Whats New in Release 2
Critical Changes in Behaviour
Generally we seek to maintain upward compatability so that ASReml 1.00 code
will continue to run. However, to deliver improved facilities,
some changes to behaviour
are unavoidable.
Workspace
ASReml will automatically increase workspace
in large jobs if it needs it and it is available.
The default workspace is 32Mbyte but the user
can specify a larger amount (see Section 3.2).
In any case, if a job runs out of workspace
when it is running, and has not already claimed the maximum amount
of workspace permitted on the computer, ASReml will restart the job
with a larger allocation.
Storage of alphabetic factor labels
In release 1.00, there was a fixed allocation for alphabetic labels
of 5000 labels of up to 20 characters each. Longer labels were
truncated and regarded as equivalent if identical in the first 20
characters.
In release 2.00, the allocation is dynamic with default provision for
2000 labels of 16 characters each (see Section 5.1). However, the allocation is increased
to make provision for the declared size of all alphabetic factors
and the label length can be set using the !LL qualifier.
If the number of levels is overspecified, the !PRUNE qualifier
may be used to readjust the sizes.
Slash operator in the model specificaton
In model specification, A/B now expands
to A A.B
Graphics
In this version, graphics have been converted from Interacter
to Winteracter
except that MENU mode has been removed, being replaced by
WinASReml.
This should not alter the appearance of any of the graphs produced.
Singularities in Average Information matrix
Singularities in ASReml arise in three contexts: the linear model,
the variance model and the Average Information (AI) matrix.
The AI matrix is used to
give updates to the variance parameter estimates.
The AI matrix is used to
give updates to the variance parameter estimates.
In release 1.00, if singularities were present in the AI matrix,
a generalised inverse was used which effectively conditioned on
whichever parameters were identified as singular.
ASReml now aborts processing if singularities appear
unless the !AISINGULARITIES qualifier is set.
Which particular parameter is singular is reported in the
variance component table printed in the .asr file.
The most common reason for singularities is that
the user has overspecified the model and is likely to misinterpret
the results if not fully aware of the situation.
Overspecification will occur in a direct product of two
unconstrained variance matrices, when a random term is confounded
with a fixed term and when there is no information in the data
on a particular component. The best action
is to reform the variance model so that the ambiguity is removed,
or to fix one of the parameters in the variance model so that the model
can be fitted.
Only rarely will it be reasonable to specifiy
the !AISINGULARITIES qualifier.
Prediction
The order predicted values are presented has changed
and are now under user control: the
factors are presented in the order they are declared on the PREDICT
statement. The predicted values are ordered so that the rightmost
factor rotates fastest, followed by the second from the right.
Extended Factor Analytic
XFA model fitting with some specific variances zero
is extended to allow direct product structures
of XFA with other structures (previously only XFA with Identity was
allowed) provided the other structure is not another XFA with some
zero specific variances, and to allow several model terms to have
XFA structures involving zero specific variances.
The work around for XFA x XFA (both with zero PSIs) is to fix 'ZERO' PSIs
in one of the terms at 0.0001.
In 3-way terms involving an XFA with zero specific variance, the
central component must be an Identity. I.e. XFA x I x C and
C x I x XFA are allowed where C is some structure and I the Identity.
ASReml update changes
Changes to the rules for when the AI update of a variance parameter
is replaced by a smaller update have been
made to make the process more robust. Consequently,
the iteration sequence may differ slightly in some jobs from
earlier versions. It should end up at effectively the same point
with only small numerical differences in the results. These rules
come into play when the simple AI update is large (10 fold)
or would put the parameter
'out of bounds'.
BLUEs in .asr file
In release 1.00, the non-zero BLUEs (estimates of fixed effects) were
reported in the .asr file as well
as in the .sln file.
In release 2.00, the BLUEs are not reported in the .asr file
unless the !BRIEF -1 qualifier is set or there are less than 10
BLUEs to report.
Return to What's New