Hi, I'm trying to do something like the following:

def RVec(iv):
 for i in range(100):
 iv=tt.set_subtensor(iv[i], tt.roll(iv[i], -WBT[i]))
 return iv


Here WBT is a vector of integers, and I want to loop though the rows of the 
array iv and shift each row by a different amount given by that entry in 
WBT.

This is unbelievably slow.  The code spends an inordinate amount of time 
in theano.sandbox.cuda.basic_ops.GpuJoin.

Is there a better way to do this?  I want to be able to slot this into a 
function like:

lfunc = -0.5*tt.sum(((AmpParams*RVec(IVec)+OffParams-TData)/SigParams)**2-1)


Does anyone have any suggestions for how to speed this up?

Many Thanks

-- 

--- 
You received this message because you are subscribed to the Google Groups 
"theano-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
For more options, visit https://groups.google.com/d/optout.

Reply via email to