Hello, everyone
Recently I started using SAS EM and applied Decision Tree/Gradient Boosting model to my data. The data's response rate is 1%. Anyhow, for some reason, the Decision Tree run was suspended for Run time error and Gradient Boosting run results shows no lift, just ramdom distribution of data.
After I oversampled the data and make the new sample's response rate to be 10% or more, the decision tree and gradient boosting run were successful.
My question is, is there any parameter in Decision Tree and Gradient Boosting, after I change it, they can run successfully when the sample data's response rate is small, say 1% or below? Thanks,
Jimmy
Your question is a common one and is discussed in part in SAS Note 47965 available at
http://support.sas.com/kb/47/965.html
In general, it is always good to check running a Tree model if your Gradient Boosting node is not running since Gradient Boosting models. The strategies described in the note above will likely help you with your original data without oversampling.
For Gradient Boosting, try the following:
- lowering the Minimum Categorical Size property to 2 or 3
- changing the Missing Values property
To investigate further, open the Gradient Boosting node results and click on
View --> SAS Results --> Log
to view the log from your Gradient Boosting node and look for notes similar to the following:
...
NOTE: Will not search for split on variable A.
NOTE: Too few acceptable cases.
NOTE: Option MINCATSIZE=5 may apply.
NOTE: Will not search for split on variable B.
NOTE: Too few acceptable cases.
NOTE: Option MINCATSIZE=5 may apply
....
We have seen that message appear when some of the samples had an insufficient number of events and non-events and Gradient Boosting was unable to iterate. Sampling is used at different points to determine split values, and then the model is fit to the whole data set. If there are not enough events, then SAS Enterprise Miner cannot determine where the splits should occur.
Are other models able to run such as a regression model or decision tree? If so, examine your regression results to see whether there are many near-zero standard errors. Some customers have the opposite problem - infinite standard errors. For more information about this problem, please review
Usage Note 22599: Understanding and correcting complete or quasi-complete separation problems
http://support.sas.com/kb/22/599.html
I hope this helps!
Doug
Your question is a common one and is discussed in part in SAS Note 47965 available at
http://support.sas.com/kb/47/965.html
In general, it is always good to check running a Tree model if your Gradient Boosting node is not running since Gradient Boosting models. The strategies described in the note above will likely help you with your original data without oversampling.
For Gradient Boosting, try the following:
- lowering the Minimum Categorical Size property to 2 or 3
- changing the Missing Values property
To investigate further, open the Gradient Boosting node results and click on
View --> SAS Results --> Log
to view the log from your Gradient Boosting node and look for notes similar to the following:
...
NOTE: Will not search for split on variable A.
NOTE: Too few acceptable cases.
NOTE: Option MINCATSIZE=5 may apply.
NOTE: Will not search for split on variable B.
NOTE: Too few acceptable cases.
NOTE: Option MINCATSIZE=5 may apply
....
We have seen that message appear when some of the samples had an insufficient number of events and non-events and Gradient Boosting was unable to iterate. Sampling is used at different points to determine split values, and then the model is fit to the whole data set. If there are not enough events, then SAS Enterprise Miner cannot determine where the splits should occur.
Are other models able to run such as a regression model or decision tree? If so, examine your regression results to see whether there are many near-zero standard errors. Some customers have the opposite problem - infinite standard errors. For more information about this problem, please review
Usage Note 22599: Understanding and correcting complete or quasi-complete separation problems
http://support.sas.com/kb/22/599.html
I hope this helps!
Doug
Don't miss out on SAS Innovate - Register now for the FREE Livestream!
Can't make it to Vegas? No problem! Watch our general sessions LIVE or on-demand starting April 17th. Hear from SAS execs, best-selling author Adam Grant, Hot Ones host Sean Evans, top tech journalist Kara Swisher, AI expert Cassie Kozyrkov, and the mind-blowing dance crew iLuminate! Plus, get access to over 20 breakout sessions.
Use this tutorial as a handy guide to weigh the pros and cons of these commonly used machine learning algorithms.
Find more tutorials on the SAS Users YouTube channel.