R help - Jun 2009 - (FULL) Need help to optimize a piece of code involving zoo objects

(Sorry, sent the message before I finished it)
Hello, everyone

I have a long script that uses zoo objects. In this script I used
simple moving averages and these I can very efficiently calculate with
filter() functions.
Now, I have to use special "exponential" moving averages, and the only
way I could write the code was with a for-loop, which makes everything
extremely slow.
I don't know how to optimize the code, but I need to find a solution.
I hope someone can help me.

The special moving average is calculated in the following way:

EMA = ( K x ( C - P ) ) + P

where,

C = Current Value
P = Previous periods EMA ? ?(A SMA is used for the first period's
calculation)
K = Exponential smoothing constant

K = 2 / ( 1 + Periods )

Below is the code with the for-loop.

-"temp" contains C
-Periods is variable "j" in the for loop (so K varies)
- I first produce a vector of simple equally weighted moving average,
and use the first non-NA value to initiate the second for-loop

x.Date <- as.Date("2003-02-01") + seq(1,1100) - 1
temp <- zoo(rnorm(1100, 0, 10)+100, x.Date)

start.time <- proc.time()

for(j in seq(5,100,by=5)){

? ? ? ?#PRODUCE FAST MOVING AVERAGE
? ? ? ?#Create equally weighted MA vector (we need only the first value)
? ? ? ?smafast <- zoo(coredata(filter(coredata(temp[,1]), filter=rep(1/j,
j), sides=1)), order.by=time(temp))

? ? ? ?#index of first non-NA value, which is the first SMA needed
? ? ? ?#which(is.na(smafast))[length(which(is.na(smafast)))]+1

? ? ? ?#Calculate decay factor K
? ? ? ?#number of periods is j
? ? ? ?K <- 2/(1+j)

? ? ? ?#Calculate recursively the EMA for the fast index (starting with
second non-NA value)
? ? ? ?for (k in
(which(is.na(smafast))[length(which(is.na(smafast)))]+2):length(smafast))
{
? ? ? ? ? ? ? ?smafast[k] <-
coredata(smafast[k-1])+K*(coredata(temp[k,1])-coredata(smafast[k-1]))
? ? ? ?}

? ? ? ?#PRODUCE SLOW MOVING AVERAGE
? ? ? ?#Create equally weighted MA vector (we need only the first value)
? ? ? ?smaslow <- zoo(coredata(filter(coredata(temp[,1]),
filter=rep(1/(j*4), (j*4)), sides=1)), order.by=time(temp))
? ? ? ?K <- 2/(1+j*4)
#Calculate EMA
? ? ? ?for (k in
(which(is.na(smaslow))[length(which(is.na(smaslow)))]+2):length(smaslow))
{
? ? ? ? ? ? ? ?smaslow[k] <-
coredata(smaslow[k-1])+K*(coredata(temp[k,1])-coredata(smaslow[k-1]))
? ? ? ?}

? ? ? ?#COMBINE DIFFERENCES OF FAST AND SLOW
? ? ? ?temp <- ? ? ? ? merge(temp, ma=smafast-smaslow)
}

proc.time()-start.time

--------------------------------------------------------------------------------------
Could someone help me to optimize the two EMA for-loops within the
bigger for-loop?
I need to cut the execution time down by at least half.

Thank you in advance for your help!

Best,
Sergey

Check out method = "recursive" in filter().

On Fri, Jun 19, 2009 at 3:36 AM, Sergey Goriatchev<sergeyg at gmail.com>
wrote:
> (Sorry, sent the message before I finished it)
> Hello, everyone
>
> I have a long script that uses zoo objects. In this script I used
> simple moving averages and these I can very efficiently calculate with
> filter() functions.
> Now, I have to use special "exponential" moving averages, and the
only
> way I could write the code was with a for-loop, which makes everything
> extremely slow.
> I don't know how to optimize the code, but I need to find a solution.
> I hope someone can help me.
>
> The special moving average is calculated in the following way:
>
> EMA = ( K x ( C - P ) ) + P
>
> where,
>
> C = Current Value
> P = Previous periods EMA ? ?(A SMA is used for the first period's
calculation)
> K = Exponential smoothing constant
>
> K = 2 / ( 1 + Periods )
>
> Below is the code with the for-loop.
>
> -"temp" contains C
> -Periods is variable "j" in the for loop (so K varies)
> - I first produce a vector of simple equally weighted moving average,
> and use the first non-NA value to initiate the second for-loop
>
> x.Date <- as.Date("2003-02-01") + seq(1,1100) - 1
> temp <- zoo(rnorm(1100, 0, 10)+100, x.Date)
>
> start.time <- proc.time()
>
> for(j in seq(5,100,by=5)){
>
> ? ? ? ?#PRODUCE FAST MOVING AVERAGE
> ? ? ? ?#Create equally weighted MA vector (we need only the first value)
> ? ? ? ?smafast <- zoo(coredata(filter(coredata(temp[,1]),
filter=rep(1/j,
> j), sides=1)), order.by=time(temp))
>
> ? ? ? ?#index of first non-NA value, which is the first SMA needed
> ? ? ? ?#which(is.na(smafast))[length(which(is.na(smafast)))]+1
>
> ? ? ? ?#Calculate decay factor K
> ? ? ? ?#number of periods is j
> ? ? ? ?K <- 2/(1+j)
>
> ? ? ? ?#Calculate recursively the EMA for the fast index (starting with
> second non-NA value)
> ? ? ? ?for (k in
> (which(is.na(smafast))[length(which(is.na(smafast)))]+2):length(smafast))
> {
> ? ? ? ? ? ? ? ?smafast[k] <-
> coredata(smafast[k-1])+K*(coredata(temp[k,1])-coredata(smafast[k-1]))
> ? ? ? ?}
>
> ? ? ? ?#PRODUCE SLOW MOVING AVERAGE
> ? ? ? ?#Create equally weighted MA vector (we need only the first value)
> ? ? ? ?smaslow <- zoo(coredata(filter(coredata(temp[,1]),
> filter=rep(1/(j*4), (j*4)), sides=1)), order.by=time(temp))
> ? ? ? ?K <- 2/(1+j*4)
> #Calculate EMA
> ? ? ? ?for (k in
> (which(is.na(smaslow))[length(which(is.na(smaslow)))]+2):length(smaslow))
> {
> ? ? ? ? ? ? ? ?smaslow[k] <-
> coredata(smaslow[k-1])+K*(coredata(temp[k,1])-coredata(smaslow[k-1]))
> ? ? ? ?}
>
> ? ? ? ?#COMBINE DIFFERENCES OF FAST AND SLOW
> ? ? ? ?temp <- ? ? ? ? merge(temp, ma=smafast-smaslow)
> }
>
> proc.time()-start.time
>
>
--------------------------------------------------------------------------------------
> Could someone help me to optimize the two EMA for-loops within the
> bigger for-loop?
> I need to cut the execution time down by at least half.
>
> Thank you in advance for your help!
>
> Best,
> Sergey
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>