This file lists notable changes to Tetrad IV since version 4.3.6.

4.3.6

* Rendered Bayes updater interface less clunky. Fixed the problem of category
buttons not shading correctly in Java 1.5 and Macs. Added a notification to let
the user know when they're trying to update a Bayes IM that's not completely
specified.

* Repaired the help button for searches so that it actually opens up the help
file for those searches. Corrected help files for most searches.

* Lined up columnar output in regression editor for platforms in which it was
not lined up.

* Specified null hypotheses for SEM chi square tests specified where used in the
interface (SEM estimator, parameters tab of pattern searches).

* Removed some items from the interface that weren't working correctly: Logistic
regression, Structural EM, CEFS. (These will be re-added when the underlying
issues are fixed.)

4.3.6-1

* Added this change log to the downloads page.

* Improved the discretizer some. If no variables are selected, all variables are
discretized. The displayed number of categories is now synced with the number
of bins displayed for discretizing continuous data.

* Changed "unfaithful" to "ambiguous" for CPC, to reflect the fact that for
finite sample sizes is detecting (and underlining) examples of near-unfaithfulness,
not necessarily unfaithfulness per se. (Made similar change to CFCI.)

* Fixed knowledge in FCI so that if X<-oY and Forbidden(X-->Y), the edge will
be oriented as X<->Y.

* Modified the case expander so that it can expand cases for mixed data set.

* Reorganized the tools menu of the Data Editor a bit and added a gadget to
split datasets by cases.

4.3.6-2

* Put the BPC algorithm back from last August to fix a subtle bug in recent
versions.

* Modified reading for discrete data so that categories for defined variables
are treated as fixed. For example, if a variable X is defined to have categories
"true" and "false," and in the data column for X the string "red" appears, this
string is treated as a missing value for the column rather as an additional
category for the column. Also, in variable definitions, the syntax for a
definition of categories is changed from X:0=off,1=on to X:off,on.

* Any session node that in the editor interface edits parameter values has been
tagged as un uncancellable operation for now so that parameter values can be
passed to the simulation algorithm. Hopefully a better way of doing this will
be conceived of eventually.

4.3.6-6

* Modified simulation study behavior so that it logs while traversing the
execution tree. This involved adding logging behavior to relevant modules,
including updating.

* Fixed knowledge bug in PC. Now knowledge is invoked to determine whether a
set S of variables consists entirely of potential parents of X or potential
parents of Y, before using that set as a conditioning set to remove the edge
from X to Y. Previously, it was insisted that S consist of potential parents
of both X and Y.

4.3.6-7

* Fixed a bug in use of knowledge in PC. Used to be that for removing A---B,
it was insisted that variables in the conditioning set be potential parents
of both A and B. Now it is insisted that they all be potential parents of A or
of B.

* Added a gadget to interpolate means of columns for missing values in the
data editor.

* Added support in SEM IM to automatically enter and calculate variable
intercepts for acyclic models. The underlying model is still parameterized
using variable means, so the intercepts are calculated.

* Added GUI support for viewing either means or intercepts in the SEM IM Editor
for acyclic SEMs. Fixed some old labeling problems for popup mean editors.

* A bug in the construction of SEM PM's was fixed. When a graph was provided
for constructing a SEM PM that contained edges not directed or bidirected,
these edges were stripped from the graph. Now an exception is thrown.

4.3.6-8

* Add new data loading interface to support mixed tabular data loading, and
faster loading of large discrete data sets, among other things.

* Fixed bug in classifier, in which the target of classification was required
to be in the test set for Bayesian classification.

4.3.6-9

* Added knowledge sensitivity to MBFS.

* Added stats tabs for discrete pattern-based searches.

* Added tool to pattern-producing search editors to let the user automatically
add orientations implied by local orientation changes.

* Added an experimental score-based reorientation procedure.

* Added checks for tabular data checks and covariance matrices so that if
these are manually edited, downstream models (like searches) are invalidated.

* Added a rudimentary graph history mechanism to the pattern searches.

* Fixed bug where when classifier editor was closed and reopened, recent
classifications were no longer displayed.

* Added "graph properties" menu item to pattern searches.

* Modified the layered drawing layout so that it lays out all connected
components.

### INTERIM VERSION; MORE CHANGES WILL BE POSTED FOR 4.3.6-9 AS THEY ARE MADE ###