changes in glmlab From 2.3 to 2.4 * Changed plotwork as the residual vs Fitted Values plot still had no Menu Bar, when all other plots did. 12 July 1999 * Changed structure of Web Site, and made appropriate changes to glmlab to reflect these. 26 February 1999 * Changed some aspects of Loading Data Files that should make it quicker. 08 February 1999 * Made startup smoother and quicker. 05 February 1999 * Added Tooltips. 05 February 1999 * Made minor changes to problems encountered with MATLAB v5.2 05 February 1999 * Fixed a bug that appeared in MATLAB v5.2 where MATLAB complained about problems with a COLON. Thanx J-MF. 04 February 1999 * Fixed a problem that appeared in MATLAB v5.2 with loading models. This made things better anyway. 04 February 1999 From 2.2 to 2.3 * Fixed a bug when loading models that sometimes ignored saved information. 02 November 1998 * Fixed a bug in Saving Models that added a .glg extension even when it was supplied, despite what the documentation said. 02 November 1998 * Fixed a bug in alias that sometimes meant aliasing of variables was reported on an initial fit only when this was not the case. 02 November 1998 * Fixed some annoying things (not quite bugs) that occurred when the inputs were not quite right; now error messages should be fewer. 31 August 1998 * Fixed a bug: sometimes FIT MODEL was disabled, and could only be enabled by adjusting the y-variable, even when there was no problem. 20 August 1998 * Added some assistance when glmlab falls over on startup so that the user knows that dummylog.m must be in a writeable directory. 20 August 1998 * Fixed a bug with loading models; the DETAILS file path was hard coded (!). 20 August 1998 * Fixed a bug in findvars that made an appearance in some version of glmlab (Noticeably Windows version). Thanx J-MF 06 August 1998 * Fixed a bug using findres that became apparent in log-linear models. Thanx Jean-Marc Fromentin. 30 July 1998 * Fixed a bug in addpmenu that made the whole thing freeze in some circumstances. 29 June 1998 * Made some changes to glmlab.m, menuwork.m and addpmenu to speed glmab up slightly, especially on starting. 29 June 1998 From 2.0 to 2.2 * Fixed a bug in saving models when the glg extension wasn't given. 21 April 1998 * Removed saveseed and loadseed to use the improved rand. 19 September 1997 * Fiddled to fix some errors with the plots uimenu. 05 September 1997 * Added the capacity to save and load models--the workspace and settings (in GLMLAB_INFO). 26 August 1997 * Allowed matrices to be entered into the Covariates window, though it is still *strongly* unrecommended. 26 August 1997 * Added Labels, Gridlines, Marker Attributes and Axis Limits uimenus to the residual plots prettying the plots. 13 August 1997 * A new output display control: toggle the variable information. 01 August 1997 * A new help window: OUTPUT VARIABLES 01 August 1997 * Some tidying of internal workings, including two new files, extcrtgl.m and resetgl.m. 01 August 1997 * Now returns the names of the X-variables and the X-matrix as the variables XVARNAMES and XMATRIX (thanx to Ross Darnell) and separates the parameter estimates (BETA) and their standard errors (SERRORS). 18 July 1997, 02 March 1998 From 1.7 to 2.0 The changes from version 1.* to 2.* also includes changes for running glmlab under MATLAB 5. * Fixed a bug that meant the DETAILS file was a bit odd. Very odd really. 8 July 1997 * glmlab now removes occurences of the same string in the X-variables window. 4 July 1997 * Moved Residuals Plots from being a button to a menu item, and added NEW MODEL as a button. 11 June 1997 * In interactive variable entry, only allowed `double' variables appear in the variable list of choices. 29 May 1997 * Introduced interactive variable entry (apart from interactions...) No doubt this will need to be tested more thoroughly. 28 May 1997 * Changed the X Variable Edit window to accept multi-line strings. 22 May 1997 * Made some much needed changes to the distributions files in /dist 15 May 1997 * Added the linear predictor as a returned variable. 15/05/1997 * Changed interact.m to use varargin so that the number of interaction terms in one interaction is unlimited. 09/05/1997 * Changed fhisto.m so that an error when using some starting points caused the x-axis numbering to go wonky. 07/05/1997 * Altered fac.m so that fac-ing a factor with missing levels works OK. (ie fac([1 2 6]) assumes 6 levels, not three as it used to). 01/05/1997 * Made more room for parameter estimates in output. 01/05/1997 * Provided error windows for user defined links and distributions when the files don't exist. 01/05/1997 * Provided a separator for user defined links (as for distributions). 01/05/1997 * Removed an annoyance in DEVLIST that showed the initial condition. 30/04/1997 * Minor changes to plotwork to pretty Inv Gsn and Binoml plots 30/04/1997 * Moved errors trapping in glmfit to a new file, editerrs. 28/04/1997 * Altered lpower so that zeroes don't cause the log's to fail. 28/04/1998 * Made some minor changes to glmlab.m now that the new on-line manual exists. 23/04/1997 * glmlab now remembers the last number entered into the scale parameter and power link function input dialogs. 23/04/1997 * Dimmed the FIT SPECIFIED MODEL button when there is no legit model to fit. 21/04/1997 * Moved some files from one directory to another (some were in some illogical places). 20/04/1997 * Added a demo in the Help menu that runs a demo on the glmlab screen. 19/04/1997 * The fitting parameters can now be changed through a dialog box in the Options menu. 18/04/1997 * Added the inverse Gaussian distribution to its box of tricks. 17/04/1997 * Fixed a bug that crept in so that changes in deviance and df wasn't displayed. 17/04/1997 * Fixed the output display that got confused on the first iteration if a power link function was used. 17/04/1997 * Added dialog box exit on a system window close with CloserequestFcn to glmlab.m and glmplot.m 14/04/1997 * Made a warning status option for turning off the annoying warning messages. 13/04/1997 * Changed glmlab.m so that only one copy of glmlab could be run at any one time. A little tricky, and not ideal, but it was difficult (for me anyhow) given the HandleVisibility property is on callback. 12/04/1997 * Changed just about everything and every file to use the UserData property. This hides glmlab's internal information from the user (so they don't delete or clear it accidentally (or otherwise :->) and foul it all up completely). 11/04/1997 * Changed HandleVisibility properties to callback in glmlab.m, glmplot.m and help windows. 09/04/1997 * Turned off RESIDUAL PLOTS... button after making important structural changed to the model. (A warning still ensues after (possible) variable changes, as sometimes the routines are called incidentally.) 09/04/1997 * Minor wording changes in glmplot.m. 09/04/1997 * Some more fine-tuning of the new fixmeup.m 09/04/1997 * Minor changes to opterr. 09/04/1997 * Some more fixing of fixmeup which failed when glmfit auto-added the transpose character. 02/04/1997 * Turned the watch off after error messages displayed. 02/04/1997 * Fixed rhisto and fhisto to work in MATLAB 5--some minor changes (like XTickLabels to XTickLabel and so on). 26/03/1997 * Distributions for which quantile residuals cannot be found (ie user- defined) issues a dialog box warning. 25/03/1997 * Fixed glmfit so that a non-finite dispersion parameter issues a warning, and hides the residuals plots button (since residuals cannot be calculated). 25/03/1997 * Ironically, had to fix up fixmeup after the change made below to remove the [ and ] that appeared in the variable names. 25/03/1997 * Fixed up newmodel which somehow acquired a bug while including the GLMLAB_INFO_ variable. 25/03/1997 * When plotting residuals vs covariates, the window asking for the covariates doesn't pop up when there is only one covariate--it just uses it. (This behaviour was already part of the last version; the recent fiddling, however, failed to do this.) 24/03/1997 * Changed the plotting window so that Transformed Residual are no longer an option when using the normal distribution (there is no transformation is that case). 24/03/1997 * glmlab now reports an error when trying to fit a model with no co-variates and no constant term. (It used to shrug, and decide to fit a constant term anyway). 24/03/1997 * Used cells to hide most of the internal glmlab variables (like the error function, link, etc...) but stashing them in the UserData variable GLMLAB_INFO_, making who and whos much cleaner (which is great, of course, when trying to see what vars are loaded). 24/03/1997 * Made sure the Residual plots didn't have plots ending up on the axis and hence sometimes obscured. 23/03/1997 * Fixed a bug in residual plotting that only plotted the first replicate of residuals in a quantile equivalents plot. 23/03/1997 * Added a dialog box to indicate that files have been successfully loaded. 23/03/1997 * When a binomial dist is specified but the response variable is not set up for it, glmlab would use the binomial, but not check the binomial dist and logit link in the uimenus (though they were permanently set). Likewise when a non-binomial is specified but the response is set for binomial. Now, a dialog box appears indicating the change, and the new distributions and links are checked. 22/03/1997 * Fixed some bugs in the residual plotting when the binomial distribution is used. 22/03/1997 * Errors appearing in opterr now appear in dialog boxes. 22/03/1997 * The variables when entered no longer appear with [ and ] enclosing. Was a dumb idea to begin with really... 22/03/1997 * When the input variables are different lengths, dialog boxes now show the error message. 22/03/1997 * On asking for a new model, a dialog box appears giving the user a chance to bail out. 22/03/1997 * Fiddled with the pushbuttons on the main glmlab window, adding a CLOSE button (so it looks a bit more like other windows), and hence needing to shift and re-size the FIT and RESIDUALS buttons. 22/03/1997 * Added a question dialog box so that the user can bail out after electing to quit and lose settings. 22/03/1997 * Added a plot to the RESIDUALS PLOTS menu--fitted values without the constant information transformation. 21/03/1997 * Changed npplot so that the legend didn't obscure the title (this only occurred with MATLAB 5 for some reason). 21/03/1997 * Changed the plotting routine so that it asks nicely for which covariates to plot against residuals, rather than guessing from the first five observations. 21/03/1997 * Fixed minors bug in plotting routine: always labelled the covariates as Covariate(1) whatever the number should have been (hmm....); fixed a bug that crept in when randomised quantile residuals were calculated (replicates were not asked for). 20/03/1997 * Re-ordered res vs covar plot so that the replicates were repeated first before the covariate. 20/03/1997 * Fixed a minor bug in plotting: Residual vs Response when y~Binomial would plot twice--once for observed counts and once for sample size. Now plots the ratio, and is labelled accordingly. 20/03/1997 * Changed things (I hope!) to be MATLAB 5 complaint. 19/03/1997 * Now the power link function shows the power used (don't know why it took so long...!) 19/03/1997 * Added access to on-line manual from help menu, and to MathWorks and glmlab Web pages from appropriate help windows. 19/03/1997 * Made some of the help windows smaller and better using textwrap. 19/03/1997 * Placed help and demo files in their own directory: glmhelp 19/03/1997 * Changed/added dialog boxes using MATLAB 5 capabilities. 18/03/1997 From 1.6 to 1.7 * Changed the way paramtrs handles the parameter values. Now, the current values are written to a file named PARAVALS.mat and stored in the top directory. This now allows the user to more easily change parameters on the run without editting the parametrs.m file. 06/02/1996 * Changed the plot directory to plotting to stop the double-up with the MATLAB function by the same name (made help plot ugly, and besides, it was a stupid thing to do in the first place). * glmlab now tells you that it's thinking by printing dots on the screen as different routines and loops are called. * Made some minor changes to some files for efficiency. From 1.5 to 1.6 * Problems with interact.m were dealt with. The algorithm now also handles interactions of up to 12 variables (though a small changes to the function line changes it to more if needed). Thanx to Janek .... * Added accelerators to some menu items. * Deleted a lot of files not used any more. * Made some functions more efficient. ------ Some broad changes to earlier versions: From 1.3 to 1.4 * Some changes to the files for efficiency reasons. * Also cleaned up the interface a little more. From 1.2 to 1.3 * Changed the graphical user interface to use pull-down menus. This cleaned up the interface, and made it less cluttered. From 1.1 to 1.2 * Changed the command line interface to a graphical user interface. Each option (distribution, link, ...) has a button to press giving options from which to choose. From 1.0 to 1.1 * Changed the GLIM-like command line interface to a prompted command line interface.