Time series delay

[Mag]

Hi

Is there a way to force Eureqa to skip lags? When predicting time
series it will always model the most recent data point:

Solution: a = sma(a, 1)
Solution: a = delay(a, 0)

where a is the target to predict: a = f(a,b,c,d,e,f) and input
variable is unchecked.

I know that it's possible to use f(delay(a, 1),b,c,d,e,f)

but that is quite limiting in rich data.

I'd hoped that: delay(a, -1) = f(a,b,c,d,e,f)

might work but the -1 jumps back to +1 during the run.

Or Is there a way to stop the target row index from becoming the
solution?

Thanks

[L]

You might do something like (delay (a,2)-delay(a,1)) - not sure if that's what you're asking?

 

---

<sig>&SIGFILE; NOT FOUND ERR 000121</sig>

---

---

From: Mag <m...@isoftdev.com>
To: Eureqa Group <eureqa...@googlegroups.com>
Sent: Monday, February 6, 2012 3:50 PM
Subject: Eureqa - Time series delay

--
Eureqa Formulize ( http://www.nutonian.com )
-------------------------------------------------
Unsubscribe: eureqa-group+unsub...@googlegroups.com
View Group: http://groups.google.com/group/eureqa-group

[Dave Nunez]

Take a look at:

http://creativemachines.cornell.edu/eureqa_ops

The description for the time delay op is there, also try to model 2
or 3 steps ahead.

hth, -d

> --
> Eureqa Formulize ( http://www.nutonian.com )
> -------------------------------------------------

> Unsubscribe: eureqa-group...@googlegroups.com
> View Group: http://groups.google.com/group/eureqa-group

[bill_080]

Assuming variables "a,b,c,d,e", and "f" have 1000 rows, why not create
a new variable "y" that is a copy of variable "a", but is advanced by
1 row.

The target is: y = f(a,b,c,d,e,f)

That will leave 999 rows for the fit (since "y" is advanced 1 row, the
last row will not have a "y" value).

> > View Group:http://groups.google.com/group/eureqa-group- Hide quoted text -
>
> - Show quoted text -

[Mag]

Thanks everyone.

I can see that delay(a, 1) on all the variables is equivalent of
stepping back one row. With sma I'm worried about peeking at future
data. Therefore I'm trying this at the moment:

a = f(delay(a, 1), delay(b, 1), delay(c, 1), delay(d, 1), delay(e,
1), delay(f, 1), delay(g, 1),
sma(delay(a, 2), 1), sma(delay(b, 2), 1), sma(delay(c, 2),
1), sma(delay(d, 2), 1), sma(delay(e, 2), 1), sma(delay(f, 2), 1),
sma(delay(g, 2), 1))

What do you think? I'm getting some pretty deep nesting.

Bill, thanks good idea. I've done that in the past when I've played
around with neural networks.

> > > View Group:http://groups.google.com/group/eureqa-group-Hide quoted text -

[Michael Schmidt]

I would definitely create a new variable with the "a" rows shifted by one like Bill suggested.  Modeling the delta (or derivative) instead, like L suggested, is also a cool idea.

Michael

--
Eureqa Formulize ( http://www.nutonian.com )