C4.5算法學習【一】

爲了記錄，也爲了日後重溫，更爲了交學習報告，決定寫下每天的學習情況

先貼點學習資料網站什麼的：

http://www2.cs.uregina.ca/~dbd/cs831/index.html（剛好有weka，kdd的介紹，更剛好又C4.5）

 

"Quinlan" "C4. 5: programs for machine learning"的電子版，可惜不能下

http://www.google.com/books?hl=zh-CN&lr=&id=HExncpjbYroC&oi=fnd&pg=PR7&dq=%22Quinlan%22+%22C4.+5:+programs+for+machine+learning%22+&ots=nJpdcSx0Xo&sig=djBDwgiSRoqHinBnnMzxJo3vOpk#v=onepage&q=&f=false

 

關於C4.5的介紹暫時略。

本來想看weka的源碼的，可以根本就是白費工，Java纔剛學，杯具.....

第一天分析：

結合這個網頁的資料：

http://www2.cs.uregina.ca/~dbd/cs831/notes/ml/dtrees/c4.5/tutorial.html

下載裏面的源碼：c4.5r8.tar.gz

 

 

因爲是"C4.5 Release 8 Installation Instructions for UNIX"

所以只好弄到我的虛擬機裏的linux（ubuntu）裏了。

 

然後按照這個：

1. Download the C4.5 source code.
2. Decompress the archive:
3. Type "tar xvzf c4.5r8.tar" (not universally supported), or, alternatively,
4. Type "gunzip c4.5r8.tar.gz" to decompress the gzip archive, and then
   Type "tar xvf c4.5r8.tar" to decompress the tar archive.
5. Change to ./R8/Src
6. Type "make all" to compile the executables.
7. Put the executables into a "bin" subdirectory and include it in the path for command-line usage.

弄好就可以用了。

 

源碼就在 R8/Src裏

 

先從 c4.5.c  這個文件開始吧。看着代碼才發現自己的C/C++水平很菜！分析的不對請指出。

 

 

 

 #include "defns.i" #include "types.i" //#include "extern.i" /* External data, described in extern.i */ short MaxAtt, /* max att number *///最大屬性數目 MaxClass, /* max class number *///最大類數 MaxDiscrVal = 2;/* max discrete values for any att */ ItemNo MaxItem; /* max data item number */ Description *Item; /* data items */ /* typedef union _attribute_value { DiscrValue _discr_val; float _cont_val; } AttValue, *Description; */ DiscrValue *MaxAttVal; /* number of values for each att */ //short char *SpecialStatus; /* special att treatment */ String *ClassName, /* class names *///類名 *AttName, /* att names *///屬性名 **AttValName, /* att value names *///屬性的值的名 FileName = "DF"; /* family name of files */ short VERBOSITY = 0, /* verbosity level (0 = none) *///冗長級別？ TRIALS = 10; /* number of trees to be grown */ //char //typedef char Boolean, *String, *Set; Boolean GAINRATIO = true, /* true=gain ratio, false=gain *///true表示信息增益率,false表示信息增益 SUBSET = false, /* true if subset tests allowed */ BATCH = true, /* true if windowing turned off */ UNSEENS = false, /* true if to evaluate on test data */ PROBTHRESH = false; /* true if to use soft thresholds */ ItemNo MINOBJS = 2, /* minimum items each side of a cut */ WINDOW = 0, /* initial window size */ INCREMENT = 0; /* max window increment each iteration */ float CF = 0.25; /* confidence limit for tree pruning *///置信界限？ Tree *Pruned; Boolean AllKnown = true; /* true if there have been no splits on atts with missing values above the current position in the tree */ main(Argc, Argv) /* ---- */ int Argc; char *Argv[]; { int o; extern char *optarg; //#include<unistd.h>裏有(在getopt.c也有 extern int optind; //#include<unistd.h>裏有(在getopt.c也有 /* extern的用法？？？？？？？不會啊，杯具！ */ Boolean FirstTime=true; short Best, BestTree(); PrintHeader("decision tree generator"); /* Process options */ //處理選項及其參數。 //unistd.h 函數說明 getopt()用來分析命令行參數 //百度了一翻 - -！ while ( (o = getopt(Argc, Argv, "f:bupv:t:w:i:gsm:c:")) != EOF ) { if ( FirstTime ) { printf("/n Options:/n"); FirstTime = false; } switch (o) { case 'f': FileName = optarg; printf("/tFile stem

 

杯具一次插不完 - -！ 略掉一部分不是很重要的繼續插 - -！

/* Initialise */ //讀入類名，屬性名，屬性值？統計實例數、屬性數？ //數據存放於？下面的說明？ /*************************************************************************/ /* */ /* Read the names of classes, attributes and legal attribute values. */ /* On completion, these names are stored in: */ /* ClassName - class names *///放類名? /* AttName - attribute names *///放屬性名? /* AttValName - attribute value names *///屬性值? /* with: */ /* MaxAttVal - number of values for each attribute */ /* */ /* Other global variables set are: */ /* MaxAtt - maximum attribute number */ /* MaxClass - maximum class number */ /* MaxDiscrVal - maximum discrete values for any attribute */ /* */ /* Note: until the number of attributes is known, the name */ /* information is assembled in local arrays */ /* */ /*************************************************************************/ GetNames(); //golf.names 在getnames.c GetData(".data"); //golf.data 在getdata.c printf("/nRead %d cases (%d attributes) from %s.data/n", MaxItem+1, MaxAtt+1, FileName); //14 4 golf /* Build decision trees *///建立決策樹 if ( BATCH ) { TRIALS = 1; OneTree(); //where? Best = 0; } else { Best = BestTree(); //where? } /* Soften thresholds in best tree */ if ( PROBTHRESH ) { printf("Softening thresholds"); if ( ! BATCH ) printf(" for best tree from trial %d", Best); printf("/n"); SoftenThresh(Pruned[Best]); printf("/n"); PrintTree(Pruned[Best]); } /* Save best tree */ if ( BATCH || TRIALS == 1 ) { printf("/nTree saved/n"); } else { printf("/nBest tree from trial %d saved/n", Best); } SaveTree(Pruned[Best], ".tree"); /* Evaluation */ printf("/n/nEvaluation on training data (%d items):/n", MaxItem+1); Evaluate(false, Best); if ( UNSEENS ) { GetData(".test"); printf("/nEvaluation on test data (%d items):/n", MaxItem+1); Evaluate(true, Best); } exit(0); }