[Bioc-devel] Bug in DEXSeq, me thinks

[Steve Lianoglou]

Hi Alejandro,

Thanks for fixing those things up.

I pulled from SVN earlier today and did notice some changes and,
indeed, the beautification of the divideWork function caught my eye
... now you just have to clean up the mixed `=` and `<-` assignment
ops and it will be picture worthy ;-)

Thanks again,

-steve

On Wed, Mar 14, 2012 at 1:17 PM, Alejandro Reyes
<alejandro.reyes at embl.de> wrote:
> Dear Steve,
>
> Thanks for the bug reports!  I think what was missing in in the muhats
> lapply was the "simplify =FALSE", what I find strange its that you are the
> first one to get an error from it, might be because there is at least one
> "try-error" in almost the datasets and this avoids it to be converted in a
> matrix.  Anyway I changed everything to lapply just to be on the safe side.
>
> Regarding the:
>
>
> length(testablegenes) %% mc.cores == 0
>
>
> I did not manage to reproduce this error, but I agree that this function
> (divideWork) was a bit messy.  With Wolfgang's help we re-engineered this
> function and took away the sapply-s, I am hoping that this won't give errors
> anymore. Let me know!
>
> Thanks again!
> Alejandro
>
>
>> And another sapply ->  lapply in divideWork (core_functions ~ line 496):
>>
>> fData(ecs)[,fattr][rows]<- do.call(c, sapply(...))
>>
>> is exploding when
>>
>> length(testablegenes) %% mc.cores == 0
>>
>> -steve
>>
>> On Tue, Mar 13, 2012 at 5:33 AM, Steve Lianoglou
>> <mailinglist.honeypot at gmail.com>  wrote:
>>>
>>> And while we're at it, I think there's another place you need to swap
>>> out an lapply for an sapply (or set simplify=FALSE) in the
>>> `divideWork` function.
>>>
>>> Specifically, it seems that when `testablegenes` is divisible by
>>> mc.cores:
>>>
>>> (1) This sapply call returns a matrix back to `rowlist`, when you
>>> really want a list, since you will `do.call(c, rowlist)` later:
>>>
>>> rowlist<- sapply(1:mc.cores, function(i){which(fData(ecs)$geneID %in%
>>> subgenes[[i]])})
>>>
>>> (2) Ditto for your `stopifnot` check. The `do.call(c, sapply(...))`
>>> explodes because the sapply returns a matrix:
>>>
>>> stopifnot(all(as.character(fData(ecs)$geneID)[rows] == do.call(c,
>>> sapply(allecs, function(x){as.character(fData(x)$geneID)}))))
>>>
>>> ... or ... at least these things are happening for me right now.
>>>
>>> I have a weird feeling that I might have hosed something, since I
>>> reckon people would have tripped over this by now since it's easy to
>>> fall into this scenario where:
>>>
>>> length(testablegenes) %% mc.cores == 0
>>>
>>> when mc.cores == 2
>>>
>>> So ... hopefully I'm not just chasing my tail then asking you to chase
>>> yours.
>>>
>>> Thanks,
>>> -steve
>>>
>>>
>>>
>>>
>>> On Tue, Mar 13, 2012 at 5:10 AM, Steve Lianoglou
>>> <mailinglist.honeypot at gmail.com>  wrote:
>>>>
>>>> And then add names back to muhats, of course:
>>>>
>>>> On Mon, Mar 12, 2012 at 8:32 PM, Steve Lianoglou
>>>> <mailinglist.honeypot at gmail.com>  wrote:
>>>>>
>>>>> Hi,
>>>>>
>>>>> It just so happened that DEXSeq, for reasons I couldn't explain, kept
>>>>> throwing errors during a call to `estimateDispersions`. I couldn't
>>>>> explain it because I really thought I didn't change much, but I've got
>>>>> some magic combination of samples I'm currently trying to
>>>>> estimateDispersion on that none of them throw any errors while
>>>>> estimate `muhats` (line 245 of core_functions.R):
>>>>>
>>>>> muhats<- sapply(testablegenes, function(gn) {
>>>>> ...
>>>>> muhat
>>>>> })
>>>>
>>>> [snip]
>>>>
>>>> names(muhats)<- testablegenes
>>>>
>>>> -steve
>>>> --
>>>> Steve Lianoglou
>>>> Graduate Student: Computational Systems Biology
>>>>  | Memorial Sloan-Kettering Cancer Center
>>>>  | Weill Medical College of Cornell University
>>>> Contact Info: http://cbio.mskcc.org/~lianos/contact
>>>
>>>
>>>
>>> --
>>> Steve Lianoglou
>>> Graduate Student: Computational Systems Biology
>>>  | Memorial Sloan-Kettering Cancer Center
>>>  | Weill Medical College of Cornell University
>>> Contact Info: http://cbio.mskcc.org/~lianos/contact
>>
>>
>>
>



-- 
Steve Lianoglou
Graduate Student: Computational Systems Biology
 | Memorial Sloan-Kettering Cancer Center
 | Weill Medical College of Cornell University
Contact Info: http://cbio.mskcc.org/~lianos/contact