What's new in Multifactor Dimensionality Reduction 3.0.2
Mar 14, 2013
- fixed bug where new batch/command line parameters such as '-discrete_significance_metric' were left disabled for non-experimental use.
New in Multifactor Dimensionality Reduction 3.0.1 (Mar 14, 2013)
- fixed bug printing permutation report when no originalAnalysis results passed in
- print permutation results with higher precision so permutations_on_cluster.py can match float comparison results
- fix bug looking up top model annotation info
- enhanced permutations_on_cluster.py to match output of running permutations on a single machine
- show cv results even on a forced search since now could be multiple or with wildcards
- change most text output from space separated to tab delimited
- change summary table header from testing to overall when no cross validation
- changed large_file_on_cluster to only do CV=1.
New in Multifactor Dimensionality Reduction 3.0 (Mar 14, 2013)
- Changed Top Models to show Overall Balanced accuracy rather than training. This allows huge speed up when doing cross-validation since only the top models need to be cross-validated.
- Added support for quantitative endpoints (Q-MDR)
- Added Network tab with network analysis methods
- Added support for wildcards in forced searches so you can see the best model which includes specified attribute(s)
- Added much more powerful API in org.epistasis.mdr.api.MDRExecutor
- Added filters multiSURF and multiSURFnTuRF
- Added elapsed/estimated timer to progress bars
- Extended the 'paired' analysis to also deal with triplets (e.g. case, control, control data). If a repeating pattern of status values is detected then the 'matched' checkbox (formerly called 'paired') will be enabled.
- Added new filter type: the ability to pass in a file. It will read in attributes from the file and filter the dataset to only contain those columns.
- Fixed bugs saving EPS graphs, saving text of network graphs
- Changed default Y axis of landscape line charts to dynamically set range rather than being set to 0 and 1
- Added warning upon data load if missing data encountered
- Improved feedback about mismatches between expert knowledge score file and dataset attribute names.
- Added new column to the Expert Knowledge 'View Scores' table. New column is 'current probability' and it updates as EDA searches update.
- Bug fixes.
New in Multifactor Dimensionality Reduction 2.0 Beta 8.4 (Aug 5, 2011)
- Fixed bug about type of units returned by TuRF. Explanation: SURFnTuRF, SURFStarnTuRF use zTransformed values. When they were
- implemented in 2.0 beta 4 2009-06-04, TuRF was also changed to use zTransformed scores. The difference is not great but
- the TuRF algorithm is documented as returning ReliefF scores so it is now being restored to return ReliefF scores.
- Fixed bug in batch mode where landscape table was missing a line break between table header and first row.
- Fixed bug in batch mode permutation output. The boundary p-values have been incorrect. The easiest way to see that
- this was wrong was to notice that at less significant p-values, the stated balanced accuracy was higher which
- makes no sense.
- Added ability to run analysis on more than one dataset in batch mode
- Added warning dialogs after loading an expert knowledge file if there were unmatched attributes.
- Added -all_models_outfile option to console version. If a filename passed in, MDR will output all models looked at with
- their training and testing balanced accuracy. Unlike landscape and Top Models this is written continuously rather than
- waiting until the end and therefore doesn't require massive amounts of memory.
- Updated BareBonesBrowserLaunch code to v3.1 as described here: http://www.centerkey.com/java/browser/
New in Multifactor Dimensionality Reduction 2.0 Beta 8.3 (Oct 27, 2010)
- Added support for batch mode to accept urls for dataset filenames
- Added simple API to do an analysis. org.epistasis.combinatoric.mdr.MDRExternalAPI
- In batch mode, dataset can be a local file path or a url.
- Fixed nullPointerException when reading in a saved analysis before any analysis had been run.