Time Series Forecasting
in [Matlab]
|
Prev: Calculate the coordinates of the 3rd vertex of a spherical triangle knowing the other 2 vertices, angles and edge lengths
Next: extension g
From: Rogelio on 8 Aug 2010 15:30 "Alex " <alemantz82(a)gmail.com> wrote in message <i3mv6s$miu$1(a)fred.mathworks.com>... > "Rogelio" <rogelio.a.mancisidor(a)student.bi.no> wrote in message <i3mlul$o0m$1(a)fred.mathworks.com>... > > "Alex " <alemantz82(a)gmail.com> wrote in message <i3l6a8$qcs$1(a)fred.mathworks.com>... > > > Greg Heath <heath(a)alumni.brown.edu> wrote in message <10a76362-64aa-4489-966a-f00b2ffc2a27(a)f42g2000yqn.googlegroups.com>... > > > > On Aug 7, 2:23 pm, "Alexandros " <alemant...(a)gmail.com> wrote: > > > > > Hello! > > > > > > > > > > I am new at Matlab, and I need to do Time Series Forecasting. > > > > > More specifically I have daily movie ratings until a date and I need to predict ratings from that day on(I suppose it has to do with time series forecasting, right?) > > > > > Until now I have only imported my data from a MySQL database and created some new matrices with train and test datasets, but I don't know how to proceed from that point and what to use. Any advise would be useful. > > > > > > > > > > Thanks! > > > > > > > > What type of models are you considering? > > > > What toolboxes do you have? > > > > Have you searched the newsgroup archives? > > > > > > > > Greg > > > > > > Thanks for the reply Greg. > > > I have searched the newsgroup archive, but I haven't found something relevant. > > > I was considering using simple models, such as moving average model, exponential smoothing or single regression. In fact, there are only two variables: the time vector and the ratings(data). Another detail about my data set is that the time vector is kind of sparse. For example, there may be a movie for which i have 1000 daily ratings within a 10 years period. > > > So I don't know if this plays a role to the decision about which model to use. Considering this, I am not sure if the models I mentioned are the appropriate ones, so I would appreciate a piece of advice on this. > > > As far as I googled it, I found a toolbox named Captain toolbox that uses regression model. Could you recommend any toolbox that I could use for my purpose (if it can be handled from a GUI, the better) ? > > > Thanks in advance for your time and help. > > > > Do you have any explanatory variable? or you only have rating variables over time? If you only have the ratings over time then you must work with ARIMA models. Are you familiar with such models? otherwise you will need to use some book to choose the model with wich you will work with and how to interprete your results. What kind of version is your matlab? the complete version offers different functions for parameter estimation and forecasting, but first you need to mention which toolboxes you have. > > Thanks for the reply Rogelio. > It's only rating variable(scale 1 to 5) over time, where not all dates are recorded. For example, > > 25/03/2004 5 > 26/03/2004 3.5 > 27/03/2004 4 > 29/03/2004 3 > > In fact, I am not familiar with ARIMA models. The Matlab version I am using is 7.6.0 and the toolboxes are the ones that come with it. I have inserted only Captain toolbox, that I mention to my previous post. Is there an already embedded to Matlab toolbox that I can use in order to work with ARIMA models? If not could you please mention such a toolbox? Thanks! check if you have the next two functions, 'garchfit' and 'garchpred'. The first one estimates the coefficientes in your ARIMA model by MLE (Actually the function is for GARCH modelling, but it also works for ARIMA models if you specify correctly the error variace). The second function will do the forcast for you. You can also apply direct formulae to obtain your forcasts.
From: TheSecondDerivative on 8 Aug 2010 12:05
So you want to use time series forecasting to predict future movie ratings? And you already have some basic data in a MySQL database? I'm not that experienced with MySQL, I have used it a little bit for my [url=http://www.eastcoastairsoft.com]airsoft guns[/url] website but not much beyond that. If you could extract the data and plug it into statistical software, that might work better. |