I think that would be a good solution --- I think that the notion of
triangularity extends well to non-square matrices.

Best,

Tamas

On Thu, Jan 15 2015, Andreas Noack <[email protected]> wrote:

> The problem is that the shapes of L and U depend on the input matrix and it
> need not be square. I'm very tempted to allow Triangular to be rectangular
> so actually the triangular matrix would be more a trapezoid. I think it
> could work but would require some efforts. It would also make it it
> possible to use the structures for blocked/paneled factorization algorithm.
>
> 2015-01-15 9:33 GMT-05:00 Tamas Papp <[email protected]>:
>
>> I am still learning about matrices in Julia so I didn't file an issue
>> becuase I am not sure this is a bug, but in
>>
>> julia> VERSION
>> v"0.3.5"
>>
>> julia> lu = lufact([1 2; 3 4])
>> LU{Float64,Array{Float64,2}}(2x2 Array{Float64,2}:
>>  3.0       4.0
>>  0.333333  0.666667,Int32[2,2],0)
>>
>> julia> typeof(lu[:U])
>> Array{Float64,2}
>>
>> julia> typeof(lu[:L])
>> Array{Float64,2}
>>
>> wouldn't it make sense for the :U and :L accessors to return triangular
>> matrices?
>>
>> Best,
>>
>> Tamas
>>

Reply via email to