As the titles goes, I have an array of sorted real positive numbers, and some information on the structure of the data.
I know that there are two constants $a>>b>0$ and two other numbers $\delta, \epsilon<<a-b$, such that almost all the numbers in the array stay in the intervals $[a-\delta,a+\delta]$ and $[b-\epsilon,b+\epsilon]$. Moreover, the elements not belonging to them are a lot less than the ones in each one of this two intervals.
Without giving any value to the parameter, I looked for an algorithm that could let a computer approximate them from the array itself.
I tried with means, median, considering the differences between consecutive numbers, but nothing really worked.
In reality the thing I'm interested in, and that I'm asking to you, is weaker than that: there exists a method to find a threshold separating the values near to $a$ to those near to $b$?