Difference between revisions of "Orange: CN2 Rule Induction"

From OnnoWiki
Jump to navigation Jump to search
 
(4 intermediate revisions by the same user not shown)
Line 1: Line 1:
 
Sumber: https://docs.biolab.si//3/visual-programming/widgets/model/cn2ruleinduction.html
 
Sumber: https://docs.biolab.si//3/visual-programming/widgets/model/cn2ruleinduction.html
  
−
Induce rules from data using CN2 algorithm.
+
Induce rule dari data menggunakan algoritma CN2.
  
−
Inputs
+
==Input==
  
−
    Data: input dataset
+
Data: input dataset
−
    Preprocessor: preprocessing method(s)
+
Preprocessor: preprocessing method(s)
  
−
Outputs
+
==Output==
  
−
    Learner: CN2 learning algorithm
+
Learner: CN2 learning algorithm
−
    CN2 Rule Classifier: trained model
+
CN2 Rule Classifier: trained model
  
−
The CN2 algorithm is a classification technique designed for the efficient induction of simple, comprehensible rules of form “if cond then predict class”, even in domains where noise may be present.
+
Algoritma CN2 adalah teknik klasifikasi yang di rancang agar secara effisien melakukan induksi dari rules yang simple & komprehesif dalam bentuk “if cond then predict class”, meskipun dalam domain dimana mungkin ada noise.
  
−
CN2 Rule Induction works only for classification.
+
CN2 Rule Induction hanya bisa jalan untuk klasifikasi saja.
  
 
[[File:CN2-stamped.png|center|200px|thumb]]
 
[[File:CN2-stamped.png|center|200px|thumb]]
  
−
    Name under which the learner appears in other widgets. The default name is CN2 Rule Induction.
+
* Name under which the learner appears in other widgets. The default name is CN2 Rule Induction.
 +
* Rule ordering:
 +
** Ordered: induce ordered rules (decision list). Rule conditions are found and the majority class is assigned in the rule head.
 +
** Unordered: induce unordered rules (rule set). Learn rules for each class individually, in regard to the original learning data.
  
−
    Rule ordering:
+
* Covering algorithm:
−
        Ordered: induce ordered rules (decision list). Rule conditions are found and the majority class is assigned in the rule head.
+
** Exclusive: after covering a learning instance, remove it from further consideration.
−
        Unordered: induce unordered rules (rule set). Learn rules for each class individually, in regard to the original learning data.
+
** Weighted: after covering a learning instance, decrease its weight (multiplication by gamma) and in-turn decrease its impact on further iterations of the algorithm.
  
−
    Covering algorithm:
+
* Rule search:
−
        Exclusive: after covering a learning instance, remove it from further consideration.
+
** Evaluation measure: select a heuristic to evaluate found hypotheses:
−
        Weighted: after covering a learning instance, decrease its weight (multiplication by gamma) and in-turn decrease its impact on further iterations of the algorithm.
+
*** Entropy (measure of unpredictability of content)
 +
*** Laplace Accuracy
 +
*** Weighted Relative Accuracy
 +
** Beam width; remember the best rule found thus far and monitor a fixed number of alternatives (the beam).
  
−
    Rule search:
+
* Rule filtering:
−
        Evaluation measure: select a heuristic to evaluate found hypotheses:
+
** Minimum rule coverage: found rules must cover at least the minimum required number of covered examples. Unordered rules must cover this many target class examples.
−
            Entropy (measure of unpredictability of content)
+
** Maximum rule length: found rules may combine at most the maximum allowed number of selectors (conditions).
−
            Laplace Accuracy
+
** Default alpha: significance testing to prune out most specialised (less frequently applicable) rules in regard to the initial distribution of classes.
−
            Weighted Relative Accuracy
+
** Parent alpha: significance testing to prune out most specialised (less frequently applicable) rules in regard to the parent class distribution.
−
        Beam width; remember the best rule found thus far and monitor a fixed number of alternatives (the beam).
+
* Tick ‘Apply Automatically’ to auto-communicate changes to other widgets and to immediately train the classifier if learning data is connected. Alternatively, press ‘Apply‘ after configuration.
−
 
 
−
    Rule filtering:
 
−
        Minimum rule coverage: found rules must cover at least the minimum required number of covered examples. Unordered rules must cover this many target class examples.
 
−
        Maximum rule length: found rules may combine at most the maximum allowed number of selectors (conditions).
 
−
        Default alpha: significance testing to prune out most specialised (less frequently applicable) rules in regard to the initial distribution of classes.
 
−
        Parent alpha: significance testing to prune out most specialised (less frequently applicable) rules in regard to the parent class distribution.
 
−
    Tick ‘Apply Automatically’ to auto-communicate changes to other widgets and to immediately train the classifier if learning data is connected. Alternatively, press ‘Apply‘ after configuration.
 
  
 
==Contoh==
 
==Contoh==
  
−
For the example below, we have used zoo dataset and passed it to CN2 Rule Induction. We can review and interpret the built model with CN2 Rule Viewer widget.
+
Dalam contoh di bawah, kita menggunakan dataset zoo dan mengirimkannya ke CN2 Rule Induction. Kita bisa me-review dan meng-interpretasi model yang dibuat dengan CN2 Rule Viewer widget.
  
−
[[File:CN2-visualize.png|center|200px|thumb]]
+
[[File:CN2-visualize.png|center|600px|thumb]]
  
−
The second workflow tests evaluates CN2 Rule Induction and Tree in Test & Score.
+
Workflow yang ke dua men-test evaluasi CN2 Rule Induction dan Tree di Test & Score.
−
 
 
−
[[File:CN2-classification.png|center|200px|thumb]]
 
  
 +
[[File:CN2-classification.png|center|600px|thumb]]
  
 
==Referensi==
 
==Referensi==
  
−
    Fürnkranz, Johannes. “Separate-and-Conquer Rule Learning”, Artificial Intelligence Review 13, 3-54, 1999.
+
* Fürnkranz, Johannes. “Separate-and-Conquer Rule Learning”, Artificial Intelligence Review 13, 3-54, 1999.
−
 
+
* Clark, Peter and Tim Niblett. “The CN2 Induction Algorithm”, Machine Learning Journal, 3 (4), 261-283, 1989.
−
    Clark, Peter and Tim Niblett. “The CN2 Induction Algorithm”, Machine Learning Journal, 3 (4), 261-283, 1989.
+
* Clark, Peter and Robin Boswell. “Rule Induction with CN2: Some Recent Improvements”, Machine Learning - Proceedings of the 5th European Conference (EWSL-91),151-163, 1991.
 +
* Lavrač, Nada et al. “Subgroup Discovery with CN2-SD”,Journal of Machine Learning Research 5, 153-188, 2004
  
−
    Clark, Peter and Robin Boswell. “Rule Induction with CN2: Some Recent Improvements”, Machine Learning - Proceedings of the 5th European Conference (EWSL-91),151-163, 1991.
 
  
−
    Lavrač, Nada et al. “Subgroup Discovery with CN2-SD”,Journal of Machine Learning Research 5, 153-188, 2004
 
  
 +
==Youtube==
  
 +
* [https://www.youtube.com/watch?v=8c1jQkI8NCY YOUTUBE: Model Constant & CN2 Rule Inducer]
  
 
==Referensi==
 
==Referensi==

Latest revision as of 06:42, 10 April 2020

Sumber: https://docs.biolab.si//3/visual-programming/widgets/model/cn2ruleinduction.html

Induce rule dari data menggunakan algoritma CN2.

Input

Data: input dataset
Preprocessor: preprocessing method(s)

Output

Learner: CN2 learning algorithm
CN2 Rule Classifier: trained model

Algoritma CN2 adalah teknik klasifikasi yang di rancang agar secara effisien melakukan induksi dari rules yang simple & komprehesif dalam bentuk “if cond then predict class”, meskipun dalam domain dimana mungkin ada noise.

CN2 Rule Induction hanya bisa jalan untuk klasifikasi saja.

CN2-stamped.png
  • Name under which the learner appears in other widgets. The default name is CN2 Rule Induction.
  • Rule ordering:
    • Ordered: induce ordered rules (decision list). Rule conditions are found and the majority class is assigned in the rule head.
    • Unordered: induce unordered rules (rule set). Learn rules for each class individually, in regard to the original learning data.
  • Covering algorithm:
    • Exclusive: after covering a learning instance, remove it from further consideration.
    • Weighted: after covering a learning instance, decrease its weight (multiplication by gamma) and in-turn decrease its impact on further iterations of the algorithm.
  • Rule search:
    • Evaluation measure: select a heuristic to evaluate found hypotheses:
      • Entropy (measure of unpredictability of content)
      • Laplace Accuracy
      • Weighted Relative Accuracy
    • Beam width; remember the best rule found thus far and monitor a fixed number of alternatives (the beam).
  • Rule filtering:
    • Minimum rule coverage: found rules must cover at least the minimum required number of covered examples. Unordered rules must cover this many target class examples.
    • Maximum rule length: found rules may combine at most the maximum allowed number of selectors (conditions).
    • Default alpha: significance testing to prune out most specialised (less frequently applicable) rules in regard to the initial distribution of classes.
    • Parent alpha: significance testing to prune out most specialised (less frequently applicable) rules in regard to the parent class distribution.
  • Tick ‘Apply Automatically’ to auto-communicate changes to other widgets and to immediately train the classifier if learning data is connected. Alternatively, press ‘Apply‘ after configuration.

Contoh

Dalam contoh di bawah, kita menggunakan dataset zoo dan mengirimkannya ke CN2 Rule Induction. Kita bisa me-review dan meng-interpretasi model yang dibuat dengan CN2 Rule Viewer widget.

CN2-visualize.png

Workflow yang ke dua men-test evaluasi CN2 Rule Induction dan Tree di Test & Score.

CN2-classification.png

Referensi

  • Fürnkranz, Johannes. “Separate-and-Conquer Rule Learning”, Artificial Intelligence Review 13, 3-54, 1999.
  • Clark, Peter and Tim Niblett. “The CN2 Induction Algorithm”, Machine Learning Journal, 3 (4), 261-283, 1989.
  • Clark, Peter and Robin Boswell. “Rule Induction with CN2: Some Recent Improvements”, Machine Learning - Proceedings of the 5th European Conference (EWSL-91),151-163, 1991.
  • Lavrač, Nada et al. “Subgroup Discovery with CN2-SD”,Journal of Machine Learning Research 5, 153-188, 2004


Youtube

Referensi

Pranala Menarik

Retrieved from "https://onnocenter.or.id/wiki/index.php?title=Orange:_CN2_Rule_Induction&oldid=61078"