New in Release 2.11
- New pruning heuristics
The methods used to avoid overfitting have been enhanced. Decision
trees in particular are generally smaller than previously, but
predictive accuracy is usually improved.
- Soft thresholds
Earlier releases contained an option to "soften" thresholds on
continuous attributes in decision trees.
Experiments with large collections of applications with continuous attributes
have shown that this leads to
a near-universal increase in predictive accuracy.
Consequently, threshold softening is now standard
-- all thresholds are soft in Release 2.11. To simplify the display
of trees, however, the thresholds are shown as single values.
A new option is available to present the complete information about
As before, threshold softening does not affect rule-based classifiers in any way.
- Further multi-threading for rulesets
The final selection of rules and the optional division of rules into utility
bands both make use of multiple threads.
- Minor changes
The cross-validation option for the Windows version See5 has been
brought into line with the Linux C5.0 procedure.
Test cases, if they are present, are now included when a
cross-validation is performed.
The feedback on classifier formation has been tweaked in an attempt to make it more informative with respect to the time remaining to complete each stage.
Click here for information on updates in previous releases.
For Licensees Only:
Licensees who are currently running Release 2.10 are welcome to upgrade to Release 2.11.
Click the appropriate link(s) below to download Release 2.11. You will be asked to re-enter your licence ID before using the system. If you use the source code for reading and interpreting See5/C5.0 classifiers, you should also download the latest version.
- C5.0 (Linux):
- See5 (Windows 7/8/10):
- Single-computer licence (32-bit)
- Single-computer licence (64-bit)
- Network licence (32-bit and 64-bit)
- Either: source code for reading/interpreting classifiers
|© RULEQUEST RESEARCH 2018||Last updated January 2018|