Overlapping of clusters in InfiniteInsight Clustering model.

How to eliminate the overlapping of clusters in SAP InfiniteInsight clustering model? What may be the reasons for overlapping. Since, conceptually overlapping should never happen in clustering model, so I am unable to find the reason. Please help with appropriate soluions.

Hi,
I think there are 2 topics in this discussion.
But first, I wanted to mention that there is no cluster overlap. Each record belong to one and only one segment (cluster).
Ashwanth, the screenshot you sent is not showing an "overlap" betwen clusters.
The bubble chart is showing the different clusters based on 2 continuous variables (X and Y) and the bubble width is by default the cluster frequency. A very large cluster will have a large frequency and thus a large diameter. On the graph, it can seem to "overlap" with other clusters but it is not the case.
You can change how the bubble width is represented (with target frequency) and you will see your graph changing and the will probably be no "overlap". (see screenshot below)
You can also click on X and Y buttons to change the variables along wich clusters are displayed.
In order to have nice presentations, you should select X and Y variables that enable a good disposition of clusters (meaning no overlapping, else it's harder to read).
What Doug mentioned refers to what is displayed in the summary report (Model Overview).
When the option "Calculate SQL Expressions" is selected (default setting), a cluster simplification is performed. During the initial steps, there are some clusters overlapping (1 record assigned artificially to more than 1 cluster)
Note, that this happens only during a temporary step of this simplification process. In the end, in the final model, all records are assigned to 1 cluster only.
Another side-effect of this option is that some records, which can be seen as outliers will be excluded from their initial cluster and will be called "Unassigned records". In your final model, you will still have records that belong to no cluster.
If you uncheck the option Calculate SQL Expressions, no overlap and no unassigned records will appear in the model overview.
However, the clusters will be more complex to interpret, and the "In database application" mode will not be available.
Armelle

Similar Messages

Clustering model ..target group creation ...error

Hello All
After i created a Clustered model successfully i went ahead to create a CRM target group and as the last step in the process I selected the Training source of my Clustering Model ,right clicked it and clicked Monitor
The result shows that the CRM target group creation was incorrect.The following screeen shows the error screen.
http://www.geocities.com/aditya_ponnam/errorscreen.JPG
Please advise.
Thank you
Aditya

Hi naveen
Thank you.How can I check the RFC setting for my CRM system?Has it got anything to do with this problem?I have just started learning SAP BW and CRM.Please suggest.As a matter of fact, this error occured in the BW system.The version of BW system is 3.1 Support Package 6.I could not figure out anything from the error description.Please look at this error screensot.
http://www.geocities.com/aditya_ponnam/errorscreen.JPG
Thank you
Regards
Aditya

Problems in using a calculated key figure in the Clustering Model

We have a problem in using a calculated key figure in the clustering model. With the limitation of the standard, in the Field tab, each calculated key figure cannot be selected from the selection screen directly. According to the SAP help, we have to use a dummy infoobject for a calculated key figure.
Are there any other ways to use a calculated key figure without using a dummy infoojbect?

Hi,
Make use of aggragation property of the keyfigure, for summing the values of that KF.
Regards,
Rajkandula

Change number of clusters

Hi
I want to change the default number of clusters (10) while creating a clustering model in pl/sql.
I have created the settings table and inserted the values (dbms_data_mining.clus_num_clusters, TO_CHAR(6))......still i'm getting result in 10 clusters.
how to override the default 10 clusters value?
thanks,
swathi.

Hi Swathi:
In reply to your question, I tried the code you posted at the Oracle Data Mining and Analytics blog using a 11g database and it works. I had to make a small change to make it run:
CREATE OR REPLACE procedure cluster_build(table_name varchar2,case_id varchar2) as
begin
execute immediate 'drop table cluster_settings ';
execute immediate 'CREATE TABLE cluster_settings (
setting_name VARCHAR2(30),
setting_value VARCHAR2(30))';
insert into cluster_settings values (DBMS_DATA_MINING.clus_num_clusters,'4');
DBMS_DATA_MINING.DROP_MODEL('km_sample');
DBMS_DATA_MINING.CREATE_MODEL(
model_name => 'KM_SAMPLE',
mining_function => dbms_data_mining.clustering,
data_table_name => table_name,
case_id_column_name => case_id,
settings_table_name => 'cluster_settings');
end;
But when I run the query:
SELECT setting_name, setting_value
FROM TABLE(DBMS_DATA_MINING.GET_MODEL_SETTINGS('km_sample'))
ORDER BY setting_name;
I get 4 for the number of clusters.
-Marcos

Sequence Clustering in Excel DM Add-In 2013, Cannot Set Sequence Key

Hi Everyone,
I was attempting to run a Sequence Clustering DM in Excel and I tried to get around the need for a nested table by flattening my normalized Sales Order and Sales Order Line Item into a single table through a query.
However, when I go to create the Mining Structure in Excel, there is no option for setting the Key Sequence. If I go try to add a Mining Model to a mining Structure, Sequence Clustering is an option but I cannot proceed unless I set the Key Sequence when
I am building the Mining Structure.
Does MS Excel DM support Sequence Clustering analysis? Or is there a workaround within Excel?
Thanks,
Billy

You cannot create Sequence Clustering models in DM Add-ins.
Tatyana Yakushev [PredixionSoftware.com]
Download Predixion Insight 3.0 - World class predictive platform for big data

Problem when clustering with O Cluster

Dear all ,
i have the following error when trying to build a clustering model on my data. I used O cluster with the default settings. The strange thing is i have have created a build before for the same data and on the same machine and it worked fine. what could be wrong ? The error description says a memory problem but my settings have not changed since last time.
here is the complete error desciption
Server task state: error
Server task detail:
ORA-40101: Data Mining System Error ORA-40101: Data Mining System Error ORA-40101: Data Mining System Error ODM_OC_CLUSTERING_MODEL-BUILD_OC.build_ocluster--20010
ORA-06512: at "SYS.DBMS_SYS_ERROR", line 105
ORA-06512: at "DMSYS.ODM_OC_CLUSTERING_MODEL", line 122
ORA-06512: at "DMSYS.ODM_OC_CLUSTERING_MODEL", line 2408
ORA-40101: Data Mining System Error ODM_OC_CLUSTERING_MODEL-BUILD_OC.ocluster--20010
ORA-06512: at "SYS.DBMS_SYS_ERROR", line 105
ORA-06512: at "DMSYS.ODM_OC_CLUSTERING_MODEL", line 122
ORA-06512: at "DMSYS.ODM_OC_CLUSTERING_MODEL", line 2312
ORA-06500: PL/SQL: storage error
ORA-04030: out of process memory when trying to allocate 16396 bytes (koh-kghu call ,pmucalm coll)
ORA-06512: at "SYS.DBMS_SYS_ERROR", line 105
ORA-06512: at "DMSYS.ODM_OC_CLUSTERING_MODEL", line 122
ORA-06512: at "DMSYS.ODM_OC_CLUSTERING_MODEL", line 2408
ORA-06512: at "SYS.DBMS_SYS_ERROR", line 105
ORA-06512: at "DMSYS.ODM_OC_CLUSTERING_MODEL", line 122
ORA-06512: at "DMSYS.ODM_OC_CLUSTERING_MODEL", line 2312
ORA-06512: at "DMSYS.DBMS_DATA_MINING", line 30--
Many thanks
Lamia Abo zaid

hi ,
In this scenario u can use the load balancer instead of fail over clustering .
I would suggest u to create apache proxy for redirect the request for many jboss instance.
Rgds
kathir

Infinite Insight Tutorials (not Regression Model)

Hi,
I have seen the official SAP Tutorial for creating a Regression model in Infinite Insight, does anybody know if there is anything similar for any ofthe other modules, Explorer, Clustering Model, Association Rules, Social or Recommendation?
I can't find any, but presume there must be some out these somewhere,
Thanks
Howard Stiles

Hi,
If you had download the zip file, you also have some pdf included (51047379.zip\DATA_UNITS\SAP_INFINITESIGHT_DOCUMENT\Help\PDF\US).
All the datasets are avaiables in the folder "Samples" in your install directory : C:\Program Files\SAP InfiniteInsight\WorkstationV6.5.5\Samples. (KAR for association Rule, KSN for social network, etc...)
You can also find some exercises on the workbook SAP Advanced Analytics on the sap partneredge website, in particular data manipulation with the creation of analytical datasets. (SAP PartnerEdge).
Hope it helps,
Edouard

Cannot Display Graphic in Viewing Model Results

In our project, we use a clustering model. After having trained the model, we try to view model result in graphic such as Influence Chart. Unfortunately, this problem occurs.
<u>Note.</u>
when clicking 'Inter and Intra Cluster Distances Information,' we find the error message 'Error displaying additional information about clusters.'
<u>Details of this message</u>
Diagnosis
Error occurred while generating graphs for displaying the additional clustering information.
Procedure
Ensure that the RFC destination IGS_RFC_DEST is configured to 6.2 or above version for the Internet Graphics Server (IGS).

If you have reimported the model or removed/added the controller to view context, the binding would have broken. Make sure you see the outward arrow visible on the model attributes of view context exist, otherwise rebind the context to component controller or custom controller structures.
Another tips is
You can try to display the size of the BAPI return output object, if it's 0 then there is no data returned. I mean in the component controller you would have written retrieve method to fetch data from backend
wdContext.node<your BAPI>_Input().
current<your BAPI>_InputElement().
modelObject().execute();
After executing above code, check your output object return code or message lines or a structure object size information. If nothing is returned then there is a problem, sometimes return code and message lines will return something if your backend BAPI is executed.
Or as above posted suggested put external breakpoints in your BAPI and make sure the input parameters are correct.
Hope this helps
Praveen
Edited by: Praveen11 on Oct 21, 2009 8:58 AM
Edited by: Praveen11 on Oct 21, 2009 9:06 AM

Unable to test model-displayName argument

I am using DM4J to create a clustering [o-cluster] model. The model is created successfully but when I attempt to execute a model test, it fails immediately with the following error.
Argument "displayName" is null. Supply a non-null value.
I am running JDeveloper 9.0.3.1 with today's DM4J applied.
I have patched my database with patch #26329311
and have applied ODM patches #2498603 and #2647739.
I successfully created the tutorial clustering model and executed a test for it.
Any help would be greatly appreciated,
Dianna

I am attempting to build a model against my own data, 8 features a mixture of numerical and categorical. When I built the model, I used default OCluster settings and I omitted my SHIPID for the training set [a unique identifier]. When I created a Clustering Apply component, I again used default settings but then I selected my SHIPID and the item that I wanted to uniquely identify my data. I trained on 10,000 rows and attempted to validate different size datasets from 400,000 to 50,000 [I reduced the size when I started having problems.] The data [both training and test] contains some missing values.
The error "displayName" argument not specified made me think that it was something in my configuration setup but since I am able to run the demo OCluster Model Build/Apply, it leads me to believe that it is data related.
Thanks,
Dianna

Can't get remote cluster machine to encode video - getting 'Media Server application unexpectedly quit'

Hi
Using Compressor 4 / Apple QuarterMaster admin etc, I have set up a Cluster and it works well to Share > Export Using Compressor Settings. That took some doing, and I'm happy that it works.
If I do all my processing in the foreground, or alternatively in the background on my FCPX computer, ie without running or using any cluster in the background, it all goes well. Quick, error free.
If I activate the cluster and only put my local machine's compressor services into it, and then send my compressor batch to the cluster, it works perfectly well. But that defeats the point of having a cluster. I want the remote machine to do all the work so my local machine doesn't slow down.
Unfortunately, if I add the remote machine's services into the cluster, the encoding always fails with 'Media Server application unexpectedly quit' in the error log. I know that the cluster is distributing segments of the file to the remote machine - this can be seen in the Share Monitor, but they never get processed and sit there waiting until I get the error message.
The remote machine is a modern 4G core 2 duo MacBook. It's never been used for this before. It has Compressor 4 and the Pro Apps Update installed and both appear to work. Both machines are running 10.6.8. The remote machine has Compressor services initialised in Apple QMaster sharing. The QMasterAdmin sees these services. They are accessible over Bonjour and appear in the Cluster. Compressor services from both the local and the remote machine are configured exactly the same way. The same QT codecs exist on both machines. I've restarted, shift-restarted, etc etc.
So I am at a loss here. The remote machine just won't compress anything it is sent.
Any ideas? Must both have FCPX installed?
Anyone actually got a remote machine in a cluster to work with Compressor 4? I can't figure it out.
Chris.

OK... finally sorted it out.
It's a bug, as far as I can tell.
Any job entered directly into Compressor 4 in the normal Compressor manner will be successfully rendered by any working cluster from any machine that can access the cluster. That's good. It means that the underlying distributed processing model works well.
HOWEVER - any job forwarded to a Compressor 4 cluster that includes non-local compressor services (ie compressor services not resident on the same machine), using Share > Export Using Compressor Settings direct from FCPX will fail.
To confirm this bug, I made a cluster on a remote machine. It was a dual core machine, so I enabled 2 compressor services on that machine, and that's all the cluster was. Simple.
I then manually entered a video file (ProRess 422 720p) into Compressor on the remote machine. I did this by physically setting up new job using the compressor user interface. A bog standard ProRes422 720p file rendered fine this way on the remote machine. As did anything else I gave it. Good.
Then on my main machine, I opened Compressor and made a job based on the same file and settings, and sent it to the remote machine's cluster. No problem at all! Great!
So now I know that both Compressor versions, and the clustering model, are working fine. In fact I can send all sorts of files to the cluster, from any other version of compressor, and have them processed on the remote machine, and get the result back on my desktop later on. Excellent.
But, if I try put this same file into a FMPX timeline, and go Share > Export Using Compressor Settings.... and select a cluster with remote (non-local) compressor services, it does not work. Rendering the video segments on the remote machine times out and fails, every time. It doesn't matter what file format I use, it just fails.
So it's a bug.
From what I can tell, Final Cut Pro X somehow messes up the Share > Export Using Compressor settings where the cluster includes non-local compressor services, causing all jobs to fail. The same Share > Export Using Compressor Settings will work quite happily if all the services on the cluster are on the same machine as FCPX, or if the job is sent to This Computer in the background. But any attempt to send files to a cluster using any remote services will fail.
I hope this saves some people from wasting as much time as I have!
One workaround is to export to ProRes then put this file manually into Compressor, sending the job to the remote cluster. This is a two step process with a large intermediary ProRes file. If one goes Export as QuickTime movie, generating the intermediary file prevents further FCPX work being done.
To get the intermediary in the background one could use Share > Export Using Compressor Settings via either This Computer, or to a cluster using only local services. Then once complete, manually add it to a compressor job.
So this is a FCPX bug as far as I can tell.
Chris.

CRM target griup creation error

Hello All
After i created a Clustered model successfully i went ahead to create a CRM target group and as the last step in the process I selected the Training source of my Clustering Model ,right clicked it and clicked Monitor
The result shows that the CRM target group creation was incorrect.The following screeen shows the error screen.
http://www.geocities.com/aditya_ponnam/errorscreen.JPG
Please advise.
Thank you
Aditya

Thnak you ashwin..
These were my steps
1.Created a clustering model
2.checked for inconsistencies and activated both the training model and training source
3.Trained source online
4.Viewed training source results-Value Distribution,Attribute data and Attribute charts and went ahead to create a target gruop
5.Created target group
6.Selected "immediate" and saved .
7.Went back and selected my training source and rightclicked to monitor it
8.The results were as shown in the figure below
http://www.geocities.com/aditya_ponnam/errorscreen.JPG
It flags that as Incorrect target group creation.Please see the figure.
What could be the problem here?
Thank you
Aditya

Need the SSAS Prediction Date

I have an SSAS BI project using SQL Server 2012 and Visual Studio 2012 (using the Sequence clustering algorithm) that is giving me some great predictions. I have several years of system failure data and it is predicting WHAT the next system failure is
predicted to be along with its probability. This is great, so far.
The failure data I have contains the: Date of failure, type of failure, region, and other info...
But I also need the WHEN! When is the predicted failure expected to occur? In other words, what is the predicted failure date? Which, of course, would also be a prediction.
I don't see a way to do this using the same Sequence Clustering model. I also looked at the Time Series algorithm, but that doesn't seem to fit either. Does anyone know how to do this in SSAS?

Hi Montana,
According to your description, it seems that you want to predicted failure date base on the current data, right? In this case, I suggest you refer to Chris Webb's blog about using AS Data Mining to ddd Forecast Values to a Cube, please see:
http://cwebbbi.wordpress.com/2008/10/01/using-as-data-mining-to-add-forecast-values-to-a-cube/
Here is a good article about understanding Time Series Forecasting Concepts for your reference:
http://sqlmag.com/sql-server-analysis-services/understanding-time-series-forecasting-concepts
Regards,
Charlie Liao
TechNet Community Support

Some question on CORE BPEL ENGINE

Hello,
I've read Chapter 4 and now I've some question:
1-Clustering: What's you intend which expression "fail-over can be configured on the engine" and How? What Kind of clustering you realize (active-active, active-passive,...)?
2-I think that I don't exactly understand what you intend with "BPEL Compensating Transaction Model" could you give me more information about this?
thank you guys!
Andrea

1- I just completed a draft of a tech note on High Availability that I can share with you that explains the clustering model in detail. At a very high level, the BPEL Server is completely stateless and reliable, including a built-in recovery service that enables failover in case of failures. You don't need to cluster the app-server to be able to cluster BPEL servers (for either high availability or scalability).
2 - BPEL provides support for compensating transactions, which is an alternative transaction model when XA-style, ACID transactions can not be used (either due to the long-running nature of a "transaction" or the inclusion of services which don't support XA/JTA style transactions. We have code examples available for this and slides from our training course that discuss it.
Send me an email if you would like access to either of these documents. I'll also try to publish them to OTN in the next few days as tech notes, but in the meantime, I'm happy to share the information directly with any individuals who want it.
Dave
David Shaffer
Product Mgr, Oracle BPEL Process Manager
[email protected]
W: 650.506.1729
http://otn.oracle.com/bpel

Protocols supported by SunOne

HI,
We are planning to built a clustering model with SunOne and use two application servers. These two server will be may be on a different network and only few protocol stack are allowed for communication between these. Is there any list of the supported protocols? Can SunOne communication with the other server in cluster using someother protocol other than tcp/ip?
Thanks,
Srivatsan.

Sir,
<suresh kumar> wrote in message news:[email protected]..
How can i connect to weblogic other than t3 protocol?
Does weblogic supports any other J2EE standard protocols??
I am using rmi protocol in my code to connect to weblogic. It is failing.It is not very likely that it is failing because of t3. It happens to be a pretty
stable thing. May be it is the way you are cooking it.
Do you have any stack traces or error messages that you could kindly share
with us?
Regards,
Slava Imeshev

APD Flatfile

Hello,
im would like to export some Query Data to a flat file, using APD.
I have read about a flatfile data target in APD. But i can not find it in the workbench.
The available data target types are:
- Characteristics / Attributes
- Transactional ODS
- CRM
- Decision Tree
- Clustering Model
- Association Model
- Scoring Model
- 3rd Party Model
I also could export the Data via an ABAP routine, however the APD process needs to have a Data Target.
Is there a way to create an ADP Process without a data target, or will i have to use a dummy target?
What is a good dummy target. Is there a target that requires no Objects, no space and only minor processing resources when executed?

The Flat file target option is relatively new, Netweaver Support Pack Stack 11 ( BI SP 13). Check to see what level you are at.
Yes, you need a target. Depending on what you are doing, if you have APD saving intermediate steps to tables, you might be able to identify the interim table thath a particular step saves it's data to, butI think you would be better off just defining a transaction ODS to be your target and write the data there.

Overlapping of clusters in InfiniteInsight Clustering model.

Similar Messages

Maybe you are looking for