Hi Michael,
thanks for your response. Indeed I have now verified the same using
randomise_parallel - exact same permutations (and p-value images) every
time.
I just noticed, however, that randomise has a --seed option, whcih I
guess could alter this scenario.
Nicola
On 8/4/2011 4:54 PM, Michael Harms wrote:
> Hi Nicola,
> It has been my empirical experience that you will indeed get exactly the
> same result (i.e., same permutations) if you run randomise using the
> identical command a 2nd time. (Use the -P flag to save the actual
> permutation vector if you are interested in comparing the permutations
> employed across repeated randomise instantiations). However, note that
> if you have multiple contrasts specified in a .con file, the
> permutations that you get for the 2nd, 3rd, etc contrasts are NOT the
> same as you would get if you ran those contrasts as the only contrast in
> the .con file. That is, within a given instance of randomise, its
> internal random seed does not appear to "reset" to the initial value for
> consecutive contrasts. (This would IMHO be a useful feature to have
> available as an optional flag -- for example if one wishes to run both
> positive and negative contrasts using the identical set of permutations
> within a single run of randomise).
>
> Don't know how this all applies with randomise_parallel...
>
> cheers,
> -MH
>
> On Thu, 2011-08-04 at 14:47 +0200, Nicola Toschi wrote:
>> Dear all,
>>
>> I was wondering how the seeding is done in randomize.
>>
>> In other words, if I run the same randomise command multiple times
>> (assuming the number of permutations is lower that the number of all
>> possible permutations), will I always get exactly the same result? Does
>> the same answer hold for randomise_parallel?
>>
>> Thanks in advance,
>>
>> Nicola
|