Hi Brennan & All, On 29 March 2010 00:46, Brennan Williams wrote: > Andrea Gavana wrote: >> As for your question, the parameter are not spread completely >> randomly, as this is a collection of simulations done over the years, >> trying manually different scenarios, without having in mind a proper >> experimental design or any other technique. Nor the parameter values >> vary only on one axis in each simulation (few of them are like that). >> > > I assume that there is a default "norm" that calculates the distance > between points irrespective of the order of the input coordinates? > > So if that isn't working, leading to the spurious results, the next step > is to normalise all the inputs so they are in the same range, e.g > max-min=1.0
Scaling the input data using their standard deviation worked very well for my case. > On a related note, what approach would be best if one of the input > parameters wasn't continuous? e.g. I have three quite different > geological distributions called say A,B and C. > SO some of my simulations use distribution A, some use B and some use C. > I could assign them the numbers 1,2,3 but a value of 1.5 is meaningless. Not sure about this: I do have integer numbers too (the number of wells can not be a fractional one, obviously), but I don't care about it as it is an input parameter (i.e., the user choose how many o2/o3/injector wells he/she wants, and I get an interpolated production profiles). Are you saying that the geological realization is one of your output variables? > Andrea, if you have 1TB of data for 1,000 simulation runs, then, if I > assume you only mean the smspec/unsmry files, that means each of your > summary files is 1GB in size? It depends on the simulation, and also for how many years the forecast is run. Standard runs go up to 2038, but we have a bunch of them running up to 2120 (!) . As we do have really many wells in this field, the ECLIPSE summary file dimensions skyrocket pretty quickly. > Are those o2w,o3w and inw figures the number of new wells only or > existing+new? It's fun dealing with this amount of data isn't it? They're only new wells, with a range of 0 <= o2w <= 150 and 0 <= o3 <= 84 and 0 <= inw <= 37, and believe it or not, our set of simulations contains a lot of the possible combinations for these 2 variables (and the other 4 variables too)... Andrea. "Imagination Is The Only Weapon In The War Against Reality." http://xoomer.alice.it/infinity77/ ==> Never *EVER* use RemovalGroup for your house removal. You'll regret it forever. http://thedoomedcity.blogspot.com/2010/03/removal-group-nightmare.html <== _______________________________________________ NumPy-Discussion mailing list NumPy-Discussion@scipy.org http://mail.scipy.org/mailman/listinfo/numpy-discussion