New in Release 2.09
- Faster, less memory
This release focuses on large applications with cases/records numbering in the millions. (In fact, you may not notice any differences between the revised 2.08 and the new release 2.09 unless your application has at least 200,000 cases.)
For large applications, 2.09 is many times faster than 2.08 and uses significantly less memory. One application with 30 million cases ran 30 times faster on 2.09 (22 minutes instead of more than 11 hours) and used 20% less memory, but without any loss of predictive accuracy.
- Minor bugs fixed
2.08 sometimes gave incorrect feedback on progress through the stages of model formation.
For some applications, the previous release could also report very large errors when committee models were selected together with the option allowing Cubist to choose whether to use composite models. The models themselves were correct and their actual error was much lower than shown.
- Revised release
Two further bugs were corrected in the revised release (August 2016):
Windows: After a cross-validation, the relative error shown on the plot window was incorrect. The value shown on the main output was correct.
Linux: When the xval script was used to run multiple cross-validations, the average error magnitude shown on the summary was correct but rounded. The revised release shows the proper number of decimal places.
Here is a summary of changes in previous releases.
For Licensees Only:
Licensees who purchased Cubist since 27 January 2015 are welcome to upgrade to Release 2.09.
Click the appropriate link(s) below to download Release 2.09. You will be asked to re-enter your licence ID before running the system. If you use our public source code for reading and interpreting Cubist models, you should also download the latest version.
- Cubist (Linux):
- Cubist (Windows Xp/Vista/7/8/10):
- Either: source code for reading/interpreting models
|© RULEQUEST RESEARCH 2016||Last updated August 2016|