Re: results from Array and ParallelArray differ
- To: mathgroup at smc.vnet.net
- Subject: [mg117068] Re: results from Array and ParallelArray differ
- From: Heike Gramberg <heike.gramberg at gmail.com>
- Date: Wed, 9 Mar 2011 06:57:30 -0500 (EST)
It looks like ParallelArray uses Join to combine the results from the
individual kernels instead of Plus. I guess it assumes that the partial
results coming from the separate kernels are expressions with head
Plus, so the final result would be Join[Plus[...],Plus[...]]=Plus[...,...]
as you would expect, but in this case the result from each kernel is a
number, so Join[result1,result2] goes wrong.
Replacing Join with Plus as you did looks like the easiest way to
solve this, but I would have expected that the ParallelArray would
check for a situation like this (as ParallelCombine does for example).
Heike.
On 7 Mar 2011, at 10:49, Peter Pein wrote:
> Dear group,
>
> I did a search using tags "ParallelArray" and "Join" and did not find
> any posting in this group; apologies, if this has already been asked.
>
> The documentation tells:
>
> "ParallelArray will give the same results as Array, except for side
> effects during the computation."
>
> I use an anonymous function and no variables; DistributeDefinitions,
> SetSharedVariable etc. can not be the cause for the trouble in the
> following toy example:
>
> In[1]:= arrsum = Array[1/Plus[##]&, {3, 3}, {1, 0}, Plus]
> Out[1]= 37/10
> :-)
>
> In[2]:= parrsum = ParallelArray[1/Plus[##]&, {3, 3}, {1, 0}, Plus]
> Out[2]= Join[35/12, 47/60]
> :-(
>
> In[3]:= parrsum /. Join->Plus
> Out[3]= 37/10
>
> Is this a bug, a feature or my thick-wittedness? What can be done?
>
> Well, I could use Total[ParallelTable[...],2] but I don't want to.
>
> Thanks in advance,
> Peter
>
> P.S.: I came across this problem while coding
> http://dl.dropbox.com/u/3030567/Mathematica/springer.nb
>