Re: st: Slowing process when running a program with multiple nested loops

[David Kantor <kantor.d@att.net>]

Hi,

Your code looks fairly straightforward, after looking at it for a minute.
At first, it looked cryptic, but once I understood it, I realized, I 
would code very similarly.
There are a few unused macros, but that's irrelevant.

We don't see the code for sma, dma, trb, or cbo. Do these get 
progressively complicated?
Is it possible that there is sudden jump in slowness when you switch 
from sma to dma, or to trb or cbo?
Or is it gradual through all the iterations?

(TBut you did say that they do about the same amount of calculation)

More importantly, do they alter the data? Do they alter (-save-) the data file?
These latter points may be most relevant.
The important question is, after one iteration, can the next one run 
without reloading (-use-ing) the data?
If not, can you rework your code (in sma, dma, trb, and cbo) to make 
it so? (That is, have them not drop or add records. If they generate 
or replace values of variables, have those be in designated variables 
that can be reset easily. The idea being that if the dataset changes 
in a significant way, then you want to be able to bring it back to 
its pre-iteration state easily -- using -drop- or -replace ... = .-. 
The last thing you should have to do is to reload the data for each 
iteration. Reloading the data may be 1000 times slower than 
continuing with the same data. (I don't have any real statistics on 
that factor, but 1000 is not unreasonable.)
If you can arrange it so that you don't need to reload on each 
iteration (or if it is already coded that way),  then you can you 
move the -use- command to the top -- before the first foreach?
Note that the repeated reloading will cause slowness, but may not 
exactly explain why it gets progressively slower. But that may be an 
operating-system issue. (It may be that after the first -use-, the 
file is in cache, enabling some fast loads; later it is knocked out of cache.)
One other point is that it is not always good to -set mem- to a high 
value. It should be high enough to get the job done, plus maybe a 
little margin of safety. Otherwise, you are grabbing space that might 
better left for the operating system to make good use of (such as for 
cacheing files) and to run everything (including your task) smoother 
and faster.
HTH
--David


At 02:52 PM 1/14/2013, you wrote:
> Thank you for your response.
> I did mean to say that there are 7 nested loops, because there are 7
> parameters that can change values, and I do not know of another way to
> have this done.
>
> So the code is as followed:
> ** Initialization
> clear all
> set mem 100m
> set more off, perm
> set autotabgraphs on, perm
> graph drop _all
> cd "C:\Users\Trades\Stata"
> sysdir set PERSONAL "C:\Users\Trades\Stata\Ado"
>
> ** Setting parameters
>  global freq "1m"
>  global fcrc "EUR USD GBP AUD NZD NZD EUR EUR AUD"  // foreign currencies
>  global bcrc "USD JPY USD USD USD JPY JPY GBP JPY"  // base currencies
>  global startdate = mdy(1,1,1994)
>  global enddate = mdy(12,31,2010)
>  global subperiod "2002jan01 2008sep01"   // specify subperiods
>
>
>  local smam "2 5 10 15 20 25 50 100 150 200 250"  // parameter m 
> for sma method
>  local smab "0 0.0005 0.001 0.005 0.01 0.05"   // parameter b for sma method
>  local smad "2 3 4 5"         // parameter d for sma method
>  local smac "5 10 25"          // parameter c for sma method
>  local sman "0"
>  local smak "0"
>
>  local dmam "2 5 10 15 20 25 50 100 150 200 250"  // parameter m 
> for dma method
>  local dman "2 5 10 15 20 25 50 100 150 200"   // parameter n for dma method
>  local dmab "0 0.0005 0.001 0.005 0.01 0.05"   // parameter b for dma method
>  local dmad "2 3 4 5"         // parameter d for dma method
>  local dmac "5 10 25"          // parameter c for dma method
>  local dmak "1000"
>  local trbn "5 10 15 20 25 50 100"      // parameter n for trb method
>  local trbb "0.0005 0.001 0.005 0.01 0.025 0.05"  // parameter b 
> for trb method
>  local trbd "2 3 4 5"         // parameter d for trb method
>  local trbc "1 5 10 25"         // parameter c for trb method
>  local trbm "1000"
>  local trbk "0"
>  local cbon "5 10 15 20 25 50 100 200"     // parameter n for cbo method
>  local cbok "0.001 0.005 0.01 0.05 0.1"     // parameter k for cbo method
>  local cbob "0.0005 0.001 0.005 0.01 0.05"    // parameter b for cbo method
>  local cbod "0 1 2"          // parameter d for cbo method
>  local cboc "1 5 10 25"         // parameter c for cbo method
>  local cbom "1000"
>
>
> ** Loops to go through all methods and all parameters
>  foreach med in sma dma trb cbo {      // loop through all the rules
>   foreach m of local `med'm {      // all the m values
>    foreach n of local `med'n {     // all the n values
>      foreach k of local `med'k {    // all the k values
>       foreach b of local `med'b {   // all the b values
>        foreach d of local `med'd {  // all the d values
>         foreach c of local `med'c { // all the c values
>          clear
>          use data
>          `med', datevar(date) m(`m') n(`n') k(`k') b(`b') d(`d') c(`c')
>
>         }
>        }
>       }
>      }
>     }
>   }
>  }
>
> `med' is calling one of the four ado files that I wrote: sma, dma,
> trb, and cbo.  It basically calculates profits based on the rule and
> the parameters fed to the program, so I think each iteration does just
> about the same amount of calculation.
>
> The next part (which I haven't written) keeps track of results from
> some methods and parameters that satisfy certain conditions. In my
> opinion, this would be a minor thing if I can get the current code to
> run in a reasonable amount of time.
>
> Any help explaining why the program slows down so significantly after
> a couple of hundreds of iterations will be much appreciated.
>
> Thank you.
>
>
>
>
>
>
>
>
>
> I think that most of would agree that we would need to see your code
> to be able to say what the problem is. Meanwhile, did you mean that
> the loops are nested to a depth of 7? That's unusually deep.
>
> Just generally speaking, with loops, there are often actions that are
> placed inside that don't need to be there; they can be moved "up" or
> "out" (sometimes requiring a bit of modification) so as to not be done
> multiple times unnecessarily. From what you describe, it seems that
> the work done in each iteration is accumulating; each iteration does a
> bit more work than the previous. There may be some unnecessary
> repetition as described above. But it also seems that there is
> something that grows and gets dragged along with each iteration --
> again possibly unnecessarily. This is analogous to a cumulative song,
> such as "The Twelve Days of Christmas"; the 12th verse is much longer
> than the first.
>
> On the other hand, does the true complexity of the task grow with each
> iteration? Do you expect the 300th iteration to naturally be more
> complex to perform than the first?
>
> Show us your code if you want more help.
>
> HTH
> --David
>
> --
> Ly Tran
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/faqs/resources/statalist-faq/
> *   http://www.ats.ucla.edu/stat/stata/
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/