Matrix entity type

[Matt Jones]

Peter,

I think I agree, but would like to wait a while before heading down this 
route just for stability in EML.  I think further experimentation with 
what is needed is in order.  In our case, we have been considering a 
matrix type for use in Monarch as well, and have even had one in various 
versions of the pipeline language we've been developing for Monarch.

Right now, the Monarch pipeline language permits inputs and outpus from 
monarch steps to be one of the following:
   Scalar (no correspondence in EML)
   Attribute
   Table
   Other

We have also seen a need for and experimented with:
   Vector
   Matrix

When comparing these, we felt that a vector was a one-dimensional set of 
values of a single atomic type (ie, a colleciton of scalars), and a 
matrix is a 2 or more dimensional set of scalar values. Matrix 
subscripts usually correspond to real-world data too, often through an 
associated Vector of the same dimension that contains the values 
corresponding to each subscript.  For example, one could imaging a 
100x100 matrix containing current flow values in space, where the x 
subscripts correspond to longitude and the y subscripts correspond to 
latitude, and there is a 100 element 'X' vector listing the longitude 
values and a 100 value 'Y' vector listing the latitude values.

However, we also felt that there was a strong correspondence between an 
Attribute and a Vector, with the major difference being that an 
Attribute is generally embedded in a table.  We currently use Attribute 
like a Vector however, and it is a little confusing.  A Table, however, 
is quite distinct from a matrix in that a Table can contain Attributes 
with multiple atomic types, while a matrix is homogenous with respect to 
its type (ie, all cells are of a single type).

We've been processing data in Matlab and could really use the Matrix 
entity type there. We've been struggling a bit with the correspondence 
with the EML entity types, but haven't really worked it out well enough 
to know what we really need.

I for one would be open to extending the entity types in a rationale and 
well-considered manner, as long as it is backwards compatible with EML 
2.0.0.

Matt

Peter McCartney wrote:
> Has anyone else felt the need for a new entity type  to describe 
> matrices? Im sitting here writing the spec for a cross-tabulation module 
> i want to add to Xylopia am realizing that while its possible to 
> describe its output using the table entity module, thats really not a 
> good fit. I really dont need to define each column as a distinct 
> attribute - they are all the same data type and meaning and Im forced to 
> use an entirely different model to define the row headers even thought 
> order of rows versus columns is really arbitrary.What i need is one 
> description for each dimension of the matrix, plus one attribute to 
> describe the cell's datatype and calculation. This module would be 
> useful for describing lots of statistical output up to N dimensions.
>  
> Peter McCartney (peter.mccartney at asu.edu <mailto:peter.mccartney at asu.edu>)
> Center for Environmental-Studies
> Arizona State University
>  
>