Improvements and new features:

• Eliminated directhash function (now internal to directcount functions)
• Added Ma entropy estimate
• Added bootstrap variance estimate
• Made shuffling and jackknifing scripts more efficient
• Added staversion function
• Added D^interval[q] in metric space method

There are few incompatibilities with earlier releases:

• The option variance_estimate_method is now called variance_estimation_method.
• The "single q" and "all q" metric space method functions have been combined. The single q/all q choice is now specified with the option parallel.
• Function metric's output structure has been modifed. The member m_list is now called categories.

• Now requires the GNU Scientific Library
• Added Best Upper Bounds method to entropy estimation methods
• Fixed bug that was suppressing diagnostic messages on Windows and Mac

• Explicit bias correction methods eliminated, now listed as separate entropy estimation methods
• Introduced general and method-specific variance estimation methods
• In earlier versions, the jackknife and bootstrap variance estimates in all estimate structures were obtained using the plug-in method instead of the method named in the estimate structure. This has been fixed.
• Added Chao-Shen entropy estimation method
• Added Wolpert-Wolf entropy estimation method
• Code reorganized to facilitate external contributions
• Contribution instructions added

• Minor bug fixes
• Addition of staverify.

• Minor bug fixes
• Minor corrections to the project repository

New features:

• Added NSB entropy estimate, related parameter nsb_precision, and its specific variance method nsb_var.
• Added global entropy option possible_words, which applies to methods bub, nsb, tpmc, and ww, and supercedes options bub_possible_words_strategy, tpmc_possible_words_strategy, and ww_possible_words_strategy (these latter three options are now deprecated).
• Added error checking to metric.m and binless.m to prevent bug in deprecated options (i.e., *_possible_words_strategy as noted above).
• Added additional output to the entropy estimate field in the output hist1d data structure.
• Added information parameter letter_cap.

Bug fixes:

• Fixed bug in Matrix2Hist2DComp.c in which the joint histogram in the output hist2d structure was miscalculated for some matrices when unoccupied_bins_strategy=0. This error effected any matrix in which the number of occupied columns is greater than the number of occupied rows.
• Fixed bug in DirectBinComp.c that added an extra bin to all binned data when n*(words_per_train*counting_bin_size)=(end_time-start_time) for any positive integer n. This resulted in an incorrect value of N (number of sub-words or letters) in the output hist1d structure. In certain situations, this bug may slightly alter entropy calculations in which the options bub_possible_words_strategy, tpmc_possible_words_strategy, or ww_possible_words_strategy depend upon the value of N. An option legacy_binning was added to recover this functionality, if desired.
• Fixed bug in directbin.c that permitted invalid values of words_per_train, and improved error checking.
• Fixed bug in entropy_bub_mx.c that resulted in parameter bub_lambda_0 being passed to the output options structure as an integer, rather than a double. This did not effect the calculation itself, but could lead to a misrepresentation of its value in user results.
• Fixed bug in entropy_ww_mx.c that resulted in parameter ww_beta being passed to the output options structure as an integer, rather than a double. This did not effect the calculation itself, but could lead to a misrepresentation of its value in user results.
• Fixed bug in DirectBinComp.c that caused a segmentation violation when options permute_spike_trains=sum_spike_trains=1.
• Fixed bug in entropy_mx.c that prevented writing options to the output structure correctly when the value was changed internally. This bug had no outwardly noticeable effect.

Minor improvements and changes:

• Improved argument checking in hist_mx.c, directcountcond.c, and infocond.c.
• Made modifications to allow for integration of C++ code.
• Improved make.m to compile files only when necessary (in the style of a true makefile), and provide additional options.
• Updated documentation, demos, and related files to fix typos and reflect the aforementioned changes.

New features:

• Added information method ctwmcmc to compute formal information via the context-tree weighting technique. In the process of porting from the original Fortran code (CTPACK-2.7.1) fixed two minor bugs in the calculation of the weighted codelengths (Lw) and the relative weightings (see CTWMCMCTreeComp.cpp for details).

Bug fixes:

• Fixed bug in binlessinfo.c that inadvertently changed the second input argument COUNTS. This bug did not effect any of the demonstrations.
• Fixed bug in MatrixToHist2DComp.c that caused Matlab to crash when binlessinfo was called with a very small dataset, in which any stratum was empty.
• Fixed bug in directcount*.c that ignored whether input was int32 or not. This did not cause Matlab to crash, but rather returned bad data. This would only be a problem if you tried to create your own input for these functions, and failed to make them int32. Use of the STAToolkit functions that create the required input to this function (e.g., directbin) avoids this bug.

Minor improvements and changes:

• Improved argument checking in gen_mx.c and directcondformal.c.
• Improved error handling in entropy_nsb_c.cpp.
• Created additional options for make.m, including information method specific compilation (e.g., make metric).
• Updated documentation, demos, and related files to fix typos and reflect the aforementioned changes.

New features:

• Functions staread and stawrite can now handle continuous data (e.g., LFP).
• The binless method functions (binless, binlessopen, binlesswarp, binlessembed, binlessinfo) can now be used to analyze continuous data.

Minor improvements and changes:

• Fixed minor memory leak issues in all gateway routines.
• Updated documentation, demos, and related files to fix typos and reflect the aforementioned changes.

New features:

• The toolkit will compile and run successfully in Octave.

Minor improvements and changes:

• Improved make.m to increase compilation speed and efficiency.
• Updated documentation, demos, and related files to fix typos and reflect the aforementioned changes.