Differences from 1.3

1. Decision tree induction can now output to the tree visualizer from SGI's MineSet.

2. Utilities now support command-line flags (options) as described in Section 2.

3. Names files now support ``discrete'' as a description for nominal attributes. As data is read, the list of values is dynamically added. Values in the test set that were not seen are considered unknown.

4. The C45Tree utility now supports the DIST_DISP option for displaying the distribution at every node.

5. The ``info'' utility now gives the majority accuracy.

6. Temporary files are now erased on interrupts. The environment variable KEEPTEMP can be set to ``yes'' to override this behavior.

7. Heuristics for determining the number of bins for discretization were added.

8. A new discretization option (t2-disc) was added. T2 is now given with the distribution (for discretization).

9. A new discretization option (c4.5-disc) was added.

10. CN2 algorithm is now given with the distribution. We thank Rick Kufrin from NCSA for modifying the original CN2 so that it easily compiles on SGI.

11. Discretization intervals were changed to conform with C4.5. The left branch is now instead of <.

12. Weights in nearest neighbor can be set ``manually,'' , through user input.

13. The C45test.awk script is no longer used nor needed.