J. R. Quinlan. Discovering rules from large collections of examples: a case study. In D. Michie, editor, Expert Systems in the Micro-electronic Age, pages 168--201. Edinburgh University Press, Edinburgh, 1979.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....algorithms are presented in Chapter 3 in more detail (Section 3.1 and 3.2) 2.2 Decision Trees Decision trees are one of the most well known and widely used approaches for learning from examples. This method was developed initially by Hunt, Marin and Stone [31] and later modified by Quinlan [49, 50]. Quinlan s ID3 [52] and C4.5 [55] are the most popular algorithms in decision tree induction. Initially, ID3 algorithm has applied to deterministic domains such as chess and games [49, 50] Later, ID3 algorithm has extended to cope with noisy and uncertain instances rather than being ....

....examples. This method was developed initially by Hunt, Marin and Stone [31] and later modified by Quinlan [49, 50] Quinlan s ID3 [52] and C4.5 [55] are the most popular algorithms in decision tree induction. Initially, ID3 algorithm has applied to deterministic domains such as chess and games [49, 50]. Later, ID3 algorithm has extended to cope with noisy and uncertain instances rather than being deterministic [52] Decision tree algorithms represents concept descriptions in the form of tree structure. Decision tree algorithms begin with a set of instances and create a tree data structure that ....

J.R. Quinlan, Discovering Rules From Large Collections of Examples: A Case Study, In D. Michie (Ed.),Expert Systems in the Microelectronic Age, Edinburgh: edinburgh University Press, 1979.

....[135] use log linear models to induce verbal transitivity. 2.2 Symbolic Machine Learning Approaches 2.2. 1 Decision Trees Decision tree based methods of supervised learning from examples represent one of the most popular approaches within the AI field for dealing with classification problems [17, 170, 171, 173]. Decision trees are a way to represent rules underlying training data, with hierarchical sequential structures that recursively partition the data. They have been used for years in several disciplines such as statistics, engineering (pattern recognition) decision theory (decision table ....

J. R. Quinlan. Discovering Rules from Large Collections of Examples. Edimburgh University Press, 1979. ???

....linear regression mainly because it is the algorithm of choice in domain studied here. The results here should be of potential interest to practitioners concerned with extracting quantitative structure activity relationships (SARs) It has been noted elsewhere in the machine learning literature [35] that finding appropriate descriptors for data is crucial to the success of an analysis method. The same concerns have FEATURE CONSTRUCTION WITH ILP 17 Data Equation with ILP attributes Number of terms in expert only equation PYR Act = 1:68MR 0 3;5 Gamma 1:66ILP009 1:15ILP017 6 Gamma ....

J.R. Quinlan. Discovering rules from large collections of examples: a case study. In D. Michie, editor, Expert Systems in the Micro-electronic Age, pages 168--201. Edinburgh University Press, Edinburgh, 1979.

....them to maintain consistency with the large data pool. In the context here, these revision operations simply result in addition and deletion of clauses. Logical windowing derives from techniques developed to enable decision tree methods (like ID3 and C4.5) to cope with large datasets [32]. There, a small sample or window of fixed size is drawn from the examples. These form the training set used to construct a decision tree. The tree is then tested on all examples. If this results in an intolerable number of errors, a small sample of the examples that are erroneously ....

J.R. Quinlan. Discovering rules from large collections of examples: a case study. In D. Michie, editor, Expert Systems in the Micro-electronic Age, pages 168--201. Edinburgh University Press, Edinburgh, 1979.

....(but not on the examples order) There is a huge bibliography in this area. We could single out, among others, the following chronologically ordered references [BFOS84, Qui86a, Utg90, Vel90, FI92, Qui93] which suppose different improvements on the basic algorithm proposed by Ross Quinlan in 1979 [Qui79]. In this context a problem of classification can be stated as follows: all the objects of the domain are described as a set of attribute value pairs, where each attribute measures a relevant feature of an object taking a (small) set of discrete, mutually incompatible values. Each object belongs ....

Quinlan, J.R. Discovering Rules From Large Collections of Examples: a Case Study. Edimburgh University Press, 1979.

....(see [27] The properties of various expert systems are given in Figure 1. The first two, MYCIN [38] and XCON [10] were built using hand coding of rules. The second two, GASOIL [39] and BMT [15] were built using software derived from Quinlan s inductive decision tree building algorithm ID3 [32]. It should be noted that the inductively constructed BMT is by far the largest expert system in full time commercial use. From the perspective of software engineering, it is also worth noting the sizeable reductions in development and maintenance times for inductively constructed systems. It ....

....Quinlan [33] has described a highly efficient program, called FOIL, which induces first order Horn clauses. The method relies on a general to specific heuristic search which is guided by an information criterion related to entropy. Quinlan sees his approach as being a natural extension of ID3 [32]. He notes that the search can be highly myopic, and is unable to learn predicates such as list reversal and integer multiplication. Attempts have been made recently by Buntine [4] Frisch and Page [11] and Muggleton and Feng [29] to find ways around Plotkin s negative RLGG results. Domain ....

J.R. Quinlan. Discovering rules from large collections of examples: a case study. In D. Michie, editor, Expert Systems in the Micro-electronic Age, pages 168--201. Edinburgh University Press, Edinburgh, 1979.

....to a leaf, the class of the case is predicted to be that record at the leaf. 2.2. 1 Decision Trees This method was developed initially by Hunt, Marin and Stone in 1966 [27] and later modified by Quinlan (1979, 1983) who applied his ID3 algorithm to deterministic domains such as chess and games [41, 42]. Quinlan s later CHAPTER 2. CONCEPT LEARNING MODELS 21 research has focused on induction on domains that are uncertain and noisy rather than deterministic. His approach is to synthesize decision trees that has been used in a variety of systems, and he has described his system ID3, the details ....

....CHAPTER 2. CONCEPT LEARNING MODELS 21 research has focused on induction on domains that are uncertain and noisy rather than deterministic. His approach is to synthesize decision trees that has been used in a variety of systems, and he has described his system ID3, the details can be found in [41, 42, 43]. A more extended version of ID3 is C4.5 [46] which can convert a decision tree to a rule base. The overall approach employed by ID3 and C4.5 is to choose the attribute that best divides the examples into classes and then partition the data according to the values of that attribute. This process ....

J.R. Quinlan, Discovering Rules From Large Collections of Examples: A Case Study, In D. Michie (Ed.), Expert Systems in the Microelectronic Age, Edinburgh: edinburgh University Press, 1979.

....constructed based on the errors of the first theory in a new sample which is a supserset of the first. Further layers of correcting theories are then added using successively larger samples until a pre specified level of overall theory accuracy is achieved. This approach is similar to what Quinlan [5] calls windowing . Clearly the minimal example requirements for layered learning will be limited by existing general lower bound PAC results. However, it is also clear that most of these examples will simply be used for testing the present stage of the theory. Only a small number of examples, ....

J.R. Quinlan. Discovering rules from large collections of examples: a case study. In D. Michie, editor, Expert Systems in the Micro-electronic Age, pages 168--201. Edinburgh University Press, Edinburgh, 1979.

....is typically followed by further generalisation and or decomposition into related sub concepts. The generality of the approach used allows CIGOL to exhibit a number of facets of Machine Learning. Thus CIGOL can be classed with systems which carry out 1. inductive concept formation such as [11, 21] 2. constructive induction such as [22, 13] 3. discovery such as [10, 9, 6] 4. generalisation of single examples using background knowledge such as [5, 15, 24] Unlike most learning systems described in the literature CIGOL uses an unrestricted form of first order Horn clause logic which allows ....

....the relative frequency of occurrence of symbol s in S as p s then, ignoring the length of the prefix table, M(S) has a length of jM(S)j N X s2sym(S) Gammap s log 2 p s bits according to Shannon information theory. There is obviously a similarity here to the entropy function used in ID3 [21], which should not be surprising given the common basis in information theory. Example 2 Let S be [crow(harry) black(X) crow(X) Then sym(S) is f : 2, 2, crow 1, black 1, X 0, harry 0, 0g, N = 10 and the corresponding relative frequencies are 0.1, 0.2, 0.2, 0.1, 0.2, 0.1, 0.1 . ....

J.R. Quinlan. Discovering rules from large collections of examples: a case study. In D. Michie, editor, Expert Systems in the Micro-electronic Age, pages 168--201. Edinburgh University Press, Edinburgh, 1979.

....large as possible. This ratio, however, may not always be defined or it may tend to favour attributes for which IV(A) is very small. The following criterion is used: from among those attributes with an averageor better gain, select the attribute that maximises the above ratio. 2.1. 2 Windowing [Qui79a] proposes the use of windowing as a method for accelerating and reducing the amount of machine storage required in the induction process. The system starts by randomly selecting a number of instances W (the window) from the training set and generates a decision tree with them. The tree is then ....

J. R. Quinlan. Discovering rules from large collections of examples: a case study. In D. Michie, editor, Expert Systems in the Micro-electronic Age, pages 168--201. Edinburgh University Press, Edinburgh, 1979.

....search are such that Shapiro failed to demonstrate that clauses as complex as the recursive clause of quick sort were learnable within reasonable time limits. To achieve greater efficiency Quinlan s FOIL greedily searches the same space guided by an information measure similar to that used in ID3 [9]. This measure supports the addition of a literal in the body of a clause on the basis of its ability to discriminate between positive and negative examples. This gains efficiency at the expense of completeness. For instance the literal partition(Head,Tail,List1,List2) in the recursive quick sort ....

J.R. Quinlan. Discovering rules from large collections of examples: a case study. In D. Michie, editor, Expert Systems in the Micro-electronic Age, pages 168--201. Edinburgh University Press, Edinburgh, 1979.

....to believe that it will never obtain very good performance 7 . 5 Decision Tree Methods Decision tree methods construct a tree which partitions the data at each level in the tree based on a particular feature of the data. CLS [30] used a heuristic lookahead method to construct decision trees. ID3 [44] extended CLS by using information content in the heuristic function. We tested the C4.5 algorithm by Ross Quinlan [45] which is an industrial strength version of ID3 designed to handle noise. 5 For an output range of 0 to 1. 6 Sequences of length zero up to the actual sequence length are ....

J.R. Quinlan. Discovering rules from large collections of examples: a case study. In D. Michie, editor, Expert Systems in the Microelectronic Age. Edinburgh University Press, Edinburgh, 1979.

....believe that it will never obtain very good performance 7 . 5 Decision Tree Methods Decision tree methods construct a tree which partitions the data at each level in the tree based on a particular feature of the data. CLS [29] used a heuristic lookahead method to construct decision trees. ID3 [44] extended CLS by using information content in the heuristic function. We tested the C4.5 algorithm by Ross Quinlan [45] which is an industrial strength version of ID3 designed to handle noise. C4.5 only deals with strings of constant length and we used an input space corresponding to the longest ....

J.R. Quinlan. Discovering rules from large collections of examples: a case study. In D. Michie, editor, Expert Systems in the Microelectronic Age. Edinburgh University Press, Edinburgh, 1979.

....Carbonell, Mitchell, 1986] Michalski, Kodratoff, 1990] In this task, a system typically learns one or more concepts by analyzing a set of examples (and possibly counterexamples) of the concepts. In fact, a number of systems for the acquisition of concepts now exist (e.g. ID3 [Quinlan, 1979], ARCH [Winston, 1975] COBWEB [Fisher, 1987] UNIMEM [Lebowitz, 1987] Independently, psychologists, psycholinguists, and cognitive scientists have examined the effects of numerous psychological limitations on human information processing. However, despite the fact that concept learning is a ....

Quinlan, J. R. (1979). Discovering Rules from Large Collections of Examples: A Case Study. In D. Michie (Ed.), Expert Systems in the Microelectronics Age Edinburgh: Edinburgh University Press.

....to Minsky and Papert [ 10 ] the parity function is unlearnable by single layer perceptrons. Recent techniques using multi layered perceptron networks [ 18 ] have been shown to be capable of learning parity effectively. However, in the paradigm of explicit rule formation, algorithms such as ID3 [ 14 ] and AQ11 [ 8 ] turn out to be rather inadequate when used to learn such functions. It has been shown [ 11 ] that whereas singlelevel concept representations of parity have a description complexity which is necessarily non polynomially dependent on the number of attributes, multi level ....

J.R. Quinlan. Discovering rules from large collections of examples: a case study. In D. Michie, editor, Expert Systems in the Micro-electronic Age, pages 168--201. Edinburgh University Press, Edinburgh, 1979.

No context found.

J. R. Quinlan. Discovering rules from large collections of examples: a case study. In D. Michie, editor, Expert Systems in the Micro-electronic Age, pages 168--201. Edinburgh University Press, Edinburgh, 1979.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC