zhixiuye / intra-bag-and-inter-bag-attentions Goto Github PK

Could you please also publish your PR curve data points . I think that will make the successors easy to compare their new models with yours when they want to plot both their and your results. Thanks~

Hello

I am trying to use a different dataset as input to the model.

What are the things I need to do?

I was trying to replicate the files that you have in the NYT_data directory but am confused with vec.bin and the first two columns in train.txt and test.txt.

Thank you in advance for any feedback you can give me

您好，我想请问一下关于P@N评测，按照代码中评测P@100应该是个整数，请问您论文中是实验多次取平均数吗？table 4中Lin2016的p@300数据好像打印错了

您好，我用自己的数据集做成您的模型需要输入的格式，训练后召回率和精确率是非法的，为什么呢？是因为我的数据集太小了吗？或者是batch_size不合适？

I think your codes has a little mistake. in the train.py.

if args.use_RA and args.sent_encoding == "pcnn": loss = model.PCNN_ATTRA(total_word, total_pos1, total_pos2, total_pcnnmask, total_shape, y_batch) if args.use_RA and args.sent_encoding == "cnn": loss = model.CNN_ATTRA(total_word, total_pos1, total_pos2, total_pcnnmask, total_shape, y_batch) if not args.use_RA and args.sent_encoding == "pcnn": loss = model.PCNN_ATT**RA**(total_word, total_pos1, total_pos2, total_pcnnmask, total_shape, y_batch) if not args.use_RA and args.sent_encoding == "cnn": loss = model.CNN_ATTBL(total_word, total_pos1, total_pos2, total_pcnnmask, total_shape, y_batch)

line 278should be "model.PCNN_ATTBL"

In preprocess/data2pkl.py line 43, 44:

sent = f.readline().strip().split(",")
sentences.append([(x+1) for x in list(map(int, sent))])

sent should be the word id list of current sentence. But why x+1 (instead of x)?

Is it because a new token '' is added to vocaburary?

Thanks.

请问extract.cpp里面的word2vec.txt，bags_train.txt，bags_test.txt　数据集里怎么没有这些文件？　还有我要换数据集的话，是不是还得处理成train.txt里面的格式？代码里好像没有处理成train.txt格式的过程。谢谢解答！

Hello! i am reading your paper and running the code, but i have a pronlem. Can you tell me How to solve this problem "RuntimeError : dimension specified as 0 but tensor has no dimensions"? Thank you!

I found that the instance number of train data of Lin2016(and many other papers) is about 520000, but that of yours is about 570000. So, where do you get your data from? Thank you.