|
PureBytes Links
Trading Reference Links
|
On Sun, 2 Jan 2000, Alan Mi wrote:
> You are right. But by "double precision" from Larry, I took it to mean
that EL variables should use 2 > 32-bit words for implementation so that
dates like 20000103 could be stored without losing accuracy > and we could
do without 1000103. That would double the memory use for all variables
including arrays.
That is true, but is a 'quirk' of the way EL works. EL is not necessarily
efficient in space or speed.
If I were actually doing it, I'd prefer to have types, as the rest of the
programming world does. Also. Long integers can do dates just fine, and
are 32 bits. If you open the types can of worms, there are other options
that might make things smaller as well as larger.
> I did make a mistake though. Memory used for native data stream series
like Close, Volume etc would > not double under the suggested scheme since
they are, according to SDK documentation, LONG's. So > Larry, I don't have
a reservation as strong as before any more. Yet still, bear in mind that
in series > functions (indicators and signals alike), each variable or
array element takes MaxBarsBack times of > the memory used by a simple
variable. That's could still be huge for certain studies.
True, but this is an often undesirable side effect of EL, NOT a
fundamental problem. If one did away with this, the space might even be
smaller. I have non-EL routines that do larger calcs than TS, and probably
use less memory because I have control over types, and don't automatically
make these huge arrays that might not be needed.
DP would be very useful for me, and I'd like types even more :-).
Larry
|