Optimal Combination Model With Constraints & Variables

  • Hi!
    I need to optimize an Excel model with 5 variables each of which cam be any value from 1 to 20. The solution space, therefore, contains 3,200,000 combinations. I have a software(Palisade Risk) which can give out a random sampling of this solution space. I need to know how many combinations can be considered as a representative sample of this solution space - e.g. can it be 1%?.
    Thanks!!!
    Dima

  • Re: Representative Sample


    Hi!


    There might be some mathematically way of calculating what is the minimum number of samples you need.


    However, one way of determning it is to run the "simulation" a few times with, lets say, with 1% of the possible samples. How much does your answer change? If it changes only a "litle", then it is enough, if it changes "alot", then it is to few... So, whats litle/alot depends on the problem

  • Re: Representative Sample


    Quote from Oskar

    Hi!


    There might be some mathematically way of calculating what is the minimum number of samples you need.


    However, one way of determning it is to run the "simulation" a few times with, lets say, with 1% of the possible samples. How much does your answer change? If it changes only a "litle", then it is enough, if it changes "alot", then it is to few... So, whats litle/alot depends on the problem



    using iterative sampling is what risk does...so I need to see how my output behaves from sample to sample?[hr]*[/hr] Auto Merged Post Until 24 Hrs Passes;[dl]*[/dl]

    Quote from shg

    How about using Solver with integer constraints?


    i am going to use Evolver or OptQuest....Solver has still yet to be evaluated...thanks!

  • Re: Optimal Combination Model With Constraints & Variables


    Let me try to explain with an example:


    If you ask 1% of the population what they will vote, is then the result representative for forecasting the result in the acctual election with 100% of the population? There probably is some mathematics behinds this, maybe you can find some more information from Palisade/ risk.


    A quick and dirty way of determining whether it is representative or not is the following:


    1. ask 1 % of the population
    2. ask another 1% of the population
    3. Compare 1 and 2; if they agree, then 1% is enough, if they are very different, then you must ask a larger % of the population. So if you have time, run the 1% case a few times and compare. Error tolerance depends on the problem...


    risk tip: Is it not also possible to look at a distribution of the choosen samples? This way, you can tell wether the 1% samples choosen spans the whole posible sample space, thereby "represents" possible outcomes.

Participate now!

Don’t have an account yet? Register yourself now and be a part of our community!