TM1 & data mining algorithims

[tjmcols]

Technology Evaluation Centers Inc listed predictive modeling as a function for IBM Cognos TM1 but I have found no evidence of any true data mining functionality. Can anyone confirm that it is up to the users to create their own predictive models calculations.

[Alan Kirk]

Technology Evaluation Centers Inc listed predictive modeling as a function for IBM Cognos TM1 but I have found no evidence of any true data mining functionality. Can anyone confirm that it is up to the users to create their own predictive models calculations.

You're talking about two completely different roles; in a slightly oversimplified nutshell, data mining is finding relevant information from vast volumes of data. Predictive modelling is doing "what if"s with a whole bunch of differently weighted "if"s.

TM1 can do both, but yes, it's up to the users, or more typically the administrators on behalf of the users, to have the imagination to develop a structure within which it's done.

For data mining, TM1 doesn't come with any built-in models. The user / administrator needs to define which data characteristics are relevant. (Time, volume, etcetera.) They define those characteristics as dimensions, and the data points as elements within the dimensions. Then they group a bunch of dimensions together into a cube, load the data into that cube, and the users use slicing and dicing to try to mine the relevant information.

For predictive models, again TM1 doesn't come with any "one size fits all" models. I'm sure that some products do, and I'm equally sure that examining chicken entrails would yield equally reliable results. Instead it's up to the users / administrators to determine what valid inputs would be, and what formulas can be applied to those inputs to yield the outputs. TM1 has a fairly rich calculation "language" (called Rules) which will allow you to do this, but yes you do need to come up with the formulas yourself.

(There are quite a few TM1 consultants who may be able to help you with that, since they'll probably have come across such models previously.)

If you're doing all of this by yourself, modelling could just as easily be done in Excel provided that the inputs and formulas aren't overwhelming. If a group of people is doing the modelling, TM1 can help because it is above all else a collaborative tool. Many people can do the inputs, all of the people can see the outputs.

For data mining, it can of course handle vastly more data than many tools and process it faster since it does everything in memory rather than having to read from disk continually.

You may want to contact a Cognos account manager and see whether you can arrange for a demo copy, and play around with it for a while.

[Mike L]

To elaborate on data mining a little, note that the term is sometimes used with a more specific meaning these days. TM1 is well suited to the kind of user-directed data mining that Alan describes, and this has long been one of its main attractions. The term (particularly "data mining algorithm") also refers to automated methods for seeking statistical relationships in the data given relatively little guidance as to what to look for. TM1 does not directly provide this functionality, but it could certainly be used to good effect in implementing it.

As for predictive modeling, TM1 is widely used for this purpose. But as Alan notes, it is not a turnkey solution. Frankly, I would be leery on anything that claimed to be, unless it was specifically tailored to a line of business and scale of operations that was a very good fit. (Some value-added-resellers of TM1 provide just such tailored solutions.)

[Marcus Scherer]

Take a look at
http://www.bissantz.com/en/
Their product DeltaMaster is the high end data mining tool. You can put in top of all common OLAP cubes, TM1 as well.
May fit your needs.