History of QUEST release March 31, 1998 --------------- QUEST 1.7 release April 07 update --------------- QUEST 1.7.1 1. bug related to the format of quest.in for minns is fixed. 2. fix bug related to F90 "where" command. 3. fix bug when all values for a categorical variable are missing. 4. the output format for numerical variables is changed. 5. fix bug related to imputation of categorical values which does not appear in the learning sample. 6. SUN version compiled with Nag F90 is available. June 22 update -------------- QUEST 1.7.2 1. The name of the data file can be surrounded by quotes. 2. Linux version for kernel release 2.0 and later is available. 3. change the number for the current year from 2 digits to 4 digits. 4. solve the problem when an end of record appears in input files. 5. exclude class labels which appear in the test sample but not in the learning sample to avoid program fail. 6. fix a bug in LaTeX tree generator when mixed data type is used. 7. fix bugs in routines of handling missing values. 8. add an option of creating a terminal id file for each case in the learning sample. 9. fix the problem related to memory leakage in CV. 10. update the manual. 11. SUN version compiled with SUN f90 V1.2 is available. 12. fix deallocation bug in the twoing procedure. 13. fix a bug in printing the coefficients for linear splits. Auguest 27 update ----------------- QUEST 1.7.3 1. modify the output format of the latex tree program. 2. change the output format to improve the interface. September 9 update ------------------ QUEST 1.7.4 1. fix a bug related to the computation of F values. 2. fix a bug in the goodness of split when there are only 2 classes and categorical splits are used. September 13 update ------------------ QUEST 1.7.5 1. fix a bug related to finding the best categorical split when exhaustive search is selected. 2. modify the user interface so that it prompts for the name of file that stores all the selected options (quest.in in the previous release). September 26 update ----------------- QUEST 1.7.6 1. set F-stat=0, if df for error is 0. 2. modify the latex tree routine so that it uses \shortstack command. 3. change the user interface such that the program prompts all the questions before building the tree. 4. fix a bug in two classes problem when exhaustive categorical split is used. October 28 update ----------------- QUEST 1.7.7 1. change the interface for the batch mode. 2. fix a bug in computing the misclassification cost when unequal costs option is selected. 3. fix a problem related to SUN machines when reading the description file. November 19 update ----------------- QUEST 1.7.8 1. option of reporting test sample estiamte of misclassification cost is added. 2. option of creating allCLEAR code is added. 3. fix a bug related to linear split on large data set. 4. fix a bug in reading file when batch mode is used. 5. fix bugs related to allocatable array which is not deallocated in routines performing linear and CART categorical split. The bugs show when SUN version is used. December 12 update ----------------- QUEST 1.7.9 1. fix a rounding error problem when goodness of split is used. 2. routines of checking the format of data files are added. 3. add a routine to print out the label of test sample used in misclassification estimate. 4. the variable type 'w' is changed to 'f' and weights have to be integers. 5. some routines are modified so that the program uses memory more efficiently and as a result it runs faster. 6. fix bugs in test sample estimates when missing values appear. 7. fix a bug in class assignment. February 3, 1999 update ----------------------- QUEST 1.7.10 1. fix bugs in CV costs estimation when estimated priors are used. 2. fix a bug caused by rounding errors when equal or given priors are used. 3. remove a line to avoid dumping CV trees. June 30, 1999 update ----------------------- QUEST 1.8 1. include the divergence family and the MPI criterion as part of splitting criteria option. The default option is changed to likelihood ratio G^2. 2. use the big_integers module from Unicomp, Inc. to increase the number of categories from 31 up to 166 so that exhaustive search on these variables are possible. 3. modify the source code so that the pointer works properly on Linux machines. 4. print the percentage of missing values in the learning sample. July 21, 1999 update ----------------------- QUEST 1.8.1 1. fix a bug in computing Gini values when there are missing covariates. August 10, 1999 update ----------------------- QUEST 1.8.2 1. add a subroutine of drawing LaTeX2e tree using pstricks package (v97 patch 9 <1998/01/07>). October 11, 1999 update ----------------------- QUEST 1.8.3 1. modify the options used in the pstricks package. 2. increase the stack size to 4000000 to avoid possible stack overflow under Win32. November 3, 1999 update ----------------------- QUEST 1.8.4 1. change the default value of minimal node size from 5 to max(5,n/100), where n is the total number of observations. March 1, 2000 update ----------------------- QUEST 1.8.5 1. fix a bug in the divergence family when lambda=0. 2. change the wording for terminal nodes. 3. variable type is checked for its legitimacy. 4. modify the output format for LaTeX tree generated by using pstricks package. June 1, 2000 update ----------------------- QUEST 1.8.6 1. modify the splitting criteria to eliminate possible selection bias toward variables with more missing values. (not complete yet!!) June 30, 2000 update ----------------------- QUEST 1.8.7 1. output the CRIMCOORD value associated with each categorical variable when linear split is used. 2. output the confusion matrices. August 31, 2000 update ---------------------- QUEST 1.8.8 1. make change in the output format of Pstricks LaTeX codes. 2. fix bugs in the output format of confusion matrix and Pstricks tree, when number of classes is more than 10. 3. a subsection related to library files is added in the manual. January 05, 2001 update ----------------------- QUEST 1.8.9 1. output the class mean or mode of splitting variable for each intermediate node. These values are useful for channeling cases with missing values. 2. change confusion matrix to classification matrix and modify its ouput format. February 08, 2001 update ------------------------ QUEST 1.8.10 1. fix a bug related to linear splits when all predictors are ordered variables. 2. add the warrenty statement. May 04, 2001 update ------------------------ QUEST 1.8.11 1. print out the total number of categories for each categorical variable. 2. fix a bug in formating the classification matrix and LaTeX trees. September 18, 2001 update ------------------------ QUEST 1.8.12 1. correct the Gini criterion when missing values occur. 2. change the output formats for priors. 3. output the number of classes on screen. November 02, 2001 update ------------------------ QUEST 1.8.13 1. sort the values in alphabetical order when exhaustive search for categorical split is used. 2. set exhaustive search as the default option for split point selection when there are only 2 classes. 3. set the Gini criteria as the default option when exhaustive variable selection is chosen. December 02, 2001 update ------------------------ QUEST 1.8.14 1. output the number of terminal nodes on screen. 2. modify the output format for misclassification costs. 3. fix a bug in the routine outputs terminal nodes information for data files containing missing class labels. 4. add a check on empty category and empty class label. December 06, 2001 update ------------------------ QUEST 1.8.15 1. fix a bug related categorical split point when CRIMCOORD option is used. 2. fix a bug in the routine outputs the categorical split points. December 16, 2001 update ------------------------ QUEST 1.8.16 1. The mean or mode of the splitting variable do not appear in the ouput when class size is zero. 2. Modify the QDA method when one of the superclass has zero variance. The algorithm is in the following. Suppose the c1 values are all equal to a (i.e., zero variance). Let x_1, x_2,..., x_n be the order statistics for superclass c2. (a) If x_1 < x_2 ... x_i <= a < x_{i+1} < ... < x_n, then: (i) If x_i < a, choose between the 2 split points 0.5*(x_i + a) and 0.5*(x_{i+1} + a). (ii) If x_i = a, choose between the 2 split points 0.5*(x_{i-1} + a) and 0.5*(x_{i+1} + a). (b) If a <= x_1, then: (i) If a = x_1, split at 0.5*(x_2 + a). (ii) If a < x_1, split at 0.5*(x_1 + a). (c) If x_n <= a, then: (i) If x_n = a, split at 0.5*(x_{n-1} + a). (ii) If x_n < a, split at 0.5*(x_n + a). 3. Entry for predicted class is added in tree display. April 17, 2002 update ---------------------- QUEST 1.8.17 1. Fix a bug when a carriage return is used as a file name. 2. Change output format for the means at each terminal node. 3. IBM workstation version is added in the list. May 6, 2002 update ---------------------- QUEST 1.8.18 1. A format statement is added to fix a bug related to the routine handling frequency variable. 2. A routine is added so that the program pauses when current job is finished. 3. Fix a bug when batch mode and a new output file is used. January 08, 2003 update ---------------------- QUEST 1.8.19 1. Fix a bug in printing the coefficients for linear splits at each node. June 16, 2003 update ---------------------- QUEST 1.8.20 1. Fix a bug in printing the specified alpha value. 2. Change the statement in the caption to the latex output. (The value beneath...) 3. add an option for test sample estimate when test sample pruning is chosen. 4. Change the output format for the split point at each intermediate node. April 27, 2004 update ------------------------ QUEST 1.9 1. Allow LaTeX tree be displayed in landscape mode when the tree size is larger than 10. 2. Use ".txt" for the description and data file name suffixes. Feburary 18, 2005 update ------------------------ QUEST 1.9.1 1. Modify the option for generating tree codes. 2. Fix a bug in Pstricks LaTeX output. LaTeX tree is always displayed in landscape mode. 3. Fix a bug in batch mode output, when test sample pruning is used. June 23, 2005 update ------------------------ QUEST 1.9.2 1. Remove the TreeTeX option. 2. Modify the statements when the dependent variable has missing values. 3. Modify the output format for LaTeX tree.