Changes in version 2.7-8 (2022-12-20)                  

    o   Fix idaFast() regression tests, using all.equal() instead of
	round(*) == round(*), notably for ATLAS-using BLAS/Lapack.

                 Changes in version 2.7-7 (2022-10-04)                  

    o   Tiny bug fix for fci.

                 Changes in version 2.7-6 (2022-04-04)                  

    o   ‘Proper’ registration of .Call()ed C routines.

    o   Not using if(class(.) = "...")

    o   Adapt to upcoming deprecation of C++ str::binary_function()

    o   dagitty masking randomDAG().

                 Changes in version 2.7-5 (2022-02-22)                  

    o   fixing bug.

                 Changes in version 2.7-4 (2021-11-25)                  

    o   updated links, removed old vignette.

                 Changes in version 2.7-3 (2021-06-01)                  

    o   adapt to R-devel's new/changed warning of
	matrix(<non-matching-length>, n,m).

                 Changes in version 2.4-2 (2016-09-26)                  

    o   print() and summary() methods for pc(), fci(), etc.

    o   new as(*, "amat") methods to get the adjacency matrix of a
	model, including a print() method for "amat"s.
	
	?amatType documents the different kinds of adjacency matrices
	in pcalg.

    o   New functions ......

    o   Deprecated LINGAM() in favor of new lingam().

    o   Deprecated allDags() in favor of pdag2allDags().

                 Changes in version 2.2-4 (2015-07-23)                  

    o   Bug fix in simulate(<gies>).

    o   Bug fix in LINGAM(., verbose=FALSE) (the default, now tested).

    o   Fix plot(pc(..), zvalue.lwd=TRUE), and better document that
	plot method.

    o   Much faster gSquareDis() and gSquareBin() for |S| in 1:4 or so.

    o   More test coverage, notably for pcSelect() and mcor().

    o   r.gauss.pardag() gets an optional argument labels.

                 Changes in version 2.2-3 (2015-07-12)                  

    o   Adapt to API change in igraph package.

    o   More test coverage and first speed tweaks.

                 Changes in version 2.2-2 (2015-06-24)                  

    o   Function gac for testing if a set of nodes satisfies the
	Generalized Adjustment Criterion (GAC). GAC is a sufficient and
	necessary criterion for estimating causal effects via covariate
	adjustment in DAGs, CPDAGs, MAGs and PAGs.

                 Changes in version 2.2-0 (2015-05-13)                  

    o   Function randDAG for generating random directed acyclic graphs
	(DAGs) with fixed expected number of neighbours. Several
	different methods are provided, each intentionally biased
	towards certain properties. The methods are based on the
	analogue functions in the igraph package.

    o   Function unifDAG and unifDAG.approx for uniform sampling of a
	labelled directed acyclic graph (DAG) with combinatorial
	enumeration.

                        Changes in version 2.1-0                        

    o   Function LINGAM for estimating the structure of Linear
	non-Gaussian Additive Models

                        Changes in version 2.0-4                        

    o   Function fciPlus for a fast (but accurate) version of fci (note
	that the already existing function rfci is a fast
	_approximation_ of fci)

                 Changes in version 2.0-1 (2014-03-06)                  

    o   Functions gds, gies, ges for score based learning

    o   Function backdoor for generalized backdoor criterion

    o   Bugfix in function wgtMatrix; this affects functions trueCov
	and rmvDAG; to reproduce old version, use option
	'back.compatible = TRUE'

    o   skeleton, pc, fci and rfci now have 'alpha' as third argument,
	then 'labels, p' where 'p' is optional; namely, derived from
	'labels' when these are specified.

    o   The default for extending the skeleton to a CPDAG is now
	udag2pdagRelaxed, which is completely deterministic (was
	udag2pdag before, which can use random orientations if sampling
	errors occur).