You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@systemml.apache.org by "Matthias Boehm (JIRA)" <ji...@apache.org> on 2017/12/01 03:48:00 UTC
[jira] [Closed] (SYSTEMML-1980) Integer matrices
[ https://issues.apache.org/jira/browse/SYSTEMML-1980?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Matthias Boehm closed SYSTEMML-1980.
------------------------------------
Resolution: Fixed
Assignee: Matthias Boehm
Fix Version/s: SystemML 1.0
thanks for catching this [~dhutchis] - the root cause of these integer/boolean matrices is now fixed in master.
> Integer matrices
> ----------------
>
> Key: SYSTEMML-1980
> URL: https://issues.apache.org/jira/browse/SYSTEMML-1980
> Project: SystemML
> Issue Type: Bug
> Reporter: Dylan Hutchison
> Assignee: Matthias Boehm
> Priority: Minor
> Fix For: SystemML 1.0
>
>
> Under rare conditions a matrix can have INT ValueType and execute correctly. For example the program
> {code}
> X= Rand( rows=2, cols=2, min=1, max=2)
> R = cbind(as.matrix(nrow(X)) * 2, as.matrix(ncol(X)))
> {code}
> would throw an exception by the HopDagValidator because the Hops produced look like
> (here, "MI" means matrix data type and integer value type):
> {code}
> ----GENERIC (lines 28-0) [recompile=false]
> ------(33) u(cast_as_matrix) ([2]) [1,1,1000,1000,-1]MI [0,0,0 -> -MB]
> ------(35) b(*) (33,[2]) [1,1,1000,1000,-1]MI [0,0,0 -> -MB], CP
> ------(37) u(cast_as_matrix) ([2]) [1,1,1000,1000,-1]MI [0,0,0 -> -MB]
> ------(38) b(cbind) (35,37) [1,2,1000,1000,-1]MI [0,0,0 -> -MB], CP
> ------(44) PWrite R (38,[target/tes...],[false],[TEXT],[false],[,]) [1,2,-1,-1,-1]MI [0,0,0 -> -MB], CP
> {code}
> If this is intended, the HopDagValidator should be relaxed to allow integer matrices, and the fact that integer matrices are legal needs to be documented and scoped.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)