def metric_fn(bmeo_label_ids, bmeo_pred_ids, attr_label_ids,
attr_pred_ids):
bmeo_indices = []
bmeo_calulate_label = []
bmeo_id2label = {}
attr_indices = []
attr_calulate_label = []
attr_id2label = {}
for k, v in bmeo_label2id.items():
bmeo_id2label[v] = k
if k == 'O' or k == '[CLS]' or k == '[SEP]':
pass
else:
bmeo_calulate_label.append(k)
bmeo_indices.append(v)
print('bmeo_calulate_label: {}'.format(bmeo_calulate_label))
print('bmeo_indices: {}'.format(bmeo_indices))
weight = tf.sequence_mask(FLAGS.max_seq_length)
bmeo_precision = tf_metrics.precision(bmeo_label_ids,
bmeo_pred_ids,
num_bmeo_labels,
bmeo_indices, weight)
bmeo_recall = tf_metrics.recall(bmeo_label_ids, bmeo_pred_ids,
num_bmeo_labels, bmeo_indices,
weight)
bmeo_f = tf_metrics.f1(bmeo_label_ids, bmeo_pred_ids,
num_bmeo_labels, bmeo_indices, weight)
for k, v in attr_label2id.items():
attr_id2label[v] = k
if k == 'O' or k == '[CLS]' or k == '[SEP]':
pass
else:
attr_calulate_label.append(k)
attr_indices.append(v)
print('attr_calulate_label: {}'.format(attr_calulate_label))
print('attr_indices: {}'.format(attr_indices))
weight = tf.sequence_mask(FLAGS.max_seq_length)
attr_precision = tf_metrics.precision(attr_label_ids,
attr_pred_ids,
num_attr_labels,
attr_indices, weight)
attr_recall = tf_metrics.recall(attr_label_ids, attr_pred_ids,
num_attr_labels, attr_indices,
weight)
attr_f = tf_metrics.f1(attr_label_ids, attr_pred_ids,
num_attr_labels, attr_indices, weight)
return {
"bmeo_eval_precision": bmeo_precision,
"bmeo_eval_recall": bmeo_recall,
"bmeo_eval_f": bmeo_f,
"attr_eval_precision": attr_precision,
"attr_eval_recall": attr_recall,
"attr_eval_f": attr_f,
# "eval_loss": loss,
}
def metric_fn(label_ids, pred_ids):
indices = []
calulate_label = []
# 加载 id2tag 字典
id2tag = {}
with open(os.path.join(FLAGS.output_dir, 'label2id.pkl'),
'rb') as f:
label2id = pickle.load(f)
print('load label2id success!')
for k, v in label2id.items():
id2tag[v] = k
if k == 'O' or k == '[CLS]' or k == '[SEP]':
pass
else:
calulate_label.append(k)
indices.append(v)
print('calulate_label: {}'.format(calulate_label))
print('indices: {}'.format(indices))
weight = tf.sequence_mask(FLAGS.max_seq_length)
precision = tf_metrics.precision(label_ids, pred_ids,
num_labels, indices, weight)
recall = tf_metrics.recall(label_ids, pred_ids, num_labels,
indices, weight)
f = tf_metrics.f1(label_ids, pred_ids, num_labels, indices,
weight)
# print('预测实体与真实实体评价结果......')
# entity_precision, entity_recall, entity_f1 = entity_metrics(input_ids, label_ids, pred_ids, id2char, id2tag)
# print("Entity Test P/R/F1: {} / {} / {}".format(round(entity_precision, 2), round(entity_recall, 2), round(entity_f1, 2)))
return {
"eval_precision": precision,
"eval_recall": recall,
"eval_f": f,
# "eval_loss": loss,
}