Enterprise Miner Interactive Grouping Node resets itself...

[awright]

When I use the interactive grouping node in Enterprise miner, it occasionally will re-run itself.  The problem is that I manually go in a reject variables, and those all pop back into the model.  Any help?  I have also tried Freezing the groupings, but that didn't help.

Thanks in advance.

[M_Maldonado]

Hi AWright,

It sounds like you are doing things right...

Give it one more try and let me know if I missed any step.

1. You drag an IG node and run it. Then you open the Interactive Grouping.

2. To manually reject variables, you select Rejected from the Variable Role Box. Then you click Apply.

3. You close and say Yes when prompted to save your groupings.

4. You change the property Use Frozen Groups to Yes.

That's the way to do it...

Once you are done doing your groupings and assigning variable roles, open the results. For the Output Variables, are all the "New Roles" what you would expect?

Thanks,

Miguel

[awright]

Hey Miguel,

It seems to be random.  I go through and reject variables, and then re-run the attached scorecard node to see what new variables are allowed in.  It's a process that sometimes takes many times.  It will be working fine, and then at random times I will run the scorecard node, and it will reset everything in the IG node.  It is frustrating because it can take hours to get through the process of vetting out variables until I get a model that I like....

[loredana]

I might have a solution for you, but it's not pretty...

Instead of freezing your groupings try saving them using this configuration in IGN (Interactive Grouping Node):

Use Frozen Groupings: Yes

Gini cutoff / Information Value Cutoff: 0 (this will allow to create groupings for all variables and then you can select manually in Interactive Grouping which one you reject or accept as input - VERY IMPORTANT TO DO THIS!)

Create Grouping Data :Yes

Create Method: Overwrite

Run the node. The groupings will be exported in the following table: EMSW<n>.IGN<m>_EXPORTGROUP (eg: EMSW1.IGN_EXPORTGROUP). Make sure you copy it in your permanent library.

(You can find <n> and <m> like this: attach a Code node to IGN, go to Code Editor, Macro Variables tab, EM_LIB contains your EMSW<n> and EM_METASOURCE_NODEID contains your IGN<m> or you can see it in IGN, Node ID.) You can delete the Code node afterwards.

Attach a Metadata node to IGN and change the New Role column to match exactly the Role column. This will create a template for your dataset, keeping the roles and levels of the variables as they are.

Going back to the same IGN, change these options:

Use Frozen Groupings: No

Import Grouping Data&colon; Yes

Import Data Set: <library>.IGN<m>_EXPORTGROUP

Create Grouping Data&colon; No

Gini cutoff / Information Value Cutoff: 0

Run it. This will create a dataset keeping your groupings but not your metadata (role, level of a var). Run the Metadata node again as it is and ... voilà. You can then use the groupings table and Metadata node on any dataset with identical structure.

[DougWielenga · Accepted solution]

There can be several reasons why a node is caused to rerun, and note that it sometimes only appears to rerun.  Anytime you run a flow, SAS Enterprise Miner will check the preceding nodes to see if they need to be rerun.  Even if they do not, the node might show a spinning green circle while it is being checked.  You can only know if a node has actually rerun if you check the timestamp next to Last Run Time in the node property sheet.   

You have the ability to Create Grouping Data and can then later can Import Grouping Data by specifying the corresponding properties in the property sheet.   Once the node has been run, you can also freeze the groupings.   If the problem is intermittent, make sure you aren't either running into disk space issues or java memory or getting disconnected from your data source.   SAS Enterprise Miner relies on views that require access to the data so losing the connection to the database keeps it from functioning as expected.