I would prefer to keep the processing within TM1 just for the sake of streamlining the process (otherwise the user has to export the data, process it and import it back into TM1). I was thinking of using slicing and VBA to perform the calculations but I don't have a maths major to put it to usable code. Has anyone implemented anything similar within TM1?
I'm curious to see what other people have to say, but I've been leery about doing anything more complicated than a rolling average when it comes to trending in TM1. For forecasted sales volume, we do interface with Logility Demand Planning, using TM1 to gather some of the inputs and manual forecast overrides, then integrating the output into our revenue models. The GMDH, though, is all done in Logility.
Given that you have to build things from scratch this sort of modelling is more difficult to do than it might first appear given TM1 is a very powerful calculation engine, when playing to its strengths. There's a thread here with a link to a linear regression model that you can download and play with. If you look you'll see that each component of the calculation gets its own measure and then the measures get combined to generate the output - kind of building up the calculation from first principles. That's just simple linear regression - looks like GMDH would be a much bigger exercise in the practical terms of the required TM1 development. Although it might be an interest academic exercise, I think that in a commercial context you'd be better of investing your hours in developing an good integration solution with one of the more specialised tools that have already done the number crunching for you.
Thanks for the comment Robin. A simple linear regression looks complicated to implement in TM1 let alone GMHD. Maybe we will leave trend prediction to the specialist tools in this case.
SPSS is probably the right tool for this Gareth but it may be a tad expensive compared to your other options.
It's a full blown statistic analysis and modeling tool.
IBM are talking about integrating it more into their TM1/BI products in future releases but any details of that would probably require you to talk to IBM under NDA (I don't know myself and not sure if many people outside the dev team would) http://www-01.ibm.com/software/analytics/spss/
1) Download and Install R.
2) Install all the packages that are relevant for forecasting. I suspect the R package forecast and TTR would be enough.
3) Spit the Source Data out from TM1 using a TI
4) Call R package via a command line from TI and output the forecast data into a output file.
5) Load the Data back into TM1.
The original plan was to use an external system but I wanted to see if it was possible to do the processing within TM1. SPSS would be nice but I don't think we can justify the investment at the moment. I will report back to my colleague with those options though, thanks!
