About Magnus Assistant system V1.6 (1992) Dunja Mladenic J.Stefan Institute Jamova 39, Ljubljana Slovenia E-mail: dunja.mladenic@ijs.ac.mail.yu Contents: 1. Introduction 2. Basic Assistant algorithm 3. Input data 3.1 Domain definition 3.2 Set of examples 4. Main menu 5. Output data 6. How to run Magnus Assistant 1. Introduction =============== Assistant [1] is an attribute-value learning algorithm for classification tree construction. The algorithm belongs to the TDIDT (Top Down Induction of Decision Trees) family [6]. Magnus Assistant [4] is an implementation of the Assistant algorithm. It enables experiments with several hundreds of attributes and thousands of examples. TDIDT family: CLS (1963) | ID3 (1979) ----------------------|------------------------------------- | | | | ACLS (1981) ASSISTANT (1985) C4 (1986) ID4 | | | Expert-ease (1983) ASSISTANT 86 (1986) ID5 EX-TRAN (1984) | | RuleMaster (1984) ASSISTANT PROFESSIONAL (1988) ... SuperExpert (1986) | MAGNUS ASSISTANT (1990) 2. Basic Assistant algorithm ============================ GIVEN: set of training examples IF all examples belong to the same class THEN this node becomes a leaf labelled with the class of examples ELSE for the next node select the best attribute (the one which minimises the expected entropy of the training set) split the set of examples in the node into disjoint subsets according to binarised values of the attribute recursively repeat the whole algorithm for each subset ENDIF OUTPUT constructed classification tree Once built, a classification tree is interpreted in the following way: each internal node of the tree contains a test on a value of an attribute. According to the result of the test, the interpretation of the tree proceeds to the left or right subtree of the node. A leaf classifies an example into the appropriate class. 3. Input data ============= Before running the system, we have to prepare input data in some editor. Domain definition is written to file ASDO????.DAT and set of examples to file ASDA????.DAT. In practical use, ???? should be replaced with the domain name. 3.1 Domain definition ===================== Domain definition contains class values and all attributes, each with its set of values (for discrete attributes) or with interval bounds (for continuous attributes). Each attribute name is followed be the number of its values or interval bounds. A "-" sign, in front of the number of bounds, gives the information about the continuity of an attribute. The class values must always be discrete. This is illustrated on the domain definition file ASDOWEAT.DAT for weather prediction. 4 % no. of class values walk % name of the first class value bike % name of the second class value bus % ... auto % ... 3 % attribute no. WEATHER % name of the first attribute 4 % no. of values sunny % name of the first attribute value cloudy % name of the second attribute value rainy % ... snowy % ... TEMPERATURE % name of the second attribute -8 % no. of values (continuous attribute) -10 % value of the first bound (-oo, -10] -5 % value of the second bound (-10, -5] 0 % ... (-5, 0] 5 % ... (0, 5] 7 % ... (5, 7] 10 % ... (7, 10] 15 % ... (10, 15] 20 % ... (15, 20], (20, +oo) WIFE TOOK THE CAR % name of the third attribute 2 % no. of values yes % name of the first attribute value no % name of the second attribute value 3.2 Set of examples =================== In the set of examples each example is represented by a single line of numbers; the first number is the number of the class value, the second number is the number of the value of the first attribute (or the actual value of the first attribute if the attribute continuous),... There can be two special value: (999 or ?) for an unknown attribute value and (-999 or *) for an unimportant attribute value Each example is assigned a real value - weight of example, which is equal to one if there is no special value (999, -999) in example. Special values influence weight of examples to be different than number of examples. If there is no special value in the example set than number of examples and weights of examples are equal. Supose that in the weather prediction problem the following examples are given: CLASS WEATHER TEMPERATURE WIFE TOOK THE CAR bike sunny 6 Don't know bus rainy Don't care yes bus snowy 1 no walk sunny 4 no auto cloudy 7 no this is encoded into the file ASDAWEAT.DAT as follows: % Cl. A1 A2 A3 2 1 6 -999 3 3 999 1 3 4 1 2 1 1 4 2 4 2 7 2 4. Main menu ============ In the right upper corner is a name of active (current) domain. All actions are preformed on this domain. EXIT - exit to OS DEFINE DOMAIN - define domain name and paths Only first four letters of domain name are used (to replaced ????) There are three paths used by the Magnus Assistant: -Input path (DOMAIN) containing (ascii) files with the domain definition (ASDO????.DAT) and the example set (ASDA????.DAT), both prepared by the user. Here is also a file of initial options and parameters setting (ASIN.DAT) if saved when running the program. This file can be changed directly in some editor or rewritten from program when using the last main-menu choice. -Temporary path (ASISTMP) containing files with internal data representations: ASAT????.DAT - domain definition, ASEX????.DAT - set of examples, ASDI????.DAT - distribution of examples over attributes and values, ASED????.DAT - distribution of example set to train and test set, ASTR????.DAT - learned classification tree, ASFS????.DAT - file sizes for internal data representation files ASSU.DAT - set up options and parameters Before compiling data for a domain, all temporary files for this domain are deleted. -Result path (RESULTS) containing (ascii) files with results (output): ASDH????.PL - Prolog input file (as used by an ILP system LINUS [3]) ASTR????.TXT - learned tree in text format ASPT????.PL - learned tree in Prolog format ASRL????.PL - learned tree transcribed into a set of rules ASTO????.TXT - classification of testing ex. using tree only method ASBO????.TXT - classification of testing ex. using Bayes only method ASTB????.TXT - classification of testing ex. using tree and Bayes method ASBT????.TXT - classification of testing ex. using Bayes in tree method ASDS????.DAT - domain statistics COMPILE DOMAIN - compile domain definition and example set into internal representation Listing of compile data can be given on screen if appropriate option in the last main-menu choice is set (useful if an error occurs while compiling). TRANSFORM - transform data between input formats (transformation from Assistant to Prolog format is implemented) (SUBMENU) GENERATE TREE - generate classification tree using either parameters from file ASSU.DAT (if this file exists on the current directory) or default parameter setting DISPLAY TREE - display classification tree to file in one of three possible formats (text, Prolog, rules) or in text format to screen (SUBMENU) PRUNE TREE - post prune generated classification tree (overwrite the original tree) [5] CLASSIFY EXAMPLE - classify single example given interactively TEST TREE - test classification tree with chosen method (tree only, Bayes only, tree and Bayes, Bayes in tree) on chosen examples (all, train, test) (SUBMENU) DOMAIN STATISTICS - calculate domain statistics (distribution of classes, distribution of attribute values, distribution over classes and values of attributes) SET UP OPTIONS - set up options and parameters [2] (SUBMENU) Load and Save: If there is the file with options and parameters setting ASIN.DAT on input path, we can load setting from this file. The current setting can be saved to the file ASIN.DAT. Set: Setting is automatically written to file (ASSU.DAT) at the current directory. -only good - if true then only good (according to the Bayes classifier) examples are used for learning -test percent - percentage of examples for testing the learned tree (randomly choose examples) -old examples distribution - if true use the old (the last one) distribution of example set to train and test -informativity pruning factor - pre-pruning parameter, determines minimal information gain required from the selected (the best) attribute, to continue with the generation of subtree (default: 3.0) -class frequency threshold - pre-pruning parameter, determines the required probability of majority class in a current node, to stop generating the subtree (default: 100) -minimal weight threshold - pre-pruning parameter, determines minimal relative sum of weights of examples in a current node with respect to the sum of weights in the root of the tree, to continue with generating the subtree (default: 2.0) -post-pruning - if true post-prune tree immediately after its generation -list domain and examples while compiling - if true list domain and examples to the screen (useful if error occurs while compiling) 5. Output data ============== There are several files produced by the Magnus Assistant system, all are placed to the RESULT path (see main-menu/Define domain). Transformation of the input data into the Prolog format produces the output file ASDH????.PL, which can be used as the input to the ILP system LINUS. This is illustrated on the weather prediction domain in the file ASDHWEAT.PL. %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % example: weat % learning( class ). taxonomy( weat ). dependence( weat, weat, [ attr( class ) ], [ attr( 'WEATHER' ), attr( 'TEMPERATURE' ), attr( 'WIFE TOOK THE CAR' ) ] ). type_def( type_class, nom, [ walk, bike, bus, auto ] ). type_def( 'type_WEATHER', nom, [ sunny, cloudy, rainy, snowy ] ). type_def( 'type_TEMPERATURE', cont, [ -10.00, -5.00, 0.00, 5.00, 7.00, 10.00, 15.00, 20.00 ] ). type_def( 'type_WIFE TOOK THE CAR', nom, [ yes, no ] ). rel_def( weat, [ class : type_class, 'WEATHER' : 'type_WEATHER', 'TEMPERATURE' : 'type_TEMPERATURE', 'WIFE TOOK THE CAR' : 'type_WIFE TOOK THE CAR' ] ). rel( weat, [ bike, sunny, 6.00, ** ] ). rel( weat, [ bus, rainy, --, yes ] ). rel( weat, [ bus, snowy, 1.00, no ] ). rel( weat, [ walk, sunny, 4.00, no ] ). rel( weat, [ auto, cloudy, 7.00, no ] ). end. Classification tree can be written on the screen node by node, directed by the user or on the file in one of three formats: text format ASTR????.TXT, Prolog format ASPT????.PL, set of rules ASRL????.PL. This is illustrated on the weather prediction domain in the following files: ASTRWEAT.TXT. Decision tree from domain: weat Tree construction parameters: Instances for testing: 0 % ALL instances selected Informativity pruning factor: 0.0 Class frequency threshold: 100% Minimal weight threshold: 0% Post pruning: NO Number of Nodes : 7 Number of Leaves: 4 Number of Nulls : 0 | WEATHER [5, 6.00] % node: [ex.no., ex.weight] | [sunny] | | TEMPERATURE [2, 3.00] | | [ V <=5.00 ] % leaf: [ex.no., ex.weight, tree prob., Bayes prob.] | | | walk [1, 1.00, 100.0, 100.0] | | [5.00 < V ] | | | bike [1, 2.00, 100.0, 100.0] | [cloudy, rainy, snowy] | | WEATHER [3, 3.00] | | [cloudy] | | | auto [1, 1.00, 100.0, 100.0] | | [rainy, snowy] | | | bus [2, 2.00, 100.0, 100.0] ASPTWEAT.PL % parameters( assis, [ % domain : 'weat', % instances_for_testing_prc : 0, % only_good_inst : yes, % informativity_pruning_factor_prc : 0.0, % class_frequency_threshold_prc : 100, % minimal_weight_threshold : 0, % post_pruning : no, % number_of_nodes : 7, % number_of_leaves : 4, % number_of_nulls : 0 ] ). tree( 'WEATHER' : 5 : 6.00, null( [] ), [sunny] : tree( 'TEMPERATURE' : 2 : 3.00, null( [] ), [( 'V')=<5.00 ] : class( [ walk : 1 : 1.00 : 100.0 : 100.0 ] ), [(5.00 <'V') ] : class( [ bike : 1 : 2.00 : 100.0 : 100.0 ] ) ) , [cloudy, rainy, snowy] : tree( 'WEATHER' : 3 : 3.00, null( [] ), [cloudy] : class( [ auto : 1 : 1.00 : 100.0 : 100.0 ] ), [rainy, snowy] : class( [ bus : 2 : 2.00 : 100.0 : 100.0 ] ) ) ) . ASRLWEAT.PL % parameters( assis, [ % domain : 'weat', % instances_for_testing_prc : 0, % only_good_inst : yes, % informativity_pruning_factor_prc : 0.0, % class_frequency_threshold_prc : 100, % minimal_weight_threshold : 0, % post_pruning : no, % number_of_nodes : 7, % number_of_leaves : 4, % number_of_nulls : 0 ] ). rule( walk , [ [ 'WEATHER' = [ sunny ], 'TEMPERATURE' = [ ( 'V')=<5.00 ] ] : 1 : 1.0 : 100.0 : 100.0 ] ). rule( bike , [ [ 'WEATHER' = [ sunny ], 'TEMPERATURE' = [ (5.00 <'V') ] ] : 1 : 2.0 : 100.0 : 100.0 ] ). rule( bus , [ [ 'WEATHER' = [ cloudy, rainy, snowy ], 'WEATHER' = [ rainy, snowy ] ] : 2 : 2.0 : 100.0 : 100.0 ] ). rule( auto , [ [ 'WEATHER' = [ cloudy, rainy, snowy ], 'WEATHER' = [ cloudy ] ] : 1 : 1.0 : 100.0 : 100.0 ] ). Testing can be done by one of the four classification methods and written to the appropriate file: ASTO????.TXT for tree only method, ASBO????.TXT for Bayes only method, ASTB????.TXT for tree and Bayes method, ASBT????.TXT for Bayes in tree method. This is illustrated on the tree only method in the file ASTOWEAT.TXT for weather prediction domain. Testing results weat classifying training examples tree was generated on 100% training examples Tree only classification: Tree construction parameters: All instances Informativity pruning factor: 0.0% Class frequency threshold: 100% Minimal weight threshold: 0 Post-pruning: NO % predicted eq. cl. predicted % class cl.prob. cl.prob. prob. classified class % 1 bike 100.00 ---> 100.00 1 OK bike 2 bus 100.00 ---> 100.00 1 OK bus 3 bus 100.00 ---> 100.00 1 OK bus 4 walk 100.00 ---> 100.00 1 OK walk 5 auto 100.00 ---> 100.00 1 OK auto No. of instances: 5 % all ex. for classification No. of correctly classified instances: 5.00 % 1 ex. == 1/eq.cl.prob. Absolute accuracy: 100.00% % correct/all ex. Relative accuracy: 100.00% % all class.prob./all ex. Domain statistics, if required, is written to the file ASDS????.DAT. It consists of three parts: distribution of classes containing weight of examples for each class, accumulate containing weight of examples for all classes and for each attribute (column) and each attribute value (row), distribution over classes and values of attributes containing weight of examples for each class and each attribute (column) and each attribute value (row). This is illustrated on the weather prediction domain in the file ASDSWEAT.TXT DOMAIN STATISTICS: Distribution of classes: % ex.weight (in percent) walk 1 16.67% bike 2 33.33% bus 2 33.33% auto 1 16.67% ACCUMULATE 6 % for all classes, ex.weight 3 0 2 % V11 V21 V31 1 0 4 % V12 V22 V32 1 0 % ... 1 3 3 0 0 0 0 % A1 A2 A3; columns Distribution over classes and values of attributes walk 1 16.67% 1 0 0 0 0 1 0 0 0 1 0 0 0 0 0 bike 2 33.33% 2 0 1 0 0 1 0 0 0 0 2 0 0 0 0 bus 2 33.33% 0 0 1 0 0 1 1 0 1 2 0 0 0 0 0 auto 1 16.67% 0 0 0 1 0 1 0 0 0 0 1 0 0 0 0 6. How to run Magnus Assistant ============================== Suppose we have an executable file asis and three directories: domain, asistmp and results. There are two input files (asdoweat.dat and asdaweat.dat) on the domain directory. The following session illustrates running of the Magnus Assistant system on the weather prediction domain. asterix:105%asis Parameters setting (can be changed upon request): Mx. attributes:100 Mx. class/attribute values:60 Mx. examples:7000 Mx. tree nodes:800 Contact person: Dunja Mladenic, J. Stefan Institute Jamova 39, 61000 Ljubljana, Slovenia Phone: (+38)(61) 159 199, Fax:(+38)(61)161 029 E-mail: Dunja.Mladenic@ijs.ac.mail.yu Press ENTER to continue MAGNUS ASSISTANT V1.6 date: 10.09.1992 (c) AI Laboratory, J.Stefan Institute, Ljubljana, Slovenia MAIN MENU no domain specified 0 - exit 1 - define domain 2 - compile data 3 - transform data 4 - generate tree 5 - display tree 6 - prune tree 7 - classify example 8 - test tree 9 - domain statistics A - options choice: 1 DEFINE DOMAIN domain name: weat data directory [../domain/]: domain/ result directory [../result/]: result/ tmp directory [../asistmp/]: asistmp/ Press ENTER to continue MAGNUS ASSISTANT V1.6 date: 10.09.1992 (c) AI Laboratory, J.Stefan Institute, Ljubljana, Slovenia MAIN MENU active domain: weat 0 - exit 1 - define domain 2 - compile data 3 - transform data 4 - generate tree 5 - display tree 6 - prune tree 7 - classify example 8 - test tree 9 - domain statistics A - options choice: 2 compiling domain... classes: 4 attributes: 3 compiling examples... examples: 5 Press ENTER to continue MAGNUS ASSISTANT V1.6 date: 10.09.1992 (c) AI Laboratory, J.Stefan Institute, Ljubljana, Slovenia MAIN MENU active domain: weat 0 - exit 1 - define domain 2 - compile data 3 - transform data 4 - generate tree 5 - display tree 6 - prune tree 7 - classify example 8 - test tree 9 - domain statistics A - options choice: 3 Transformation: 0-exit 1-Assistant-Proffesional to Magnus-Assistant 2-Magnus-Assistant to Assistant-Proffesional 3-Assistant-Proffesional to DHDB choice: 3 Transformation: 0-exit 1-Assistant-Proffesional to Magnus-Assistant 2-Magnus-Assistant to Assistant-Proffesional 3-Assistant-Proffesional to DHDB choice: 0 Press ENTER to continue MAGNUS ASSISTANT V1.6 date: 10.09.1992 (c) AI Laboratory, J.Stefan Institute, Ljubljana, Slovenia MAIN MENU active domain: weat 0 - exit 1 - define domain 2 - compile data 3 - transform data 4 - generate tree 5 - display tree 6 - prune tree 7 - classify example 8 - test tree 9 - domain statistics A - options choice: 4 learning... Nodes: 3 ( Leaves: 1, Nulls: 0 ) Already built: 16.67% Nodes: 4 ( Leaves: 2, Nulls: 0 ) Already built: 50.00% Nodes: 4 ( Leaves: 2, Nulls: 0 ) Nodes: 6 ( Leaves: 3, Nulls: 0 ) Already built: 66.67% Nodes: 7 ( Leaves: 4, Nulls: 0 ) Already built: 100.00% Nodes: 7 ( Leaves: 4, Nulls: 0 ) Nodes: 7 ( Leaves: 4, Nulls: 0 ) Nodes: 7 ( Leaves: 4, Nulls: 0 ) calculating bayes prob. for each node ... Press ENTER to continue MAGNUS ASSISTANT V1.6 date: 10.09.1992 (c) AI Laboratory, J.Stefan Institute, Ljubljana, Slovenia MAIN MENU active domain: weat 0 - exit 1 - define domain 2 - compile data 3 - transform data 4 - generate tree 5 - display tree 6 - prune tree 7 - classify example 8 - test tree 9 - domain statistics A - options choice: 5 Format of writing the decision tree: 0-exit 1-text 2-prolog 3-rules 4-show only on screen choice: 1 show on screen (y/n) [n]: y ----- file: result/astrweat.txt ----- Decision tree from domain: weat Tree construction parameters: Instances for testing: 30 % ALL instances selected Informativity pruning factor: 3.0 Class frequency threshold: 100% Minimal weight threshold: 2% Post pruning: NO Number of Nodes : 7 Number of Leaves: 4 Number of Nulls : 0 | WEATHER [5, 6.00] | [sunny] | | TEMPERATURE [2, 3.00] | | [ V <=5.00 ] | | | walk [1, 1.00, 100.0, 100.0] | | [5.00 < V ] | | | bike [1, 2.00, 100.0, 100.0] | [cloudy, rainy, snowy] | | WEATHER [3, 3.00] | | [cloudy] | | | auto [1, 1.00, 100.0, 100.0] | | [rainy, snowy] | | | bus [2, 2.00, 100.0, 100.0] ----- file: result/astrweat.txt ----- Format of writing the decision tree: 0-exit 1-text 2-prolog 3-rules 4-show only on screen choice: 2 show on screen (y/n) [n]: ----- file: result/asptweat.pl ----- Format of writing the decision tree: 0-exit 1-text 2-prolog 3-rules 4-show only on screen choice: 4 Decision tree from domain: weat Tree construction parameters: Instances for testing: 30 % ALL instances selected Informativity pruning factor: 3.0 Class frequency threshold: 100% Minimal weight threshold: 2% Post pruning: NO Number of Nodes : 7 Number of Leaves: 4 Number of Nulls : 0 Press ENTER to continue LEVEL: 1 branch node with attribute: WEATHER [ no. of ex., weight of ex. ] [ 5, 6.00 ] Press ENTER to continue for each class: [ no. of ex., weight of ex., percent of weight, Bayes prob.] bike [ 1, 2.00, 33.3, 33.3 ] bus [ 2, 2.00, 33.3, 33.3 ] walk [ 1, 1.00, 16.7, 16.7 ] auto [ 1, 1.00, 16.7, 16.7 ] Press ENTER to continue for each attribute: [ attr. ok, attr. inf. ] WEATHER - discrete attribute [ TRUE, 1.00 ] TEMPERATUR- continuous attribute [ TRUE, 1.00 ] WIFE TOOK - discrete attribute [ TRUE, 0.25 ] 0-exit 1-father node 2-left son node 3-right son node 4-null leaf choice: 2 values:[ sunny ] LEVEL: 2 branch node with attribute: TEMPERATUR [ no. of ex., weight of ex. ] [ 2, 3.00 ] Press ENTER to continue for each class: [ no. of ex., weight of ex., percent of weight, Bayes prob.] bike [ 1, 2.00, 66.7, 66.7 ] walk [ 1, 1.00, 33.3, 33.3 ] Press ENTER to continue for each attribute: [ attr. ok, attr. inf. ] TEMPERATUR- continuous attribute [ TRUE, 0.92 ] WIFE TOOK - discrete attribute [ TRUE, 0.25 ] WEATHER - discrete attribute [ FALSE, 0.00 ] 0-exit 1-father node 2-left son node 3-right son node 4-null leaf choice: 2 values:[ V <=5.00 ] LEVEL: 3 leaf for each class: [ no. of ex., weight of ex., percent of weight, Bayes prob.] walk [ 1, 1.00, 100.0, 100.0 ] 0-exit 1-father node 2-left son node 3-right son node 4-null leaf choice: 3 THIS IS A LEAF, YOU MAY GO TO THE FATHER NODE !!! LEVEL: 3 leaf for each class: [ no. of ex., weight of ex., percent of weight, Bayes prob.] walk [ 1, 1.00, 100.0, 100.0 ] 0-exit 1-father node 2-left son node 3-right son node 4-null leaf choice: 1 LEVEL: 2 branch node with attribute: TEMPERATUR [ no. of ex., weight of ex. ] [ 2, 3.00 ] Press ENTER to continue for each class: [ no. of ex., weight of ex., percent of weight, Bayes prob.] bike [ 1, 2.00, 66.7, 66.7 ] walk [ 1, 1.00, 33.3, 33.3 ] Press ENTER to continue for each attribute: [ attr. ok, attr. inf. ] TEMPERATUR- continuous attribute [ TRUE, 0.92 ] WIFE TOOK - discrete attribute [ TRUE, 0.25 ] WEATHER - discrete attribute [ FALSE, 0.00 ] 0-exit 1-father node 2-left son node 3-right son node 4-null leaf choice: 0 Format of writing the decision tree: 0-exit 1-text 2-prolog 3-rules 4-show only on screen choice: 0 Press ENTER to continue MAGNUS ASSISTANT V1.6 date: 10.09.1992 (c) AI Laboratory, J.Stefan Institute, Ljubljana, Slovenia MAIN MENU active domain: weat 0 - exit 1 - define domain 2 - compile data 3 - transform data 4 - generate tree 5 - display tree 6 - prune tree 7 - classify example 8 - test tree 9 - domain statistics A - options choice: 6 tree was reduced to 100.00% of previous size Press ENTER to continue MAGNUS ASSISTANT V1.6 date: 10.09.1992 (c) AI Laboratory, J.Stefan Institute, Ljubljana, Slovenia MAIN MENU active domain: weat 0 - exit 1 - define domain 2 - compile data 3 - transform data 4 - generate tree 5 - display tree 6 - prune tree 7 - classify example 8 - test tree 9 - domain statistics A - options choice: 7 type of classification: 0-exit 1-tree only 2-bayes only 3-tree and bayes 4-bayes in tree choice: 2 Domain:weat tree was generated on 70% training examples Bayes only classification: Tree construction parameters: All instances Informativity pruning factor: 3.0% Class frequency threshold: 100% Minimal weight threshold: 2 Post-pruning: YES Tree was reduced to 100.00% of previous size Class values: 1. walk 2. bike 3. bus 4. auto ( 999, ? ) as unknown value, ( -999, * ) as don't care value choice: ? Attribute: WEATHER values: 1. sunny 2. cloudy 3. rainy 4. snowy ( 999, ? ) as unknown value, ( -999, * ) as don't care value choice: 1 Attribute: TEMPERATURE values: any real number ( 999, ? ) as unknown value, ( -999, * ) as don't care value value: 15 Attribute: WIFE TOOK THE CAR values: 1. yes 2. no ( 999, ? ) as unknown value, ( -999, * ) as don't care value choice: 1 probability( walk ) = 0.25 probability( bike ) = 0.25 probability( bus ) = 0.25 probability( auto ) = 0.25 classify another example (y/n) [y]: n type of classification: 0-exit 1-tree only 2-bayes only 3-tree and bayes 4-bayes in tree choice: 0 Press ENTER to continue MAGNUS ASSISTANT V1.6 date: 10.09.1992 (c) AI Laboratory, J.Stefan Institute, Ljubljana, Slovenia MAIN MENU active domain: weat 0 - exit 1 - define domain 2 - compile data 3 - transform data 4 - generate tree 5 - display tree 6 - prune tree 7 - classify example 8 - test tree 9 - domain statistics A - options choice: 8 type of classification: 0-exit 1-tree only 2-bayes only 3-tree and bayes 4-bayes in tree choice: 4 test on: 1-training examples 2-testing examples 3-all examples choice: 1 show on screen (y/n) [n]: y ----- file: result/asbtweat.txt ----- Testing results weat classifying training examples tree was generated on 70% training examples Bayes in tree classification: Tree construction parameters: All instances Informativity pruning factor: 3.0% Class frequency threshold: 100% Minimal weight threshold: 2 Post-pruning: YES Tree was reduced to 100.00% of previous size 1 bike 100.00 ---> 100.00 1 OK bike 2 bus 100.00 ---> 100.00 1 OK bus 3 bus 100.00 ---> 100.00 1 OK bus 4 walk 100.00 ---> 100.00 1 OK walk 5 auto 100.00 ---> 100.00 1 OK auto No. of instances: 5 No. of correctly classified instances: 5.00 Absolute accuracy: 100.00% Relative accuracy: 100.00% ----- file: result/asbtweat.txt ----- type of classification: 0-exit 1-tree only 2-bayes only 3-tree and bayes 4-bayes in tree choice: 0 Press ENTER to continue MAGNUS ASSISTANT V1.6 date: 10.09.1992 (c) AI Laboratory, J.Stefan Institute, Ljubljana, Slovenia MAIN MENU active domain: weat 0 - exit 1 - define domain 2 - compile data 3 - transform data 4 - generate tree 5 - display tree 6 - prune tree 7 - classify example 8 - test tree 9 - domain statistics A - options choice: 9 show on screen (y/n) [n]: ----- file: result/asdsweat.txt ----- Press ENTER to continue MAGNUS ASSISTANT V1.6 date: 10.09.1992 (c) AI Laboratory, J.Stefan Institute, Ljubljana, Slovenia MAIN MENU active domain: weat 0 - exit 1 - define domain 2 - compile data 3 - transform data 4 - generate tree 5 - display tree 6 - prune tree 7 - classify example 8 - test tree 9 - domain statistics A - options choice: a OPTIONS: 0-exit 1-load options 2-set options 3-save options choice: 2 SETUP OPTIONS selecting only good examples (y/n) [n]: instances for testing (0 - 99) [30%]: old example distribution (y/n) [n]: informativity pruning factor (0 - 9.9) [ 3.0]: class frequency threshold (1 - 100) [ 100%]: minimal weight threshold (0 - 99) [ 2%]: post pruning (y/n) [n]: list domain and examples while compiling (y/n) [n]: y OPTIONS: 0-exit 1-load options 2-set options 3-save options choice: 0 Press ENTER to continue MAGNUS ASSISTANT V1.6 date: 10.09.1992 (c) AI Laboratory, J.Stefan Institute, Ljubljana, Slovenia MAIN MENU active domain: weat 0 - exit 1 - define domain 2 - compile data 3 - transform data 4 - generate tree 5 - display tree 6 - prune tree 7 - classify example 8 - test tree 9 - domain statistics A - options choice: 0 asterix:105% [1] Cestnik, B., Kononenko, I. and Bratko, I. (1987) ASSISTANT 86: A Knowledge-Elicitation Tool for Sophisticated User. In Progress in Machine Learning, ed. by I.Bratko and N.Lavrac, Sigma Press, Wilmslow. [2] Kononenko, I., Cestnik, B. and Bratko, I. (1988) Assistant Professional User's Guide [3] Lavrac, N. (1990) Principles of knowledge acquisition in expert systems. Doctoral dissertation, University of Maribor, Faculty of Technical Sciences. [4] Mladenic, D. (1990) Machine learning system Magnus Assistant (in Slovene). BSc Thesis, University of Ljubljana, Faculty of Electrical Engineering and Computer Sciences. [5] Niblett, T.B. and Bratko, I. (1986) Learning decision rules in noisy domains. Expert Systems 86, Cambridge University Press. [6] Quinlan, J.R. (1986) Induction of decision trees. Machine Learning 1.