Multiplicative???

[jzen188]

Hi macro @scubamaggo , I found there is only decomposition providing multiplicative functionality. Chris said you and him talk quite a lot for that... Am I missing anything on github here?

[scubamaggo]

Hi Junjie,

it has been a while since I've done anything with the TS part of iNZight, but if I remember correctly Chris and I only discussed the multiplicative part for decompositon. It does not make sense for all of the plots anyway. Rawplot is independent of multiplicative, forecastplot could change. But here we used the built-in predict() function, which does not support multiplicative predictions as far as I know. So we would have to extend this ourselves to this case, not sure how complicated that would be.

On Sun, Mar 2, 2014 at 9:38 PM, Junjie Zeng notifications@github.comwrote:

Hi macro @scubamaggo https://github.com/scubamaggo , I found there is only decomposition providing multiplicative functionality. Chris said you and him talk quite a lot for that... Am I missing anything on github here?

Reply to this email directly or view it on GitHubhttps://github.com/iNZightVIT/iNZightTS/issues/1 .

[jzen188]

Thank you, macro.... One more question, I notice the algorithm in the function to calculate the trend/season/random value using loess, it is kind of 'stl' way? What purpose for using this kind of calculation compared with using other algorithm?

[scubamaggo]

It uses stl() for timeseries with frequencies greather than 1, since freq=1 is not supported by stl(). For freq=1 loess() is used and it tries to roughly do the same as stl() for the other cases. loess() is used because stl() uses loess as well

On Mon, Mar 3, 2014 at 10:58 PM, Junjie Zeng notifications@github.comwrote:

Thank you, macro.... One more question, I notice the algorithm in the function to calculate the trend/season/random value using loess, it is kind of 'stl' way? What purpose for using this kind of calculation compared with using other algorithm?

Reply to this email directly or view it on GitHubhttps://github.com/iNZightVIT/iNZightTS/issues/1#issuecomment-36565264 .

[jzen188]

Get it.

On 4 March 2014 11:07, scubamaggo notifications@github.com wrote:

It uses stl() for timeseries with frequencies greather than 1, since freq=1 is not supported by stl(). For freq=1 loess() is used and it tries to roughly do the same as stl() for the other cases. loess() is used because stl() uses loess as well

On Mon, Mar 3, 2014 at 10:58 PM, Junjie Zeng <notifications@github.com

wrote:

Thank you, macro.... One more question, I notice the algorithm in the function to calculate the trend/season/random value using loess, it is kind of 'stl' way? What purpose for using this kind of calculation compared with using other algorithm?

Reply to this email directly or view it on GitHub< https://github.com/iNZightVIT/iNZightTS/issues/1#issuecomment-36565264>

.

Reply to this email directly or view it on GitHubhttps://github.com/iNZightVIT/iNZightTS/issues/1#issuecomment-36566344 .

[jzen188]

Hi macro @scubamaggo , when you said,

But here we used the built-in predict() function, which does not support multiplicative predictions as far as I know

Are you meaning HoltWinter() or loess() ? I can see the predict.HoltWinter function, which should support the prediction of HoltWinter() setting argument seasonal = "multiplicative". As loess(), yes, we have no way to do the prediction but it only used in decomposition, and the latter, decomposition() is using in multiseries and seasonalplot, which is not requiring the prediction.

[scubamaggo]

We are using the predict() function with a HoltWiner() object, so we are using predict.holtwiner. If this predict function supports multiplicative TS, that is good news. It is not yet implemented though (multiplicative functionality for forecast). We (Chris and I) only discussed multiplicative TS in the context of decomposition and thus that is the only place where it is implemented.

[jzen188]

Hi macro @scubamaggo , Did you write any animation of the recompose() ?

[scubamaggo]

Nope

Am 12.03.2014 01:59 schrieb "Junjie Zeng" notifications@github.com:

Hi macro @scubamaggo https://github.com/scubamaggo , Did you write any animation of the recompose() ?

Reply to this email directly or view it on GitHubhttps://github.com/iNZightVIT/iNZightTS/issues/1#issuecomment-37365044 .