Data vs fit results

[stefanocovino]

Dear friends,

I apologize for the trivial question.

In this package, what is the best way to compare input data and fit results? For instance, to obtain the residuals shown in the diagnostic plots.

Thanks a lot,
Stefano

[raphaelsaavedra]

I haven't looked at this in a while, but I believe if you have a naive model you can either call model.residuals or call get_standard_residuals(model) which will return the standardized residuals (normalized by variance). If you have a non-standard model or just a filter you can get instead get the innovations (i.e. filter residuals) using get_standard_residuals(filter_output). See

StateSpaceModels.jl/src/kalman_filter_and_smoother.jl

Line 83 in b15f942

get_standard_residuals(fo::FilterOutput) = get_standard_innovations(fo)

[stefanocovino]

Hi Raphael, Many thanks for your rapid reply. I'm using the SARIMA model. However, again my ignorance, I wonder which is the scale of the innovations. I found something close to my expectation reading; get_predictive_state(filter_output)[:,1]. But it was just after a browsing of the available functions. Could you maybe provide an example? It could be of some interest possibly in the example section of the documentation too. Bye, Stefano Il giorno lun 10 feb 2025 alle ore 11:03 Raphael Saavedra < ***@***.***> ha scritto:

…

I haven't looked at this in a while, but I believe if you have a naive model you can either call model.residuals or call get_standard_residuals(model) which will return the standardized residuals (normalized by variance). If you have a non-standard model or just a filter you can get instead get the innovations (i.e. filter residuals) using get_standard_residuals(filter_output). See https://github.com/LAMPSPUC/StateSpaceModels.jl/blob/b15f942b9314861b321435766a2b364bf19d5133/src/kalman_filter_and_smoother.jl#L83 — Reply to this email directly, view it on GitHub <#333 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAGVYHDC3LEFBRBOWZJQTQ32PB2NRAVCNFSM6AAAAABWZ5LL4CVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDMNBXGUYDAMBXGE> . You are receiving this because you authored the thread.Message ID: ***@***.***>

-- ——————————————————————————— Il tempo è l'immagine dell'eternità Platone Stefano Covino INAF / Osservatorio Astronomico di Brera Via Emilio Bianchi 46, 23807 Merate (LC) - Italy Tel.: +39 02 72320475 (office) FAX: +39 02 72320401 Cell: +39 331 6748534 E-mail: ***@***.*** URL: http://www.merate.mi.astro.it/∼covino -----------------------------------------------------------------------------------

[stefanocovino]

I beg your pardon. I've understood how innovation is produced. A real example could help anyway, though. Thanks anyway, Stefano Il giorno lun 10 feb 2025 alle ore 11:34 Covino, Stefano < ***@***.***> ha scritto:

…

Hi Raphael, Many thanks for your rapid reply. I'm using the SARIMA model. However, again my ignorance, I wonder which is the scale of the innovations. I found something close to my expectation reading; get_predictive_state(filter_output)[:,1]. But it was just after a browsing of the available functions. Could you maybe provide an example? It could be of some interest possibly in the example section of the documentation too. Bye, Stefano Il giorno lun 10 feb 2025 alle ore 11:03 Raphael Saavedra < ***@***.***> ha scritto: > I haven't looked at this in a while, but I believe if you have a naive > model you can either call model.residuals or call > get_standard_residuals(model) which will return the standardized > residuals (normalized by variance). If you have a non-standard model or > just a filter you can get instead get the innovations (i.e. filter > residuals) using get_standard_residuals(filter_output). See > https://github.com/LAMPSPUC/StateSpaceModels.jl/blob/b15f942b9314861b321435766a2b364bf19d5133/src/kalman_filter_and_smoother.jl#L83 > > — > Reply to this email directly, view it on GitHub > <#333 (comment)>, > or unsubscribe > <https://github.com/notifications/unsubscribe-auth/AAGVYHDC3LEFBRBOWZJQTQ32PB2NRAVCNFSM6AAAAABWZ5LL4CVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDMNBXGUYDAMBXGE> > . > You are receiving this because you authored the thread.Message ID: > ***@***.***> > -- ——————————————————————————— Il tempo è l'immagine dell'eternità Platone Stefano Covino INAF / Osservatorio Astronomico di Brera Via Emilio Bianchi 46, 23807 Merate (LC) - Italy Tel.: +39 02 72320475 (office) FAX: +39 02 72320401 Cell: +39 331 6748534 E-mail: ***@***.*** URL: http://www.merate.mi.astro.it/∼covino -----------------------------------------------------------------------------------

-- ——————————————————————————— Il tempo è l'immagine dell'eternità Platone Stefano Covino INAF / Osservatorio Astronomico di Brera Via Emilio Bianchi 46, 23807 Merate (LC) - Italy Tel.: +39 02 72320475 (office) FAX: +39 02 72320401 Cell: +39 331 6748534 E-mail: ***@***.*** URL: http://www.merate.mi.astro.it/∼covino -----------------------------------------------------------------------------------

[raphaelsaavedra]

I think what you mean by scale of innovations is that get_standard_residuals (which calls get_standard_innovations) standardizes the innovations? If you want the raw innovations you can call get_innovations on the filter output. I suggest taking a look at the code in StateSpaceModels.jl/src/kalman_filter_and_smoother.jl where these operations are used.