I'd like to start with the vlist implementation Mark mentioned previously,
it would be cool if this can help speed up the compiler. But since I'm new
here, would anybody be kind enough to give me some clue on how to integrate
a c-version vlist into current system? or show me some documents so that I
can find the answer myself.
> On Sun, 2012-08-12 at 22:31 +0800, rushan chen wrote:
> > Hi Mark,
> > Very appreciate for your reply.
> > I see you mention that it's useful to implement a larger library of
> > efficient data structure, and I'm interested in that very much. I used to
> > work on projects which involve complicated but very interesting data
> > structures, implementing them could be challenging, but once done I feel
> > great sense of achievement.
> > One such project is implementing a language model (LM) which is a core
> > component of speech recognition and machine translation. I don't know if
> > you heard of it before. Unfortunately, I can't cover it too detailed
> > that would complicate things too much.
> > Basically, one of the key operations LM supports is it should return a
> > probability associated with any given id sequence. All id sequences are
> > the same length, and there are a mass amount of such id sequences (a
> > commonly-seen LM may contain billions of them). So it's required to store
> > LM in a concise way, and at the same time make the search for each id
> > sequence very quickly.
> OK, it's very good
> > Trie is finally chosen to be the data structure for LM (there were many
> > papers discussing this issue). All id sequences with the same prefix
> > the same internal node, for example, for <1, 2, 3, 4> and <1, 2, 3, 5>,
> > only one copy of <1, 2, 3> will be stored in LM, and a search for a id
> > sequence is done by a sequence of binary search until the leaf is met.
> > extra thing worth mentioning is that I store the whole trie structure in
> > single large piece of memory (usually around 2 gigabytes), which makes
> > it convenient to write out to disk and load into memory by simply using
> > mmap, and I think it also makes the system faster than if you allocate
> > memory every time it's needed.
> Seems we don't have any prefix-tree implementation yet?
> Maybe some hero too busy to share it? ;-)
> I'd like to see you make it, or I must write myself one.
> IIRC, many guys here wrote their own data-structure/algorithm
> Sharing makes our world better.
> But, sometimes we reinvent wheels just for fun.
> So just do what you want to do if it's interesting to you.
> > There are some other projects I worked or working on like Spell
> > which also involve complicated data structures, but due to privacy
> > I can't say much about it.
> If something force you not to share, you may rewrite it all by
> friends will see your creativity, and your work be enhanced by others.
> > All in all, I'm very interested in it, and I really really hope I can
> > Looking forward to your reply. Thanks in advance.
> > Have fun!
> happy hacking!
> > Rushan Chen