commons-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Gilles <>
Subject Re: [math] Usage of Marquardt Optimizer for an Equation
Date Sun, 13 Sep 2015 11:34:08 GMT
On Sun, 13 Sep 2015 12:45:55 +0200, Thom Brown wrote:
> Sorry, for double posting, but maybe the solution is not sooo bad 
> after
> all.

Thanks for keeping me informed about your progress!

> My goal is not a forecast itself but rather smoothing the values.
> Thus, I tried alpha = 0, beta = 0 and gamma = 0 as initial
> parameterization. The result was alpha = 0.0015, beta = 0.5812, gamma
> = ‚Äč0.6624. Despite the parameters are far off the solution posted on 
> the
> page I referenced earlier (they use alpha = 0.7556, beta = 0, gamma =
> 0.9837), the result itself is acceptable. So I can definitely use the
> optimizer. There is probably more than just one acceptable solution.

Did you try using the reference results as a starting point?
If it's a local minimum, the optimizer should not move from there.
If it does, it would be worrying; assuming that the reference values 
correct, it could indicate a bug either in the code, or in the data.

> Indeed, I don't care much about parameterization as long as the 
> results
> work for me. I tried another run with using an extreme seasonality, 
> e.g., I
> peaked extremely every 3rd quarter, and the results were still good 
> with
> what the program gave me. After all, thanks a lot, Gilles, you put a 
> lot of
> effort into helping me and I would like to thank you very much. I had 
> no
> idea how to start off and now I got a better understanding. :)

You are welcome.

At some point, if you are confident that it works as expected, you 
want to propose the triple exponential smoothing use-case as a new 
of the user-guide...

Best regards,

> Greetings,
> Thom

To unsubscribe, e-mail:
For additional commands, e-mail:

View raw message