Subject: LBFGS for use in large maximum likelihood problem
From: dlakelan
Date: Tue, 19 Aug 2008 15:55:24 -0700
dlakelan wrote:
> Robert Dodier wrote:
>> On 8/8/08, dlakelan <dlakelan at street-artists.org> wrote:
> ...
>> Dan, I think the problem in this case is that diff doesn't treat
>> lsum the same as sum --- specifically that diff distributes over
>> sum but not lsum.
> ....
>> Anyway hope this sheds some light on the problem.
>
> Hi Robert,
>
> It seems to me that it takes longer to do each step of the LBFGS when I
> use the "sum" noun than when I construct the huge expression, although
> the sum noun is of course much faster to construct. I didn't profile it
> though since I was working with my main problem not the simpler example
> one that I posted.
....
aha, I've thought of one thing that's relevant here... indexing the list
is quadratic in the length of the list... as far as I know... so
sum(f(data[i]),i,1,length(data)) takes O(length(data)^2)... right? is
there a way I can formulate this to take a single pass down the data
list, and still not construct a massive symbolic sum?