bys0318 / squire Goto Github PK

您好，我在生成规则时，按照给出的命令，总是报错：
reading params from file config-learn.properties
Exception in thread "main" java.io.FileNotFoundException: path1/path_log (No such file or directory)
at java.io.FileOutputStream.open0(Native Method)
at java.io.FileOutputStream.open(FileOutputStream.java:270)
at java.io.FileOutputStream.(FileOutputStream.java:213)
at java.io.FileOutputStream.(FileOutputStream.java:101)
at java.io.PrintWriter.(PrintWriter.java:184)
at de.unima.ki.anyburl.LearnReinforced.main(LearnReinforced.java:141)
请问您是否知道原因？

错误代码
RuntimeError: Error(s) in loading state_dict for TransformerModel:
size mismatch for encoder.weight: copying a param with shape torch.Size([15021, 256]) from checkpoint, the shape in current model is torch.Size([13646, 256]).
训练完进行测试为什么会显示大小不对呢

Yushi你好，对你的这篇工作很感兴趣，不知道是否可以分享一下AnyBURL生成的rules.dict文件，没用过AnyBURL，十分感谢！

你好，我想问一下这个path-1000是不是没有给出来呢？

File "utils.py", line 34, in filter_rule
with open(path, 'r') as f:
FileNotFoundError: [Errno 2] No such file or directory: 'path1/path-1000'

while executing the run.sh file, i get the following error msg.

FileNotFoundError: [Errno 2] No such file or directory: 'path1/path-1000'

您好，感谢您发布的代码。
我们在train.py的evaluate()函数中，观察max-len参数，实则整个计算预测概率的model.logits()应该只运行了两次，先后得到两个在所有token上的预测概率。其中在evaluate()函数中for l in range(2, max_len)的参数，max_len看到训练与测试使用默认赋值3，那么则会在该函数candiates的计算也就只有一次，显然并不符合实验结果，因为每次调用train.py中的logits(）仅输出一个token（关系or实体）的概率。因此这里是需要修改max-len参数吗，修改为6？

Hi,

Thanks for sharing the codes. Your work is quite interesting. However, as I was reproducing the experiments, I found that every time after performing the iterative training, the GPU memory increases significantly and will not drop back to what it was before the iterative training. This might cause GPU out of memory issues. Would you investigate this problem and see if it is reproducible on your side?

b/w: could you also share the hyperparameters for other datasets such as FB15K237-20? Thanks!

Best,
Zhongyu

你好，在model.TransformerModel.get_loss函数里实现label smoothing的时候，(1 - self.lable_smooth) / (self.ntoken - 1) * lprobs.sum(dim=-1)这一项乘积是包括了目标token的对数概率吗，似乎与论文里的公式不符，还望解答，感谢~

为什么论文中提到的benchmark模型的实验结果比它们原文中的好那么多？