source: proiecte/Parallel-DT/R8/Src/Modifications @ 26

Last change on this file since 26 was 26, checked in by (none), 14 years ago

blabla
File size: 7.0 KB
Line 
1Modifications since book was published:
2
3--------------------------------------------------------------------------------
4(1)     17 August 1992:  fixed bug in prunerule.c
5
6        In routine Satisfies about line 434:
7        moved statement
8                    t->Outcome = -1;
9        to before the for loop
10--------------------------------------------------------------------------------
11(2)     2nd Feb 1993: fixed errors reported by Dick Jackson
12
13        c4.5rules.c line 34: changed ';' to ','
14        getnames.c: moved CopyString() declaration to head
15--------------------------------------------------------------------------------
16(3)     19th June 1993: fixed error reported by Guillermo Irisarri
17
18        ANSI C doesn't like "exit()" with no args in average.c, xval-prep.c
19--------------------------------------------------------------------------------
20(4)     5th July 1993: fixed bug in c4.5rules reported by Ray Mooney
21
22        SaveRules() was invoked before EvaluateRulesets(), but the latter
23        can delete globally unhelpful rules.  SaveRules() was moved to
24        after evaluation of rules on training data
25            (Note: this change affects only the use of consultr with the
26             saved rules; experimental results are unaltered.)
27--------------------------------------------------------------------------------
28(5)     13th July 1993: changed rules.c to improve printing with -s option
29
30        When tests on discrete attributes use value groups, the standard
31        form of test is
32            "<attribute> in {<value>, <value>, ...}".
33        If there is only one value, this should appear as
34            "<attribute> = <value>".
35        This has already been changed in trees; a similar change has now
36        been made to function PrintCondition() in rules.c
37--------------------------------------------------------------------------------
38(6)     28th July 1993: killed very large confusion matrices
39
40        confmat.c line 19: added copout if number of classes > 20
41--------------------------------------------------------------------------------
42(7)     9th September 1993: fixed problems notified by Mike Jankulak.
43
44        * Added checks for reasonable parameter values in c4.5, c4.5rules.
45          Check in GetNames() for discrete N: N must be at least 2.
46
47        * consult, consultr don't work with attributes of type discrete N !
48          Added routines in trees.c to save and restore values of attributes
49          of this type when saving / reading trees.
50          Modified rules.c to invoke these routines when saving / reading
51          rulesets.
52
53          NOTE: old .tree, .unpruned and .rules files must be regenerated
54                if they are to be used by the modified programs.
55--------------------------------------------------------------------------------
56(8)     3rd November 1993; problem notified by Jason Catlett
57
58        c4.5rules prints an incorrect confusion matrix for the training
59        set when rules are dropped.  Altered testrules.c.
60--------------------------------------------------------------------------------
61(9)     21st December 1993; tidying up only
62
63        Changed definition of Log() in defns.i so that argument of log()
64        is guaranteed float.
65--------------------------------------------------------------------------------
66(10)    5th February 1994; problem notified by George John
67
68        Calculation of Gain in build.c can be negative rather than zero
69        due to FP rounding.  Changed tests "Gain[Att] >= 0" to
70        "Gain[Att] > -Epsilon".
71--------------------------------------------------------------------------------
72(11)    25th May 1994; problem notified by Ronny Kohavi
73
74        Similar problem in info.c with -g option.  Changed test
75        "ThisGain > 0" to "ThisGain > -Epsilon".
76--------------------------------------------------------------------------------
77(12)    30th May 1994; tidying up
78
79        Removed explicit Outcomes field from rules.  This simplifies
80        the code somewhat with little decrease in efficiency.
81--------------------------------------------------------------------------------
82(13)    18th July 1994; problem notified by Ronny Kohavi
83
84        Average gain evaluated incorrectly when all attributes have
85        many discrete values.  In build.c, introduced MultiVal to check
86        for this contingency.
87--------------------------------------------------------------------------------
88(14)    18th-20th July 1994; modifications to siftrules.c
89
90        (a) Changed coding of exceptions:
91            * added cost of encoding total number of errors to cost of
92              identifying false positives and false negatives.
93            * applied penalty to non-representative theories as described
94              in my ML'94 paper.
95        (b) Introduced a new form of local greedy search for finding
96            good subsets when there are more than 10 rules.  This is
97            faster than simulated annealing and replaces it as the default:
98            simulated annealing is still available via a new option -a.
99--------------------------------------------------------------------------------
100
101***********************  Release 6  July 1994  *********************************
102
103--------------------------------------------------------------------------------
104(15)    11th August 1994; bug reported by KaiMing Ting and Zijian Zheng
105
106        In subset.c, DiscrKnownBaseInfo() can be called when KnownItems = 0.
107        Trapped such calls.
108--------------------------------------------------------------------------------
109(16)    21st January 1995; bug reported by Tom Fawcett
110
111        Very large trees can cause the short int in TreeSize to overflow.
112        Changed to int.
113--------------------------------------------------------------------------------
114(17)    6th April 1995; bug reported by Ronny Kohavi
115
116        Exit status not being set properly.  Modified the following:
117            c4.5.c, c4.5rules.c, consult.c, consultr.c.
118--------------------------------------------------------------------------------
119(18)    19th April 1995; bug reported by Kim Horn
120
121        For very small values of CF less than 0.1%, confidence levels
122        are computed erratically.  Modified stats.c.
123--------------------------------------------------------------------------------
124(19)    June 1995: modifications to siftrules.c (again!)
125
126        Scheme described above in 14(a) amended in line with my ML'95
127        paper, available by anonymous ftp from ftp.cs.su.oz.au, directory
128        pub/ml, file q.ml95.ps.Z.
129--------------------------------------------------------------------------------
130
131***********************  Release 7  June 1995  *********************************
132
133--------------------------------------------------------------------------------
134(20)    6th July 1995; bug reported by Andrew Taylor
135
136        Tree printing can have problems when attribute names are very long.
137        Modified trees.c.
138--------------------------------------------------------------------------------
139(21)    18th October 1995: modifications to contin.c
140
141        Altered the calculation of gain for continuous attributes (described
142        in "Improved Use of Continuous Attributes in C4.5").
143--------------------------------------------------------------------------------
144
145***********************  Release 8  October 1995  ******************************
146
147--------------------------------------------------------------------------------
148(22)    26th Feb 1996; minor glitches reported by Ron Kohavi of SGI
149
150        Fn declared extern in rules.c
151        -lm removed from consult, consultr, xval-prep
152--------------------------------------------------------------------------------
Note: See TracBrowser for help on using the repository browser.