You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This is low priority issue for me, but when DiskArrays are used for memory arrays, the it might be useful to adapt the default chunksize for transposed arrays:
For a transposed array I would expect that the first chunk size of (1:31, 1:4000). I guess one can use stride to check this.
I am note sure if OffsetArray is supposed to be sported but currently all indices from eachchunk are ones-based
using DiskArrays, OffsetArray
DiskArrays.default_chunk_size[] =1
OA =OffsetArray(ones(4000,4000),(-1,-1));
c =eachchunk(OA);
OA[last(c)...]
# ERROR: BoundsError: attempt to access 4000×4000 OffsetArray(::Matrix{Float64}, 0:3999, 0:3999) with eltype Float64 with indices 0:3999×0:3999 at index [1:4000, 4000:4000]
None of these issue affect me currently in any production code, but I thought it might be useful to notify you about it.
I am using DiskArrays v0.3.22
The text was updated successfully, but these errors were encountered:
Thanks, yes we should test on OffsetArray. Correctness fixes are probably more important at this stage, the cost of wrong stride in Array will be lost in the cost of reading from disk (assuming this mostly happens in a broadcast/zip with a real disk-based array). But yes one day we could correct for that.
This is low priority issue for me, but when DiskArrays are used for memory arrays, the it might be useful to adapt the default chunksize for transposed arrays:
For a transposed array I would expect that the first chunk size of
(1:31, 1:4000)
. I guess one can usestride
to check this.I am note sure if
OffsetArray
is supposed to be sported but currently all indices fromeachchunk
are ones-basedNone of these issue affect me currently in any production code, but I thought it might be useful to notify you about it.
I am using DiskArrays v0.3.22
The text was updated successfully, but these errors were encountered: