StructTS hang?

Ein eingebundener Text mit undefiniertem Zeichensatz wurde abgetrennt.
Name: nicht verf?gbar
URL: <https://stat.ethz.ch/pipermail/r-help/attachments/20100125/737acde8/attachment.pl>
Hi!

You have probably sorted out your problem a different way.
However I have tried dealing with the same problem for a while and writing down the reason for this here may be of use to someone.

In some other help-mail I read that time series objects behave unexpectedly in some situations if the frequency is too high. As a result of this, a call of StructTS (with seasonality) hangs if and only if the frequency is larger than 24 (if I remember the number right). This is what I found with my own time series using several frequencies.

So, this may not be helpful using StructTS on a time series, but it may save somebody some time.
Even better: report a reproducible example to the package maintainer who 
is certainly happy to fix bugs in the package.

Best wishes,
Uwe Ligges
Regards,
Richard

----------------------------------------------------------------
Irgendwann kommt jeder drauf! WWW.ENERGIEWELT.DE
-----------------------------------------------------------------------------------
Vorsitzender des Aufsichtsrates: Dr. Rolf Martin Schmitz
Vorstand: Dr. Johannes Lambertz (Vorsitzender),
Prof. Dr. Gerd Jaeger, Antonius Voss,
Erwin Winkel

Sitz der Gesellschaft: Essen und Koeln
Eingetragen beim Amtsgericht Essen
Handelsregister-Nr. HRB 17420
Eingetragen beim Amtsgericht Koeln
Handelsregister-Nr. HRB 117

	[[alternative HTML version deleted]]

______________________________________________
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.
No problem!

Working:
z<-ts(sin(1:1000), start=1, freq=10)
x<-StructTS(z, type='BSM')
Hang:
z<-ts(sin(1:1000), start=1, freq=100)
x<-StructTS(z, type='BSM')
(Both on WinXP, R 2.9.2)

I don't, however, know who the package maintainer is, nor can I track down where exactly the hang is caused.

Maybe somebody else can help me with that.

Cheers,
Richard Kiefer

-----Urspr?ngliche Nachricht-----
Von: Uwe Ligges [mailto:ligges at statistik.tu-dortmund.de]
Gesendet: Dienstag, 26. Januar 2010 10:32
An: Kiefer, Richard
Cc: r-help at r-project.org
Betreff: Re: [R] StructTS hang?
Hi!

You have probably sorted out your problem a different way.
However I have tried dealing with the same problem for a while and writing down the reason for this here may be of use to someone.

In some other help-mail I read that time series objects behave unexpectedly in some situations if the frequency is too high. As a result of this, a call of StructTS (with seasonality) hangs if and only if the frequency is larger than 24 (if I remember the number right). This is what I found with my own time series using several frequencies.

So, this may not be helpful using StructTS on a time series, but it may save somebody some time.
Even better: report a reproducible example to the package maintainer who
is certainly happy to fix bugs in the package.

Best wishes,
Uwe Ligges
Regards,
Richard

----------------------------------------------------------------
Irgendwann kommt jeder drauf! WWW.ENERGIEWELT.DE
-----------------------------------------------------------------------------------
Vorsitzender des Aufsichtsrates: Dr. Rolf Martin Schmitz
Vorstand: Dr. Johannes Lambertz (Vorsitzender),
Prof. Dr. Gerd Jaeger, Antonius Voss,
Erwin Winkel

Sitz der Gesellschaft: Essen und Koeln
Eingetragen beim Amtsgericht Essen
Handelsregister-Nr. HRB 17420
Eingetragen beim Amtsgericht Koeln
Handelsregister-Nr. HRB 117

      [[alternative HTML version deleted]]

______________________________________________
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.
----------------------------------------------------------------
Irgendwann kommt jeder drauf! WWW.ENERGIEWELT.DE
-----------------------------------------------------------------------------------
Vorsitzender des Aufsichtsrates: Dr. Rolf Martin Schmitz
Vorstand: Dr. Johannes Lambertz (Vorsitzender), 
Prof. Dr. Gerd Jaeger, Antonius Voss, 
Erwin Winkel

Sitz der Gesellschaft: Essen und Koeln
Eingetragen beim Amtsgericht Essen
Handelsregister-Nr. HRB 17420
Eingetragen beim Amtsgericht Koeln
Handelsregister-Nr. HRB 117
No problem!

Working:
z<-ts(sin(1:1000), start=1, freq=10)
x<-StructTS(z, type='BSM')
Hang:
z<-ts(sin(1:1000), start=1, freq=100)
x<-StructTS(z, type='BSM')
(Both on WinXP, R 2.9.2)

I don't, however, know who the package maintainer is, nor can I track down where exactly the hang is caused.

Maybe somebody else can help me with that.
Apologies, thought that code was from a contributed package rather than 
from the base distribution of R.
In this case, and given the code, we should try to debug a little bit.

My first impression is that it will take some time given experiments 
with smaller numbers it looks roughly exponential in runtime. I'll be 
off for a meeting now and will look at results in 1-2 hours (when I 
expect them to appear on my machine).

Uwe Ligges
Cheers,
Richard Kiefer

-----Urspr?ngliche Nachricht-----
Von: Uwe Ligges [mailto:ligges at statistik.tu-dortmund.de]
Gesendet: Dienstag, 26. Januar 2010 10:32
An: Kiefer, Richard
Cc: r-help at r-project.org
Betreff: Re: [R] StructTS hang?

On 25.01.2010 11:03, Richard.Kiefer at rwe.com wrote:
Hi!

You have probably sorted out your problem a different way.
However I have tried dealing with the same problem for a while and writing down the reason for this here may be of use to someone.

In some other help-mail I read that time series objects behave unexpectedly in some situations if the frequency is too high. As a result of this, a call of StructTS (with seasonality) hangs if and only if the frequency is larger than 24 (if I remember the number right). This is what I found with my own time series using several frequencies.

So, this may not be helpful using StructTS on a time series, but it may save somebody some time.

Even better: report a reproducible example to the package maintainer who
is certainly happy to fix bugs in the package.

Best wishes,
Uwe Ligges

Regards,
Richard

----------------------------------------------------------------
Irgendwann kommt jeder drauf! WWW.ENERGIEWELT.DE
-----------------------------------------------------------------------------------
Vorsitzender des Aufsichtsrates: Dr. Rolf Martin Schmitz
Vorstand: Dr. Johannes Lambertz (Vorsitzender),
Prof. Dr. Gerd Jaeger, Antonius Voss,
Erwin Winkel

Sitz der Gesellschaft: Essen und Koeln
Eingetragen beim Amtsgericht Essen
Handelsregister-Nr. HRB 17420
Eingetragen beim Amtsgericht Koeln
Handelsregister-Nr. HRB 117

       [[alternative HTML version deleted]]

______________________________________________
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.
----------------------------------------------------------------
Irgendwann kommt jeder drauf! WWW.ENERGIEWELT.DE
-----------------------------------------------------------------------------------
Vorsitzender des Aufsichtsrates: Dr. Rolf Martin Schmitz
Vorstand: Dr. Johannes Lambertz (Vorsitzender),
Prof. Dr. Gerd Jaeger, Antonius Voss,
Erwin Winkel

Sitz der Gesellschaft: Essen und Koeln
Eingetragen beim Amtsgericht Essen
Handelsregister-Nr. HRB 17420
Eingetragen beim Amtsgericht Koeln
Handelsregister-Nr. HRB 117

Exponential, you say?

Maybe I am just not patient enough.
After being unresponsive for several minutes, which in the larger examples that I work with has probably appeared like a hang, the code I just mentioned has given a result (a convergence error, but still).
So, I must say I can no longer reproduce the hang.

I remember to be able to reproduce a significant change in behaviour upon the frequency passing some threshold. Maybe by chance, this threshold parting "hang" from normal operation agreed with a number someone claimed ts-objects could not handle properly.
But I can't reproduce this now. I'm sorry...

-----Urspr?ngliche Nachricht-----
Von: Uwe Ligges [mailto:ligges at statistik.tu-dortmund.de]
Gesendet: Dienstag, 26. Januar 2010 11:15
An: Kiefer, Richard
Cc: r-help at r-project.org
Betreff: Re: AW: [R] StructTS hang?
No problem!

Working:
z<-ts(sin(1:1000), start=1, freq=10)
x<-StructTS(z, type='BSM')
Hang:
z<-ts(sin(1:1000), start=1, freq=100)
x<-StructTS(z, type='BSM')
(Both on WinXP, R 2.9.2)

I don't, however, know who the package maintainer is, nor can I track down where exactly the hang is caused.

Maybe somebody else can help me with that.
Apologies, thought that code was from a contributed package rather than
from the base distribution of R.
In this case, and given the code, we should try to debug a little bit.

My first impression is that it will take some time given experiments
with smaller numbers it looks roughly exponential in runtime. I'll be
off for a meeting now and will look at results in 1-2 hours (when I
expect them to appear on my machine).

Uwe Ligges
Cheers,
Richard Kiefer

-----Urspr?ngliche Nachricht-----
Von: Uwe Ligges [mailto:ligges at statistik.tu-dortmund.de]
Gesendet: Dienstag, 26. Januar 2010 10:32
An: Kiefer, Richard
Cc: r-help at r-project.org
Betreff: Re: [R] StructTS hang?

On 25.01.2010 11:03, Richard.Kiefer at rwe.com wrote:
Hi!

You have probably sorted out your problem a different way.
However I have tried dealing with the same problem for a while and writing down the reason for this here may be of use to someone.

In some other help-mail I read that time series objects behave unexpectedly in some situations if the frequency is too high. As a result of this, a call of StructTS (with seasonality) hangs if and only if the frequency is larger than 24 (if I remember the number right). This is what I found with my own time series using several frequencies.

So, this may not be helpful using StructTS on a time series, but it may save somebody some time.

Even better: report a reproducible example to the package maintainer who
is certainly happy to fix bugs in the package.

Best wishes,
Uwe Ligges

Regards,
Richard

----------------------------------------------------------------
Irgendwann kommt jeder drauf! WWW.ENERGIEWELT.DE
-----------------------------------------------------------------------------------
Vorsitzender des Aufsichtsrates: Dr. Rolf Martin Schmitz
Vorstand: Dr. Johannes Lambertz (Vorsitzender), 
Prof. Dr. Gerd Jaeger, Antonius Voss, 
Erwin Winkel

Sitz der Gesellschaft: Essen und Koeln
Eingetragen beim Amtsgericht Essen
Handelsregister-Nr. HRB 17420
Eingetragen beim Amtsgericht Koeln
Handelsregister-Nr. HRB 117
Exponential, you say?

Maybe I am just not patient enough.
After being unresponsive for several minutes, which in the larger examples that I work with has probably appeared like a hang, the code I just mentioned has given a result (a convergence error, but still).
So, I must say I can no longer reproduce the hang.

I remember to be able to reproduce a significant change in behaviour upon the frequency passing some threshold. Maybe by chance, this threshold parting "hang" from normal operation agreed with a number someone claimed ts-objects could not handle properly.
But I can't reproduce this now. I'm sorry...
Right, your example takes roughly 30 minutes on my machine.

Uwe

-----Urspr?ngliche Nachricht-----
Von: Uwe Ligges [mailto:ligges at statistik.tu-dortmund.de]
Gesendet: Dienstag, 26. Januar 2010 11:15
An: Kiefer, Richard
Cc: r-help at r-project.org
Betreff: Re: AW: [R] StructTS hang?

On 26.01.2010 10:55, Richard.Kiefer at rwe.com wrote:
No problem!

Working:
z<-ts(sin(1:1000), start=1, freq=10)
x<-StructTS(z, type='BSM')
Hang:
z<-ts(sin(1:1000), start=1, freq=100)
x<-StructTS(z, type='BSM')
(Both on WinXP, R 2.9.2)

I don't, however, know who the package maintainer is, nor can I track down where exactly the hang is caused.

Maybe somebody else can help me with that.

Apologies, thought that code was from a contributed package rather than
from the base distribution of R.
In this case, and given the code, we should try to debug a little bit.

My first impression is that it will take some time given experiments
with smaller numbers it looks roughly exponential in runtime. I'll be
off for a meeting now and will look at results in 1-2 hours (when I
expect them to appear on my machine).

Uwe Ligges

Cheers,
Richard Kiefer

-----Urspr?ngliche Nachricht-----
Von: Uwe Ligges [mailto:ligges at statistik.tu-dortmund.de]
Gesendet: Dienstag, 26. Januar 2010 10:32
An: Kiefer, Richard
Cc: r-help at r-project.org
Betreff: Re: [R] StructTS hang?

On 25.01.2010 11:03, Richard.Kiefer at rwe.com wrote:
Hi!

You have probably sorted out your problem a different way.
However I have tried dealing with the same problem for a while and writing down the reason for this here may be of use to someone.

In some other help-mail I read that time series objects behave unexpectedly in some situations if the frequency is too high. As a result of this, a call of StructTS (with seasonality) hangs if and only if the frequency is larger than 24 (if I remember the number right). This is what I found with my own time series using several frequencies.

So, this may not be helpful using StructTS on a time series, but it may save somebody some time.

Even better: report a reproducible example to the package maintainer who
is certainly happy to fix bugs in the package.

Best wishes,
Uwe Ligges

Regards,
Richard

----------------------------------------------------------------
Irgendwann kommt jeder drauf! WWW.ENERGIEWELT.DE
-----------------------------------------------------------------------------------
Vorsitzender des Aufsichtsrates: Dr. Rolf Martin Schmitz
Vorstand: Dr. Johannes Lambertz (Vorsitzender),
Prof. Dr. Gerd Jaeger, Antonius Voss,
Erwin Winkel

Sitz der Gesellschaft: Essen und Koeln
Eingetragen beim Amtsgericht Essen
Handelsregister-Nr. HRB 17420
Eingetragen beim Amtsgericht Koeln
Handelsregister-Nr. HRB 117