Prev: question about intergal
Next: Event driven simulation
From: dpb on 1 Jun 2010 14:36
Luna Moon wrote:
> In theory, Matlab's vectorized operations should be able to allow us
> process gigantic size tick data fast.

What is "tick data", pray tell???

Ticks are nasty little buggers that carry various infectious diseases...

> However, Matlab's limited by memory size so it is not as good as other
> software for processing tick data.

What makes Matlab any more memory-size limited than any other analysis
tool? There's only so much memory available on any given system...

> However, if we process data in trunks Matlab should be still usable?

OK, now what does "trunk" mean in this context? You mean "chunks" or
piecewise? In that case, yes, that would be one way w/ Matlab (as with
any other tool).

--
From: dpb on 2 Jun 2010 09:20
dpb wrote:
> Luna Moon wrote:
....

>> However, Matlab's limited by memory size ...
>
> What makes Matlab any more memory-size limited than any other analysis
> tool? There's only so much memory available on any given system...
....

Actually, I still don't know what "tick data" is supposed to be but...

There is one size limitation w/ ML that has always seemed somewhat
frustrating (and afaik hasn't been modified significantly in later
versions) and that is that the default data size of DOUBLE means that
for much computation where SINGLE is adequate that only half the data
that otherwise could be handled in memory can be.

That the SINGLE class is a somewhat poor cousin w/ no math operations
means one either has to be able to do without or write the operations
oneself.

Which raises a question...since at least most rudimentary functions like
sum(), mean(), etc., do work on single, what justification does TMW use
for not implementing the basic operators? And, do the functions make
temporary double arrays behind the scenes to operate? Just curious
about the "why" and "how" of some of these choices TMW made...

--
From: dpb on 2 Jun 2010 09:59
Steven Lord wrote:
> "dpb" <none(a)non.net> wrote in message
> news:hu5lvl$5ad$1(a)news.eternal-september.org...
....

>> That the SINGLE class is a somewhat poor cousin w/ no math operations
>> means one either has to be able to do without or write the operations
>> oneself.
>
> If I remember correctly, you're using an older version. We implemented
> non-double arithmetic (including single precision and the smaller six
> integer types) a while ago:
>
> http://www.mathworks.com/access/helpdesk/help/techdoc/rn/f7-998197.html
>
> Release R14, when we introduced that functionality, came out exactly six
> years ago today.

Well, I _almost_ went and looked but the load time of web pages w/ my
dialup connection convinced me not to... :( Sorry.

Anyway, it's good to know it was done and that would allow the OP in
this thread could handle larger datasets by treating most as single
precision, perhaps.

Presuming the "tick data" is time-stamped data the subject of multiple
other threads, datenum() needs the double but the corollary data might
well be ok as single precision; certainly for simply plotting/visualization.

--
 | 
Pages: 1
Prev: question about intergal
Next: Event driven simulation