i.maxlik

[Michael Shapiro]

I have seen this "problem" with i.class but  this  is  the  first
with  i.cluster.   Both i.class and i.cluster generate covariance
matrices based on  the  pixels  assigned  to  a  class.  However,
i.maxlik  has more stringent requirement that the that the matrix
simply be a covariance (eg  symmetric)  matrices.  They  must  be
invertible  and  have  non-negative  eigen  vectors.  i.class can
easily generate such matrices and now apparently i.cluster can as
well.   i.maxlik  will  "ignore" classes defined by such matrices
and reduce the number of class in the final map by the number  of
offending classes.

|
|I have been running i.cluster on 3week-avhrr composites for 1988. When running i.maxlik I got this message :
|
|LOCATION: gr                MAXIMUM LIKELIHOOD                MAPSET: china.gvi
|
|SIGNATURE
|Enter the signature file to be used for classification
|Enter 'list' for a list of existing subgroup [3week] signature files
|Enter 'list -f' for a list with titles
|Hit RETURN to cancel request
|> list
|<list>
|----------------------------------------------
|subgroup [3week] signature files available in mapset china.gvi:
|sighu3   sighu3w
|
|----------------------------------------------
|
|Enter the signature file to be used for classification
|Enter 'list' for a list of existing subgroup [3week] signature files
|Enter 'list -f' for a list with titles
|Hit RETURN to cancel request
|> sighu3w
|<sighu3w>
|WARNING: signature 1 is not valid (ill-conditioned)
|WARNING: signature 6 is not valid (ill-conditioned)
|You may continue, but the invalid signatures will not be used
|Continue? (y/n) [n] 
|--------------------------------------------------------------------------
|
|Do you know what this means? i.cluster ran without giving me any error message.
|
|Dominique Bachelet
|dominiq at heart.cor.epa.gov
|
|



Michael