rattle (2.5.36) unstable; urgency=low * Add Bill Venables' searchCRAN example code. * Improve error message when we find duplicate variable names in a loaded file, which might result when there is no header line. * Add help item for Projects. * On Evaluate with supplied file, use the hdr specified on the Data tab. -- Graham Williams Mon, 12 Jul 2010 06:43:06 +1000 rattle (2.5.35) unstable; urgency=low * Add utility lss function to list object sizes. * Add options text entry for SVM to easily allow other options. * Better formatting of the Log tab. * Use a set.seed for SVM to ensure same model each time. * Add option to random forest to impute missing values rather than simply ignoring the observations. * On Evalaute with supplied file, use the sep specified on the Data tab, thus allowing TXT files. * On loading a new dataset for evaluation be sure to add in any missing columns, and unify the levels. * Improve binning documentation. * Make RGtk2, cairoDevice, colorspace all dependencies so we can get rattle started and then rattle will prompt to install other packages that are mssing when it needs them. -- Graham Williams Thu, 01 Jul 2010 15:34:50 +1000 rattle (2.5.34) unstable; urgency=low * When a package is missing, there is now the option to install it right then, and it continues as normal after it gets installed. * Change Suggests to Depends so all used pacakges get loaded on loading rattle, in an attempt to make it easier to install Rattle. Then the r-cran-rattle package on Debian/Ubuntu will have all required dependencies and a normal install.packages will get all dependencies also, rather than having to use dependencies=c('Depends', 'Suggests'). Penalty is it takes 20 seconds to do 'library(rattle)' on a server and 90 seconds on a netbook - so revert back to not doing this. * Ensure the new train/validate/test scneario is saved across projects. -- Graham Williams Wed, 09 Jun 2010 07:04:08 +1000 rattle (2.5.33) unstable; urgency=low * Bug fix rf.cmd. * Improve scoring functionality: The dataset can have NA's for target, and these can now get scored by rf on Evaluate tab. Loading a CSV file to be scored no longer needs to have the target column included (previously it needed to be there and have non-NA values). Thanks to Chris Snijders. -- Graham Williams Mon, 31 May 2010 06:22:54 +1000 rattle (2.5.32) unstable; urgency=low * Remove dependency on car - not actually being used at the moment. * For random forest, allow sample size text entry as a single integer or a list, as per randomForest. * Use na.omit with cforest, as is done with randomForest. * For randomForest turn subsampling with replacement off since it is more likely to produce biased importance measures, as explained in by the cforest papers. * Fix bug with multiple "contact support" lines in error popups. * When showing the randomForest importance values, sort on the accuracy measure rather than the Gini measure, since the Gini is biased in favour of categoric variables with many categories. * ada boost seed should be 42, like all other seeds. * Tidy up some ada output. * Bug fix - save project for rf failing (looking for rf_sampsize_entry). * Remove text from toolbar by default. * Change order of Forest/Boost buttons on Model tab. * Add tooltips for all toolbar buttons. -- Graham Williams Fri, 28 May 2010 15:47:15 +1000 rattle (2.5.31) stable; urgency=low * Add rattle.info() to list information for debugging purposes. * Bump pmml to 1.2.22 * Fixes from wenching.lin@zementis.com: Extension in Header should be first element. Coefficients in regression models should not be NA (as will be for singularities), but replace with, and so no impact of change. * Ensure Survival defaults are reset appropriately. -- Graham Williams Wed, 19 May 2010 09:50:39 +1000 rattle (2.5.30) stable; urgency=low * On MS/Windows with Japanese, read.csv needs encoding option set with file rather than with read.csv (for UTF-8) but seems okay under other scenarios. * On MS/Windows with Japanese (UTF-8) the encoding of the variables selected for transforming needs to be UTF-8 for much of the process, but "unknown" when using Rtxt and sprintf (when substituting the variable names) to ensure resulting message is correctly matched for encodings. -- Graham Williams Wed, 19 May 2010 09:47:12 +1000 rattle (2.5.29) stable; urgency=low * Add the translation file. * Fix an Encoding/sprintf issue for Japanese on MS/Windows. * Allow crv$NOTEBOOK.MODEL.NAME to be overridden by other packages (RStat). * When dispatch fails be sure to include the Tab label on which it fails. * Ensure HClust Options are re-enabled on loading a project. -- Graham Williams Sat, 24 Apr 2010 07:32:02 +1000 rattle (2.5.28) stable; urgency=low * Minor format changes for glm and rf model output. * Capture additional survival model error and suggest a solution. * Remove spurious additional plot for Survival Residual plot. * Update log tab labels to be more generic. * Update tooltips to be generic and add survival tooltips. -- Graham Williams Thu, 22 Apr 2010 06:21:58 +1000 rattle (2.5.27) unstable; urgency=low * Further translation fixes. In particular, use Encoding(...getText()) <- "UTF-8" to ensure strings from the GUI ate UTF-8, and not unknown. * Ensure training dataset rather than sample dataset nomenclature is now used. * Ensure execute button can only be clicked once while it is processing. * Survival plot buttons need to be made sensitive as appropriate. * For Japanese on MS/Windows do not use monospace font since this ends up vertically cenbtering periods and commas (and all other characters). Need a fixed width font that does not do this, but for now we put up with variable width font. * Revert to using only English for all hidden tab labels. * Improved identification of current plot number. * Bug fix multiple vars selected for asnumeric and ascategoric transforms. -- Graham Williams Thu, 22 Apr 2010 06:17:20 +1000 rattle (2.5.26) unstable; urgency=low * Add Cross Tab option to Explore tab to generate cross tabulations of each categoric variable by the target variable. (Luke Lake) * Bug fix - improve how we obtain the plot number from the title, particularly in the context of translations. * Further translation markup. * Clean up the use of dfedit. * Minor improvement to spacing in Log tab. -- Graham Williams Tue, 30 Mar 2010 21:37:29 +1100 rattle (2.5.25) unstable; urgency=low * Start using the RGtk2DfEdit for the View and Edit buttons of the Data tab, and the Enter/Score option of the Evaluate tab. RGtk2DfEdit provides a spreadhseet like interface to the data. Various data editing options are available. Also press = to run an arbitrary R command on selected data (e.g. select two columns of data and issue the plot command). * Add further markup of text for translations. * Support specification of the character used for decimal points (to suit some European usage). * Fix bug in exporting XML - replace & with & * Survival plots - split survival chart plot from residuals plots, and plot all residuals. * Fix logic behind what is greyed out in the Test tab. -- Graham Williams Mon, 29 Mar 2010 19:37:25 +1100 rattle (2.5.24) stable; urgency=low * Revamp the help text, and put into the Rtxt translation framework. * Fix the height of the data name widget (the library option was growing the width for some reason). * For Evaluate, add Full and Enter as dataset options. Enter will pop up an editor with the final row from the dataset, allowing you to add rows or modify the supplied row. We supply the row so that we have an example to work from. Full uses the whole original dataset. -- Graham Williams Sat, 06 Mar 2010 14:17:12 +1100 rattle (2.5.23) stable; urgency=low * Catch "arules" error in converting data to transactions when baskets contain repeated items. * When data tab is executed, and so crs$rpart is reset to NULL, always remove the Draw/Rules button from the Tree option of the Predictive tab. * Add code to fix translations that are not being loaded when using RGtk2 on MS/Windows. All is okay on GNU/Linux, but RGtk2 seems not to get the right locale for loaded Glade file. The fix is to traverse the GUI and change all labels, on starting up Rattle. RGtk2 authors tried to fix but it remains an issue. * Ensure rpart is reset on resetting rattle. * Rework handling of tab pages because a Japanese translation on MS/Windows is having issues with the following call (nd=notebook) nb$getTabLabelText(nb$getNthPage(nb$getCurrentPage())) returning what looks like Shift-JIS encoding of the string rather than UTF-8, and hence not string matching the expected tab label. * Fix spelling errors on help menu and ensure help for all topics is covered. * For nnet, use MaxNWts=10000 (default is 1000) to allow larger nets by default, and capture the error message when this is exceeded and better explain what to do. * Ensure we don't export an empty dataset when choosing export on the data tab. * Capture arules error message when there are repeated items in one basket, and explain this more clearly. * For rpart use information as the default split rather than Gini - makes little if any difference. * Allow showHelpPlus to have an extra/alternative question that is displayed. * All random seeds should be 42. * Reset kmeans tab on loading a project. * Add dozen more weather stations to the weatherAUS dataset. * Improve the logic for the display of the Report radio buttons on the Evaluate tab. * Spelling correction to a number of tooltips. -- Graham Williams Wed, 03 Mar 2010 06:50:58 +1100 rattle (2.5.22) stable; urgency=low * Default window height is 650, but not forced so that the window nicely fills a netbook screen if maximised. * Bump R dependency to 2.8.0 in line with update of the CITATION file. -- Graham Williams Sat, 13 Feb 2010 09:48:00 +1100 rattle (2.5.21) stable; urgency=low * Re-enable gettext on MS/Windows, even though RGtk2 2.12.18 has not fixed the bindtextdomain problem with glade files and package supplied translations. * Change the tree plot to us "< =>" and ">= <" to clearly identify which branch the "=" results go. Could not figure out how to get expression to us a "ge" symbol. * Improve formatting of the PvO plots. * Use the pairs.panels function from the psych package for the default scatterplot on the Explore tab. * Add INSTALL file. -- Graham Williams Sun, 07 Feb 2010 15:03:22 +1100 rattle (2.5.20) stable; urgency=low * Restore missing weather.csv file. * Add to Google code: weather.R ChangeLog NEWS ToDo upload_uwe.sh upload_cran.sh. -- Graham Williams Sun, 31 Jan 2010 11:07:55 +1100 rattle (2.5.19) unstable; urgency=low * Ensure the right labels (Time/Risk rather than Class/Prob) displayed in filechooser when exporting a survival model. * Model tab renamed as Predictive. * Ensure boxplots have same "by ..." in the main title. * Update the weather dataset and include many more weather stations in the weatherAUS dataset. * Rtxt does no translations when running on MS/Windows (for now). -- Graham Williams Sat, 30 Jan 2010 09:28:18 +1100