Copying and reusing nodes in Enterprise Miner screwing up model selection

Ullsokk — Thu, 24 Aug 2017 07:35:30 GMT

I am using SAS Enterprise Miner Clien 14.2.

When copying an analysis node, e.g. decision tree, to resuse the settings and tweek them to se what works best in a model comparison, the ROC curve goes crazy and seems to jump between the nodes of the same name. Is there a way to copy and paste the basic settings in such a node (number of splits, depth, etc) without this happening? Or do I have to manually point and click my way each time, increasing the chance of errors and slowing down my pace immencely?

Re: Copying and reusing nodes in Enterprise Miner screwing up model selection

DougWielenga — Fri, 25 Aug 2017 18:36:18 GMT

I'm not aware of anyone else who has reported this behavior, but I would recommend the following:

1 - Check whether it is limited to a single project (e.g. if you build a new project, does the situation persist?)

2 - Check whether you are running a supported version of Java (e.g. Java always tries to update itself but is not backwards compatible so 'newer' versions of Java can cause applications running on Java to behave strangely)

3 - Check your available system resources such as available RAM, unused disk space, etc... (when you run low on memory be it virtual or disk memory, you can get strange behavior)

4 - Check how much memory is being allocated to Java (e.g. apps running Java can behave strangely when Java runs out of memory) by submitting the code below in a SAS Code node in SAS Enterprise Miner

proc options group=memory;

run;

How Much Memory?

Memory for Client and Server Machines

The SAS Enterprise Miner 14.1 product is designed to run in a client/server environment. Here are some recommendations for memory on SAS Enterprise Miner 14.1 client and server machines.

• Client — a minimum of 4 GB of physical RAM dedicated to the SAS Enterprise Miner 14.1 product running on the client machine.

• Server — a minimum of 6 GB of physical RAM, plus 2 GB per concurrent user.

• Mid-Tier — a minimum of 8 GB of physical RAM.

With regards to how much memory you need, it depends on the nodes being process and the size of the data sources. I am quoting the SAS Enterprise Miner 14.1 Reference Help from page 28 of the PDF below:

/*** BEGIN HELP EXCERPT ***/

Configuring the MEMSIZE Parameter

After sizing and configuring memory for your server, you must allocate memory for the SAS Enterprise Miner 14.1 application. Use the MEMSIZE parameter in your config.sas file to specify the memory allocation for SAS Enterprise Miner 14.1. See the Base SAS documentation for the location of the config.sas file in your SAS 9.4 installation.

The MEMSIZE option is the upper limit to the amount of memory the SAS System can use on a per-session basis. Each session uses memory as required up to the specified MEMSIZE limit. The configuration file is located in the root directory of your SAS Enterprise Miner 14.1 installation.

• Windows Server — The MEMSIZE parameter is set to 0 by default, which allows each SAS session to take all available memory on the server. The MEMSIZE parameter recommendation is 512 MB.

• UNIX Server — The MEMSIZE parameter is set to a given value by default. The value varies according to different UNIX versions of SAS. Depending on the memory requirements for your project, it might be necessary to increase the value specified for the MEMSIZE option. The MEMSIZE parameter recommendation is the amount of RAM equal to 80% of available host memory. Refer to the host documentation for more information about this option.

Never set the MEMSIZE option to a value that is too high. If multiple SAS sessions are running and every SAS session consumes the maximum memory allotted by MEMSIZE, it is possible to exceed the amount of physical RAM on some servers. On Windows servers, the sum of SAS Enterprise Miner 14.1 MEMSIZE allocations for the peak concurrent users should never exceed the physical amount of RAM, as such conditions result in severe I/O bottlenecking.

/*** END HELP EXCERPT ***/

In general terms 8G is usually sufficient.

You can test memsize at the individual level by having the user create a sasv9.cfg in his HOME directory and specify:

-MEMSIZE=8G

-MAXMEMQUERY=2G

The analogy I make is that MEMSIZE is your account balance. MAXMEMQUERY is how much you can get out of the atm at one time.

For Java memory, check the JavaArgs in the em.ini file and try to increase the Java memory shown there. For example, try setting the values

JavaArgs_1=-Xms512m

JavaArgs_2=-Xmx1024m

or even go as high as 1024 and 2048 respectively)

If the situation persists, contact SAS Technical Support and they will likely ask you to do a WebEx session so they can see the behavior first-hand.

Hope this helps!

Doug

topic Re: Copying and reusing nodes in Enterprise Miner screwing up model selection in SAS Data Science

Copying and reusing nodes in Enterprise Miner screwing up model selection

Re: Copying and reusing nodes in Enterprise Miner screwing up model selection