Minh Le
Minh Le
For a solution for the second problem, see issue #20.
No, that's the flatten call in the forward() function.
When I run it on real data, I got this error: ``` $ ./oxlm.job ################################ # Config Summary # order = 5 # feature context size = 5 # word_width...
Turns out that I need to specify hash-space (default is 0 which causes division error). A meaningful error message would be useful.
There should be something like: public synchronized Article[] getArticles(final String names[], final int properties) public synchronized SimpleArticle[] readData(final String names[], final int properties) Combining calls is not just about efficiency...
I'm curious about how these features actually work. Looking at the [cort](https://github.com/smartschat/cort) analysis of CoNLL-2012 data, many mentions of the same name are not linked. For some reason, most of...
My friend and I found the problem: some unaligned pointers ended up in Eigen and broke it. This is a stack trace: (`__P=0x2aaab8d64bc8` is not 16-aligned pointer). ``` #0 0x00002aaaabe7c49b...
Yeah I'm thinking along the line of memory allocation too. If you know before hand the number of nodes and edges, you might benefit from allocating everything once at the...