topic Re: How come no Segment Profile after I set "Cluster Variable Role" = Input in Cluster node? in SAS Data Science

How come no Segment Profile after I set "Cluster Variable Role" = Input in Cluster node?

ken88 — Tue, 14 Oct 2014 16:57:50 GMT

How come no Segment Profile after I set "Cluster Variable Role" = Input in Cluster node?

Thank You

Re: How come no Segment Profile after I set "Cluster Variable Role" = Input in Cluster node?

M_Maldonado — Tue, 14 Oct 2014 17:31:13 GMT

Hey Ken,

Not sure exactly what you mean.

If your question is, why would a Segment Profile Node does not show plots after a Cluster node with Cluster Variable Role set to input?

Well, it is just because of that . The Segment Profile Node needs a segment variable to analyze.

To do both things in one run I would leave the cluster variable role as segment on my flow, check it out with the segment profile node, and then use the metadata node to change the role from segment to input.

Your flow will look something like: Data->Cluster->Segment Profile->Metadata(change role of _segment_ variable)->Regression (or anything you had planned for your modeling).

The other way around (using the metadata role of _segment_ to segment) also works.

Would that answer your question?

Thanks,

Miguel

Re: How come no Segment Profile after I set "Cluster Variable Role" = Input in Cluster node?

ken88 — Wed, 15 Oct 2014 13:39:23 GMT

Thanks Miguel,

what the effect on the resulting cluster for diff setting in Cluster Variable Role?

so set Cluster Variable Role= segment == supervised segmentation

and set Cluster Variable Role = input == unsupervised segmentation?

Thank You

Re: How come no Segment Profile after I set "Cluster Variable Role" = Input in Cluster node?

gergely_batho — Wed, 15 Oct 2014 13:45:53 GMT

No effect at all. I affects only the ROLE of the segment variable.

It is unsupervised.

Re: How come no Segment Profile after I set "Cluster Variable Role" = Input in Cluster node?

ken88 — Wed, 15 Oct 2014 15:01:51 GMT

what criteria should I use for selecting segment variable?

is segment variable similar to target variable??

Thank You

Re: How come no Segment Profile after I set "Cluster Variable Role" = Input in Cluster node?

M_Maldonado — Wed, 15 Oct 2014 16:06:13 GMT

Ken,

Like Gergely said, clustering is unsupervised because you don't need a target.

Cluster node creates a variable called _segment_ with a role of segment to flag each of your observations.

You use the property Cluster Variable Role to tell Enterprise Miner what to do with that variable. If you specify "segment", a node like Segment Profile node can use it to give you an anlysis based on segment. If you specify "input" you can use it as one more input variable for any of your models.

To make it more clear, look at the exported data set of the Cluster node. Notice that you have 3 new variables: _segment_, segment_label, distance (to the centroid of that cluster). Cluster variable role is the way you tell EM what to do with _segment_ variable. By default _segment_label and distance will have a role of rejected, _segment_ will have a segment role.

I hope this helps,

Miguel

Re: How come no Segment Profile after I set "Cluster Variable Role" = Input in Cluster node?

ken88 — Thu, 16 Oct 2014 14:06:08 GMT

Is there a way to keep the variable in the cluster train data without using it.

I want to keep the id of the customer in the dataset without using it for data analysis

Re: How come no Segment Profile after I set "Cluster Variable Role" = Input in Cluster node?

M_Maldonado — Thu, 16 Oct 2014 14:26:57 GMT

All three variables that the cluster node creates for you are kept automatically.

The "id" of the cluster (_segment_) is kept, and only used for data analysis if you specify the Cluster Variable Role as input.

Take a look at the exported train data coming out of your cluster node. It will make things more clear.

Good luck!

Re: How come no Segment Profile after I set "Cluster Variable Role" = Input in Cluster node?

RalphAbbey — Wed, 22 Oct 2014 20:56:33 GMT

If you set the id of the customer as an ID Variable, then it should stay with the data set, but not be used by any of the nodes for data analysis.