 0., 0.
n_pred = 0
for pred, obsv in zip(predicted, observed):
src, dst = obsv[-2], obsv[-1]
rank_raw = 0
rank_flt = 0
for e in pred:
if e == dst:
break
else:
if (src, e) not in filter_pool:
rank_flt += 1
rank_raw += 1
if rank_raw == 0:
hit_1_raw += 1
if rank_raw < 5:
hit_5_raw += 1
if rank_raw < 10:
hit_10_raw += 1
mr_raw += (rank_raw + 1)
mrr_raw += 1. / (rank_raw + 1)
if rank_flt == 0:
hit_1_flt += 1
if rank_flt < 5:
hit_5_flt += 1
if rank_flt < 10:
hit_10_flt += 1
mr_flt += (rank_flt + 1)
mrr_flt += 1. / (rank_flt + 1)
n_pred += 1
hit_1_raw /= n_pred
hit_5_raw /= n_pred
hit_10_raw /= n_pred
mr_raw /= n_pred
mrr_raw /= n_pred
hit_1_flt /= n_pred
hit_5_flt /= n_pred
hit_10_flt /= n_pred
mr_flt /= n_pred
mrr_flt /= n_pred
return hit_1_raw, hit_5_raw, hit_10_raw, mr_raw, mrr_raw, \
hit_1_flt, hit_5_flt, hit_10_flt, mr_flt, mrr_flt