-
Notifications
You must be signed in to change notification settings - Fork 1
/
inference.py
161 lines (139 loc) · 7.02 KB
/
inference.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
import tensorflow as tf
import tensorflow.contrib as tc
import numpy as np
import pandas as pd
import warnings
import time
import os
import logging
from models.rnn_module import cu_rnn
from models.nn_module import dense, seq_loss
from models.attention_module import self_transformer
warnings.filterwarnings(action='ignore', category=UserWarning, module='tensorflow')
os.environ["TF_CPP_MIN_LOG_LEVEL"] = '3'
logger = logging.getLogger('Medical')
logger.setLevel(logging.INFO)
formatter = logging.Formatter('%(asctime)s - %(name)s - %(levelname)s - %(message)s')
console_handler = logging.StreamHandler()
console_handler.setLevel(logging.INFO)
console_handler.setFormatter(formatter)
logger.addHandler(console_handler)
data_dir = 'data/raw_data/inference'
model_dir = 'multi_task/DIMM_final/models'
n_index = 205
n_medicine = 241
batch_size = 8
max_len = 720
class InfModel(object):
def __init__(self):
self.id = tf.placeholder(tf.int32, [None])
self.index = tf.placeholder(tf.float32, [batch_size, max_len, n_index])
self.medicine = tf.placeholder(tf.float32, [batch_size, max_len, n_medicine])
self.seq_len = tf.placeholder(tf.int32, [None])
self.labels = tf.placeholder(tf.int32, [None])
self.n_hidden = 64
self.n_layer = 2
self.n_label = 2
self.N = tf.shape(self.id)[0]
self.max_len = tf.reduce_max(self.seq_len)
self.mask = tf.sequence_mask(self.seq_len, self.max_len, dtype=tf.float32, name='masks')
self.padding = tf.sequence_mask(self.seq_len, self.max_len, dtype=tf.int32, name='padding')
self.index = tf.slice(self.index, [0, 0, 0], tf.stack([self.N, self.max_len, n_index]))
self.medicine = tf.slice(self.medicine, [0, 0, 0], tf.stack([self.N, self.max_len, n_medicine]))
self.initializer = tc.layers.xavier_initializer()
self._build_graph()
def _build_graph(self):
start_t = time.time()
self._encode()
self._rnn()
self._step_attention()
self._seq_label()
self._compute_loss()
logger.info('Time to build graph: {} s'.format(time.time() - start_t))
def _encode(self):
with tf.variable_scope('input_encoding', reuse=tf.AUTO_REUSE):
with tf.variable_scope('index', reuse=tf.AUTO_REUSE):
self.index = dense(self.index, hidden=self.n_hidden, initializer=self.initializer)
self.index = tf.reshape(self.index, [-1, self.max_len, self.n_hidden], name='2_3D')
with tf.variable_scope('medicine', reuse=tf.AUTO_REUSE):
self.medicine = dense(self.medicine, hidden=self.n_hidden, initializer=self.initializer)
self.medicine = tf.reshape(self.medicine, [-1, self.max_len, self.n_hidden], name='2_3D')
self.i2m = self._input_attention(self.index, self.medicine, self.n_hidden, 'i2m_attention')
self.m2i = self._input_attention(self.medicine, self.index, self.n_hidden, 'm2i_attention')
self.index = self._input_attention(self.index, self.index, self.n_hidden, 'i2i_attention')
self.medicine = self._input_attention(self.medicine, self.medicine, self.n_hidden, 'm2m_attention')
self.input_encodes = tf.concat([self.index, self.medicine, self.i2m, self.m2i], 2)
self.input_encodes = tf.nn.dropout(self.input_encodes, 1.0)
def _input_attention(self, input_x, input_y, n_unit, scope):
with tf.variable_scope(scope, reuse=tf.AUTO_REUSE):
input_encodes = self_transformer(input_x, input_y, self.mask, 4, n_unit, 1, 1.0, True, False)
return input_encodes
def _rnn(self):
with tf.variable_scope('rnn', reuse=tf.AUTO_REUSE):
self.seq_encodes, _ = cu_rnn('bi-gru', self.input_encodes, self.n_hidden, batch_size, False, self.n_layer)
self.n_hidden *= self.n_layer
self.seq_encodes = tf.nn.dropout(self.seq_encodes, 1.0)
def _step_attention(self):
with tf.variable_scope('step_attention', reuse=tf.AUTO_REUSE):
self.seq_encodes = self_transformer(self.seq_encodes, self.seq_encodes, self.mask, 4, self.n_hidden,
4, 1.0, True, False)
def _seq_label(self):
with tf.variable_scope('seq_labels', reuse=tf.AUTO_REUSE):
self.seq_encodes = tf.reshape(self.seq_encodes, [-1, self.n_hidden])
self.outputs = dense(self.seq_encodes, hidden=self.n_label, scope='output_labels',
initializer=self.initializer)
self.outputs = tf.reshape(self.outputs, tf.stack([-1, self.max_len, self.n_label]))
self.labels = tf.tile(tf.expand_dims(self.labels, axis=1), tf.stack([1, self.max_len]))
self.label_loss = seq_loss(self.outputs, self.labels, self.mask)
def _compute_loss(self):
self.all_params = tf.trainable_variables()
self.soft_outputs = tf.stop_gradient(tf.nn.softmax(self.outputs))
self.pre_labels = tf.argmax(self.outputs, 2)
self.pre_scores = self.outputs[:, :, 1]
self.loss = self.label_loss
class Inference(object):
def __init__(self):
self.model = InfModel()
sess_config = tf.ConfigProto(allow_soft_placement=True)
sess_config.gpu_options.allow_growth = True
self.sess = tf.Session(config=sess_config)
saver = tf.train.Saver()
saver.restore(self.sess, tf.train.latest_checkpoint(model_dir))
def response(self, file_path):
sess = self.sess
model = self.model
ids, indexes, medicines, seq_lens, labels, names = self.prepro(file_path)
pre_labels, probs = sess.run([model.pre_labels, model.soft_outputs],
feed_dict={model.id: ids, model.index: indexes, model.medicine: medicines,
model.seq_len: seq_lens, model.labels: labels})
return pre_labels, probs
def prepro(self, data_path):
b_ids = np.zeros(batch_size, dtype=np.int64)
b_index = np.zeros((batch_size, n_index), dtype=np.float32)
b_medicine = np.zeros((batch_size, n_medicine), dtype=np.float32)
b_seq_len = np.zeros(batch_size, dtype=np.int64)
b_labels = np.zeros(batch_size, dtype=np.int64)
b_names = []
for i, file in enumerate(os.listdir(data_path)):
if file.startswith('0'):
dead = 0
else:
dead = 1
raw_sample = pd.read_csv(os.path.join(data_path, file), sep=',')
raw_sample = raw_sample.fillna(0)
medicine = raw_sample.iloc[:, 209:].as_matrix()
index = raw_sample.iloc[:, 3:208].as_matrix()
b_ids[i] = i
b_index[i] = index
b_medicine[i] = medicine
b_seq_len[i] = index.shape[0]
b_labels[i] = dead
b_names.append(file)
return b_ids, b_index, b_medicine, b_seq_len, b_labels, b_names
if __name__ == "__main__":
infer = Inference()
labels, probs = infer.response(data_dir)
print('Labels')
print(labels)
print('Probs')
print(probs)