保存模型，默认使用“模型名字+时间”作为文件名

可加载指定路径的模型

Using Attention to get all bags embedding in a batch

parse the records in data

parse the records in data

calculate the precision and recall for p-r curve reglect the NA relation

Apply the evalation method in Lin 2016

train process

get all bags embedding in one batch before Attention

x: all instance in a Bag

padding the sentences

refer: github.com/SharmisthaJat/RE-DS-Word-Attention-Models

clip the postion range: : -limit ~ limit => 0 ~ limit * 2+2 : -51 => 1 : -50 => 1 : 50 => 101 : >50:

: word embedding : postion embedding return: sen list pos_left pos_right

: word embedding : postion embedding return: sen list pos_left pos_right

use xavier to init

use xavier to init

reading from vec.bin add two extra tokens: : UNK for unkown tokens

reading from vec.bin add two extra tokens: : UNK for unkown tokens : BLANK for the max len sentence

refer: https://github.com/thunlp/OpenNRE A fast piecewise pooling using mask

refer: https://github.com/thunlp/OpenNRE A fast piecewise pooling using mask

Apply the prediction method in Lin 2016

test process

custom for DataLoader

padding the sentences

clip the postion range: : -limit ~ limit => 0 ~ limit * 2+2 : -51 => 1 : -50 => 1 : 50 => 101 : >50:

a util function for generating a LongTensor Variable

user can update the default hyperparamter

piecewise pool into 3 segements x: the batch data insPool: the batch Pool

old version piecewise

Apply the prediction method in Zeng 2015