Zhibin (Louis) Lu

I don't see you do truncate when the sequence longer than 512 lengths?

1

We know that when loading pretrained bert, max_seqence=512, but when you are dealing with the sentences from data set, each sentence is padding based on the longest sentence of the...

I am just wondering why there is not the test set score for DistilBert (Sanh 2019) in Table3 in the paper?

在题目矩阵中的路径(12)的算法描述上好像有点不对

2

算法应该是在匹配到第一个字符后，才真正开始递归dfs，而在匹配不到第一个字符的时候，是双循环继续在矩阵中查找第一个字符，这个时候并没从矩阵当前字符的4个方向去走。当然匹配到一半失败的，也会在双循环中继续往下重新找第一个字符。我的意思，寻找第一个匹配字符失败的时候，并不是朝4个方向走。

1.Notebook 07, Question 21: if s[0] == "UP": x -= int(s[1]) -> y+= if s[0] == "DOWN": x += int(s[1]) -> y-= if s[0] == "LEFT": y -= int(s[1]) ->...

I have been waiting for a longtime, something like scroll automatique the output to bottom. @DonJayamanne please...

疑问，我仔细看了您的程序，test_set参与了idf和pmi的计算，这是否合理？

2

我感觉您的test_set更应该像valid_set的功能，因为参与了idf和pmi的计算过程，所以论文中的performance应该是valid_set的。 但是，如果要另外分出一个真正的test_set去和其他论文比较多话(比如我将这模型应用在其他数据集)，公平起见的话，test_set不应该参与训练期间的idf和pmi的计算。所以这里有个难点，因为做评估的时候预测值是直接从node获得的，test_set如何做到不参与训练，并且有node在graph中呢？ 请指教！

Can we control the collapse setting?

1

When I unfold a node, all the child nodes (includes all recursive descendants) under this node will be unfolded. I think the better feature is unfold the direct child nodes...

The quicklook window is too big and can not remembered after resize

2

as the title. Thanks for the tool!