CHANGELOG:
v1.2-1 Changes
  Changes:
    - Minor (non-visible) changes for CRAN compliance.
    
v1.2 - Changes, New Functions, and New Features
  Changes:
    - Minor (non-visible) changes to ensure PROTECT compliance.
  New Functions:
    - as.sociomatrix.eventlist: converts an event list to a time-marginalized
      adjacency matrix.
  New Features:
    - The simulate() method for rem.dyad now supports conditioning on initial
      events (sender/receivers, timing, or both).

v1.1 - Changes, New Features, and Bug Fixes
  New Features:
    - A simulate() method has been added for rem.dyad; it should
      support all rem.dyad families, although it is not always fast.
    - Relatedly, rem.dyad can now be used to produce "skeleton" models
      for use with simulate(); these can be used to perform ab initio
      simulation.
  Changes:
    - Changed default name given to each list for multiple event lists in 
      REM (was "Ego n," now "Sequence n").
    - Stack usage has been reduced in many cases, which should reduce 
      overflow errors for sequences with very large numbers of events.R_NativePrimitiveArgType
    - C entry points are now registered, per coercion.
  Bug Fixes:
    - rem.dyad in some cases gave incorrect deviance values for ordinal models
      when gof=TRUE.
    - Standard errors reported for fit.method==MLE were off by a factor of 2
      (BPM and other methods are unchanged).
      
v1.0-4 - Changes and Bug Fixes
  Changes:
    - Minor (non-visible) changes to ensure CRAN compliance.
  Bug Fixes:
    - rem included exogenous final events when computing the null deviance for
      ordinal time models.

v1.0-3 - Changes
  Changes:
    - Minor (non-visible) changes to ensure CRAN compliance.

v1.0-2 - Bug Fixes and Changes
  Changes:
    - Print output has been changed, hopefully in a more pleasing way.
    - rem.dyad now removes NAs and loops (with a warning).
    - rem.dyad now exits when called with poorly ordered events.
  Bug Fixes:
    - Calling rem.dyad with NAs, loops, or unordered events (all unsupported)
      could produce failure.
    - Null deviance for non-ordinal case was not sensible.  (An intercept model
      is now used.)
    - Print and summary methods would fail on a null model object.
    - Reported deviance for non-ordinal case did not include right-censoring
      (although this did not affect estimation).

v1.0-1 - Bug Fixes and New Features
  New Features:
    - rem.dyad now allows the initial coefficient "seed" for the optimization
      process to be manually set.
    - When called with infinite degrees of freedom, rem.dyad will fit a model
      with Gaussian priors.  
  Bug Fixes:
    - acl.ps could fail when called with an edgelist containing NAs.
    - .onLoad had unfortunate behavior with recent R versions.

v1.0 - External Release Candidate
  Changes:
    - Added NAMESPACE support, though we hate it.
    - As promised at 0.9, rem is now rem.dyad,and rem.ego is now rem.  This
      better reflects the actual nature of the models in question (since the
      old "rem" fit the dyadic special case, and "rem.ego" had become a
      generic routine).  Likewise, internal files, routines, and help
      information has been altered to reflect this change.
    - Set the BSIR example for rem.dyad not to run by default (it takes too
      long for CRAN).
    - Set BPM as the default estimation method for both rem and rem.dyad.  All
      tests so far show superior performance versus the MLR (even on 
      frequentist grounds), and there seems no reason not to switch.
  Bug Fixes:
    - Fixed an effect labeling issue in rem.  [Submitted by Chris DuBois]
    - Fixed a minor labeling bug in rem.ego.

v0.9-2 - Internal Release
  Bug Fixes:
    - Minor documentation updates for rem and rem.ego.

v0.9-1 - Internal Release
  Bug Fixes:
    - rem.ego failed on BSIR with some degenerate models.

v0.9 - Internal Release
  New Features:
    - rem.ego now supports fully Bayesian analysis using MCMC and SIR.
    - Verbosity is now selectable in rem.ego.
  Changes:
    - Backend deviance routines have been renamed to be consistent with the
      upcoming 1.0 switch.
    - rem.ego deviance calculations are now backended (and are much faster).
  Bug Fixes:
    - Made a minor adjustment to the BSIR weights in rem.
  Notes:
    - For 1.0, function nomenclature and internal organization will change.
      rem.ego (which is really a generic rem routine and not "egocentric" in
      any important sense) will become "rem."  The current rem (which is 
      actually a special case function for dyadic data) will become "rem.dyad."
      This will also be reflected in some changes to the package structure,
      none of which will be internally visible.

v0.8 - Internal Release
  New Features:
    - Interval likelihood now supported in rem.ego.
    - rem.ego now supports very primitive Bayesian estimation; specifically,
      it is now possible to obtain posterior mode estimates (under independent
      t parameter priors), along with (hypothetically) asymptotic posterior
      standard deviations.  The rem.ego summary method will automatically
      adjust its presentation based on the type of estimation employed.
    - rem.ego now supports exogenous events; these should be indicated by event
      codes of 0 in the input data.  These could be handled indirectly by the
      user previously in the ordinal case, but are now managed internally (and
      their contribution to the interval likelihood is taken into account).
    - rem.ego is now documented.

v0.7 - Internal Release
  Bug Fixes:
    - Protection stack was overflowing for large models with triad census terms.
    - rem_n2llik_samp_R was not working (internal coercion problem).
    
v0.6 - Internal Release
  Bug Fixes:
    - The PSAB-BY term was not being detected in rem.
    - In some cases, rem did not generate necessary triad statistics (depending
      on what other terms were in a model that included triads), and could
      crash.

v0.5 - Internal Release
  New Functions:
    - rem.ego: analysis of egocentric relational event sequences.  (Print and
      summary methods also included).
  Changes:
    - The relevent package now depends on the trust and sna packages.
    
v0.4 - Internal Release
  New Features:
    - Support for multiple covariates (of each type) has been added to rem;
      covariates can also vary with the event history.
    - Support for generic eventwise covariates has been added to rem.
    - An option has now been added to rem to allow conditioning on the first
      portion of an event sequence.
    - rem now calculates the AICC as an additional model selecton criterion.
  Changes:
    - Effects in rem are now specified via much more convenient and intuitive
      method.  Regrettably, this will break existing code.

v0.3 - Internal Release
  New Features:
    - rem now supports use of full timing information (previously, only the
      ordinal likelihood was used).
    - man page now added for rem.

v0.2-1 - Internal Release
  Changes:
    - Removed some trace messages in accum_ps_R.
  Bug Fixes:
    - Normalized total degree stats were off by a factor of 2 (normalization
      was inadvertantly by number of events, not twice the number of events).
      This had the main effect of cutting the estimated parameter values in
      half, where applicable.
    - The NTDegRec effect label had a typo.

v0.2 - Internal Release

v0.1 - Initial Internal Release