[Biojava-dev] OutOfMemory when using a big Weight Matrix solved; Curious distribution weights

[Bruno Aranda - ebioIntel]

Thanks Mark for your clear explanation, I will try to help as much as 
possible with my limited knowledge,

Bruno

Matthew Pocock wrote:

>Glad things are working better now. If you have any
>ideas how the docs could be fixed to make this clearer
>(appart from the problems caused by version squeue)
>then drop them in and we'll fold them into the
>javadocs.
>
>As far as the 0.300....4 thing goes, powers of 2 can't
>accurately represent loads of common decimal numbers -
>I'm sure mutually prime isn't a real word, but
>basically some numbers in one base has as a factor
>that can't be represented in the other as a finite sum
>(e.g. 1/3 in base 10 is 3/10, but 1/3 in base 2 is 1/2
>+ 1/4 - 1/8 ... You're seeing some squeue between them
>at the 17th decimal - this means that any answer you
>get is in the order of 10^-17 inaccurate, which
>naively means once in every 10^-17 prediction made
>with this weight matrix, it will classify something
>wrong. There aren't that number of nucleotides in the
>world, let alone in a single genome, so I don't think
>it matters. Besides, your estimate of 0.3 from limited
>data is likely to be vastly more than 10^-17 from the
>actual biological ratios in all examples of this motif
>known to nature (not just to you).
>
>Besides, BigDecimal has sucky performance :)
>
>Matthew
>
> --- Bruno Aranda - ebioIntel <biodev at ebiointel.com>
>wrote: > Thanks, now everything works ok when finding
>motifs
>  
>
>>Weights of the distribution should be: w(A)=0,
>>w(C)=0.5, w(G)=0.2, w(T)=0.3
>>
>>But with biojava we get all ok except T, with a
>>value of 
>>w(T)=0.30000000000000004...
>>    
>>
>
>  
>
>>Thanks!
>>
>>Bruno Aranda
>>ebioIntel
>>
>>    
>>
>
>________________________________________________________________________
>BT Yahoo! Broadband - Free modem offer, sign up online today and save £80 http://btyahoo.yahoo.co.uk
>  
>