trains (Q6032908): Difference between revisions

**Author**: \N**Source**: Unknown - \N**Please cite**: \N\N1. Title: INDUCE Trains Data set\N \N 2. Sources:\N - Donor: GMU, Center for AI, Software Librarian,\N Eric E. Bloedorn (bloedorn@aic.gmu.edu)\N - Original owners: Ryszard S. Michalski (michalski@aic.gmu.edu)\N and Robert Stepp\N - Date received: 1 June 1994\N - Date updated: 24 June 1994 (Thanks to Larry Holder (UT Arlington)\N for noticing a translation error)\N \N 3. Past usage:\N - This set most closely resembles the data sets described in the following\N two publications:\N 1. R.S. Michalski and J.B. Larson "Inductive Inference of VL\N Decision Rules" In Proceedings of the Workshop in Pattern-Directed \N Inference Systems, Hawaii, May 1977. Also published in SIGART\N Newsletter, ACM No. 63, pp. 38-44, June 1977.\N 2. Stepp, R.E. and Michalski, R.S. "Conceptual Clustering: Inventing \N Goal-Oriented Classifications of Structured Objects" In \N R.S. Michalski, J.G. Carbonell, and T.M. Mitchell (Eds.) "Machine\N Learning: An Artificial Intelligence Approach, Volume II". Los\N Altos, Ca: Morgan Kaufmann. \N \N Both of these papers describe a set of 10 trains, 5 east-bound and 5 west\N bound. Both refer to the same 10 trains as seen by the figures in these\N publications. The differences are:\N 1) This dataset has 10 attributes, no wheel, or load color attributes\N 2) Reference 2 (Stepp, Michalski): does not completely list the \N attributes used, but does mention wheel color - an attribute not \N present in this dataset.\N 3) Reference 1 (Michalski, Larson): 12 attributes mentioned, but only 6\N are explicitly described. These 6 are included in the dataset below \N and the Stepp and Michalski set.\N \N Results:\N [1] Michalski and Larson found the following decision rules:\N (1) There exists car1, car2, lod1 and lod2 such that\N [infront(car1, car2)][lcont(car1, lod1)][lcont(car2,lod2)]\N [load-shape(lod1)=triangle][load-shape(lod2)=polygon]=>[dir=east]\N (2) There exists a car1 such that\N [ln(car1)=short][car-shape(car1)=closed-top]=>[dir=east]\N (3) [ncar=3]v There exists car1 such that [car1(car-shape(car1)=jagged-\N top] =>[dir=west]\N There exists car1 such that \N (4) [#cars(ln=long)=2][cshape(car1)=open,trapezoind,u-shaped] v\N [location(car1)=2][cshape(car1)=closed, rectangle]=>[dir=west]\N (The first selector in rule 4 uses a meta descriptor generated by\N the program that counts the number of long cars in a train)\N [2] The goal of the cluster research is to develop a general method\N for clustering structured objects that can generate conjunctive\N descriptions that occur in human classifications or invent new\N concepts that have similar appeal. CLUSTER/S was able to find the\N following cognitively appealing clusters: 1) a) "There are two \N different car shapes in the train" b) "There are three or more\N different car shapes in the train" 2) a) Wheels on all cars have \N the same color, b) wheels on all cars do not have the same color."\N \N 4. Relevant information:\N - Additional "background" knowledge is supplied that provides a partial\N ordering on some of the attribute values.\N - We are providing this dataset both in its original form and in a form\N similar to the more typical propositional datasets in our repository.\N Since the trains dataset records relations between attributes, this\N transformation was somewhat challenging. However, it may shed some\N insight on this problem for people who are more familiar with the simple\N one-instance-per-line dataset format.\N - Hierarchy of values:\N if (cshape is one of {openrect,opentrap,ushaped,dblopnrect}\N then cshape is opentop\N if (cshape is one of {hexagon,ellipse,closedrect,jaggedtop,slopetop,\N engine}\N then cshape closedtop\N - Prediction task: Determine concise decision rules distinguishing \N trains traveling east from those traveling west.\N \N 5. Number of instances: 10\N \N 6. Number of attributes:\N - 10, not including the class attribute\N 1. ccont(train idx1, car idx2): car idx is contained in train idx\N 2. ncar(train idx): # of trains in car train idx (int)\N 3. infront(car idx1, car idx2): relative positions of cars in train\N 4. loc(car idx): absolute position of car in train (int)\N 5. nwhl(car idx): # of wheels of car idx (int)\N 6. ln(car idx): length of car idx (long, short)\N 7. cshape(car idx): shape of car (engine, dblopenrect, \N closedrect, openrect, opentrap, ushaped,\N hexagon, ellipse, jaggedtop, slopetop,\N opentop, closedtop) \N 8. npl(car idx): number of loads in car idx\N 9. lcont(car idx, load idx): description of which cars hold which loads\N 10. lhshape(load idx): description of load shape (trianglod, \N rectanglod, circlelod, hexagonlod)\N Class: direction (east, west)\N \N The following format was used for the "transformed" dataset representation\N as found in trains.transformed.data (one instance per line):\N \N Attributes: 33\N 1. Number_of_cars (integer in [3-5])\N 2. Number_of_different_loads (integer in [1-4])\N 3-22: 5 attributes for each of cars 2 through 5: (20 attributes total)\N - num_wheels (integer in [2-3])\N - length (short or long) \N - shape (closedrect, dblopnrect, ellipse, engine, hexagon,\N jaggedtop, openrect, opentrap, slopetop, ushaped)\N - num_loads (integer in [0-3])\N - load_shape (circlelod, hexagonlod, rectanglod, trianglod)\N 23-32: 10 Boolean attributes describing whether 2 types of loads are on\N adjacent cars of the train\N - Rectangle_next_to_rectangle (0 if false, 1 if true)\N - Rectangle_next_to_triangle (0 if false, 1 if true)\N - Rectangle_next_to_hexagon (0 if false, 1 if true)\N - Rectangle_next_to_circle (0 if false, 1 if true)\N - Triangle_next_to_triangle (0 if false, 1 if true)\N - Triangle_next_to_hexagon (0 if false, 1 if true)\N - Triangle_next_to_circle (0 if false, 1 if true)\N - Hexagon_next_to_hexagon (0 if false, 1 if true)\N - Hexagon_next_to_circle (0 if false, 1 if true)\N - Circle_next_to_circle (0 if false, 1 if true)\N 33. Class attribute (east or west)\N \N The number of cars vary between 3 and 5. Therefore, attributes referring\N to properties of cars that do not exist (such as the 5 attriubutes for\N the "5th" car when the train has fewer than 5 cars) are assigned a value\N of "-".\N \N 7. Distribution of classes:\N - There are 5 east-bound trains and 5 west-bound trains \N (i.e., 50% east, 50% west)\N \N\N Information about the dataset\N CLASSTYPE: nominal\N CLASSINDEX: last