[R] evaluation + Re: hourly prediction time series

[AURORA GONZALEZ VIDAL]

  Thank you, it works fine.

Now, I am trying to evaluate the performance of the model across time. So
as to do that I use rolling window which I understand as sort of a "leave
one out".

The example:

The data are from the 1st of January to nowadays so, I use data from the
1st of January to the 1st of December to fit the model and then I predict
the temperatures of the 2nd of December. As I have the real ones, I can
compute RMSE or other metrics.
Then, I use data from 1st of January to the 2nd of December in order to
predict the 24 values of temperature on the 3rd of December, and later I
compute again the RMSE (between predicted and real of the 3rd).
So on untill I have no more data.
Then, I have several RMSE, I compute their mean and sd and I consider this
as the evaluation of the model's performance.

The question is: do you know any book or documentation where I can cosult
how many times should I do this process so as to know where I should start.
Should I start before December to do the rolling? I mean, is there any
agreement? For example, if I have 400 days of data, meaning 9600 (400 * 24)
observations maybe I could choose a 10 % of the windows so as to start
evaluating, which means, do the process 40 times starting with the day 360.

Any source of information will be appreciated.

Sean Porter <sporter en ori.org.za> escribió:

> Try the auto.arima function in the forecast package..
>
> Regards,
>
> DR SEAN PORTER
> Scientist
>
> South African Association for Marine Biological Research
> Direct Tel: +27 (31) 328 8169   Fax: +27 (31) 328 8188
> E-mail: sporter en ori.org.za Web: www.saambr.org.za[1]
> 1 King Shaka Avenue, Point, Durban 4001 KwaZulu-Natal South Africa
> PO Box 10712, Marine Parade 4056 KwaZulu-Natal South Africa
>
> -----Original Message-----
> From: R-help [mailto:r-help-bounces en r-project.org] On Behalf Of AURORA
> GONZALEZ VIDAL
> Sent: 05 February 2016 10:50 AM
> To: r-help en r-project.org
> Subject: [R] hourly prediction time series
>
> Dear R users,
>
> I am fronting my firts time series problem. I have hourly temperature
> data for 3 years (from 01/01/2013 to 5/02/2016). I would like to use
> those in order to PREDICT TEMPERATURE OF THE NEXT HOURS according to the
> observations.
>
> A subset of the data look like this:
>
> date <- rep(seq(as.Date("14-01-01"), as.Date("14-01-03"), by="days"),
> 24) hour <-rep(c(paste0("0",0:9,":00:00"), paste0(10:23,":00:00")),3)
> temperature <- c(6.1, 6.8, 6.5, 7.2, 7.1, 7.9, 5.9, 6.8, 7.7, 9.5, 12.6,
>                 14.0, 15.9, 17.3, 17.5, 17.2, 15.0, 14.1, 13.1,
11.7,
> 10.9,
>                 11.0, 11.6, 11.0, 11.2, 11.0, 11.0, 11.4, 12.2,
13.7,
> 12.9,
>                 12.9, 12.8, 13.4, 13.9, 14.9, 16.6, 16.0, 15.2,
15.4,
> 14.7,
>                 14.6, 13.3, 13.0, 13.8, 13.1, 12.0, 11.9, 11.8,
11.6,
> 11.0,
>                 11.2, 11.6, 10.6, 9.5, 9.8, 9.9, 11.7, 15.3,
18.6, 20.7,
>                 22.2, 22.2, 20.8, 20.2, 18.3, 15.6, 13.6, 12.8,
13.1,
> 13.7, 14.7)
>
> dfExample <- data.frame(date, hour, temperature)
>
> So as to plot 3 years ( from 01/01/2013 to 31/12/2015) I use this code
> and obtained the attached picture. It is observed seasonality.
>
> tempdf4 <- ts(df4$temperature, frequency=365*24*3)
> plot.ts(tempdf4)
>
> Am I doing it well? Could you help me with any information in this type
> of problem (mainly with the prediction). For example, if I want to use
> Arima, according with my data structure, what are the arguments of the
> funcion??
>
> fit=Arima(df4$temperature, seasonal=list(order=c(xxx,xxx,xxx),period=xxx)
> plot(forecast(fit))
>
> I could use also some predictions from other source that I am collecting
> since January, 2016. But I would prefer to understand the simplest way
> to solve the problem and then, progressively, understand more complex
> approaches.
>
> Thank you very much for any kind of help.
>
> ------
> Aurora González Vidal
> Phd student in Data Analytics for Energy Efficiency
>
> Faculty of Computer Sciences
> University of Murcia
>
> @. aurora.gonzalez2 en um.es
> T. 868 88 7866www.um.es/ae[2]



Vínculos:
---------
[1] http://www.saambr.org.za
[2] http://7866www.um.es/ae


------
Aurora González Vidal
Phd student in Data Analytics for Energy Efficiency

Faculty of Computer Sciences
University of Murcia

@. aurora.gonzalez2 en um.es
T. 868 88 7866
www.um.es/ae

	[[alternative HTML version deleted]]