-
Notifications
You must be signed in to change notification settings - Fork 1
/
multi_util.py
146 lines (129 loc) · 7.33 KB
/
multi_util.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
import tensorflow as tf
import numpy as np
from sklearn.metrics import accuracy_score, roc_auc_score, confusion_matrix, precision_recall_curve, auc
def get_record_parser(max_len, dim):
def parse(example):
features = tf.parse_single_example(example,
features={
'patient_id': tf.FixedLenFeature([], tf.int64),
'index': tf.FixedLenFeature([], tf.string),
'medicine': tf.FixedLenFeature([], tf.string),
'seq_len': tf.FixedLenFeature([], tf.int64),
'label': tf.FixedLenFeature([], tf.int64)
})
index = tf.reshape(tf.decode_raw(features['index'], tf.float32), [max_len, dim[0]])
medicine = tf.reshape(tf.decode_raw(features['medicine'], tf.float32), [max_len, dim[1]])
label = tf.to_int32(features['label'])
seq_len = tf.to_int32(features['seq_len'])
patient_id = features['patient_id']
return patient_id, index, medicine, seq_len, label
return parse
def get_batch_dataset(record_file, parser, config):
num_threads = tf.constant(config.num_threads, dtype=tf.int32)
dataset = tf.data.TFRecordDataset(record_file).map(parser, num_parallel_calls=num_threads).shuffle(
config.capacity).batch(config.train_batch).repeat()
return dataset
def get_dataset(record_file, parser, config):
num_threads = tf.constant(config.num_threads, dtype=tf.int32)
dataset = tf.data.TFRecordDataset(record_file).map(
parser, num_parallel_calls=num_threads).batch(config.dev_batch).repeat(config.epochs)
return dataset
def evaluate_batch(model, num_batches, eval_file, sess, data_type, handle, str_handle, is_point, logger):
losses = []
pre_scores, pre_labels, ref_labels = [], [], []
metrics = {}
pre_points = {3: [], 18: [], 36: [], 72: [], 144: [], 216: []}
ref_points = {3: [], 18: [], 36: [], 72: [], 144: [], 216: []}
for _ in range(num_batches):
patient_ids, loss, labels, scores, seq_lens = sess.run([model.id, model.loss, model.pre_labels,
model.pre_scores, model.seq_len],
feed_dict={handle: str_handle} if handle is not None else None)
losses.append(loss)
for pid, pre_label, pre_score, seq_len in zip(patient_ids, labels, scores, seq_lens):
sample = eval_file[str(pid)]
if is_point:
pre_labels.append(pre_label)
ref_labels.append(sample['label'])
else:
ref_labels += [sample['label']] * seq_len
pre_labels += pre_label[:seq_len].tolist()
pre_scores += pre_score[:seq_len].tolist()
for k, v in pre_points.items():
if seq_len >= k:
v.append(pre_label[k - 1])
ref_points[k].append(sample['label'])
metrics['loss'] = np.mean(losses)
metrics['acc'] = accuracy_score(ref_labels, pre_labels)
metrics['roc'] = roc_auc_score(ref_labels, pre_scores)
(precisions, recalls, thresholds) = precision_recall_curve(ref_labels, pre_scores)
metrics['prc'] = auc(recalls, precisions)
metrics['pse'] = np.max([min(x, y) for (x, y) in zip(precisions, recalls)])
for k, v in pre_points.items():
logger.info('{} hour confusion matrix. AUCROC : {}'.format(int(k / 3), roc_auc_score(ref_points[k], v)))
logger.info(confusion_matrix(ref_points[k], v))
logger.info('Full confusion matrix')
logger.info(confusion_matrix(ref_labels, pre_labels))
return metrics
def multi_evaluate(model, num_batches, eval_file, sess, handle, str_handle, is_point):
losses = []
task_metrics = {}
task_labels = {}
task_scores = {}
task_point_pre = {}
task_point_ref = {}
task_point_score = {}
hour_metrics = {}
# weight_matrix = []
tasks = ['4019', '41401', '25000', '5849']
for t in tasks:
task_metrics[t] = {'loss': 0, 'acc': 0.0, 'roc': 0.0, 'prc': 0.0, 'pse': 0.0}
task_labels[t] = {'true': [], 'pred': []}
task_scores[t] = []
task_point_pre[t] = {3 * k: [] for k in range(1, 73)}
task_point_ref[t] = {3 * k: [] for k in range(1, 73)}
task_point_score[t] = {3 * k: [] for k in range(1, 73)}
hour_metrics[t] = []
for _ in range(num_batches):
patient_ids, loss, labels, scores, seq_lens = sess.run([model.id, model.loss, model.pre_labels,
model.pre_scores, model.seq_len],
feed_dict={handle: str_handle} if handle is not None else None)
losses.append(loss)
# weight_matrix.append(weights)
for pid, pre_label, pre_score, seq_len in zip(patient_ids, labels, scores, seq_lens):
sample = eval_file[str(pid)]
task = sample['task']
if is_point:
task_labels[task]['pred'].append(pre_label)
task_labels[task]['true'].append(sample['label'])
else:
task_labels[task]['true'] += [sample['label']] * seq_len
task_labels[task]['pred'] += pre_label[:seq_len].tolist()
task_scores[task] += pre_score[:seq_len].tolist()
for k, v in task_point_pre[task].items():
if seq_len >= k:
task_point_pre[task][k].append(pre_label[k-1])
task_point_ref[task][k].append(sample['label'])
task_point_score[task][k].append(pre_score[k-1])
avg_loss = np.mean(losses)
for t in tasks:
task_metrics[t]['acc'] = accuracy_score(task_labels[t]['true'], task_labels[t]['pred'])
task_metrics[t]['roc'] = roc_auc_score(task_labels[t]['true'], task_scores[t])
(precisions, recalls, thresholds) = precision_recall_curve(task_labels[t]['true'], task_scores[t])
task_metrics[t]['prc'] = auc(recalls, precisions)
task_metrics[t]['pse'] = np.max([min(x, y) for (x, y) in zip(precisions, recalls)])
for k, v in task_point_pre[t].items():
hour_metrics[t].append(cal_metrics(task_point_ref[t][k], task_point_score[t][k], v))
return avg_loss, task_metrics, hour_metrics
# loss_sum = tf.Summary(value=[tf.Summary.Value(tag='{}/loss'.format(data_type), simple_value=metrics['loss']), ])
# acc_sum = tf.Summary(value=[tf.Summary.Value(tag='{}/acc'.format(data_type), simple_value=metrics['acc']), ])
# auc_sum = tf.Summary(value=[tf.Summary.Value(tag='{}/roc'.format(data_type), simple_value=metrics['roc']), ])
# prc_sum = tf.Summary(value=[tf.Summary.Value(tag='{}/prc'.format(data_type), simple_value=metrics['prc']), ])
# return metrics, (loss_sum, acc_sum, auc_sum, prc_sum)
def cal_metrics(ref, pred_scores, pred_labels):
metrics = {}
metrics['acc'] = accuracy_score(ref, pred_labels)
metrics['roc'] = roc_auc_score(ref, pred_scores)
(precisions, recalls, thresholds) = precision_recall_curve(ref, pred_scores)
metrics['prc'] = auc(recalls, precisions)
metrics['pse'] = np.max([min(x, y) for (x, y) in zip(precisions, recalls)])
return metrics