Maximum value that satisfies a specific function

Question

I want to find the maximum value for a that doesn't change the output from another function. (_find_mean simply finds the location of the maximum element in a matrix)

This is how I am currently doing it and was wondering if there was a better way.

def _find_max_a(matrix1, matrix2):
 a = np.arange(0,10, 0.01)[::-1]
 for val in a:
 mat = matrix1 + val * matrix2
 if _find_mean(matrix1) == _find_mean(mat):
 return val
 else:
 pass

• holroy: Doesn't this code return for any value within a which doesn't change the output from the other function?
• holroy: And your loop doesn't change a, it loops on values of a. So is your description correct?

Answer 1

• arange can build a descending vector. Building an ascending vector and reversing it seems wasteful. Consider

  a = np.arange(10, 0, -0.01)
  

• matrix1 is never modified, but its _find_mean is recomputed on each iteration. Yet another waste. Compute it once.

• Instead of linear search, you may want to consider a bisection of the search interval.

• else: pass achieves nothing, and can be safely omitted.

• _find_mean is a kinda strange name for a function which

  finds the location of the maximum element in a matrix

• kyrill: Now the question is how to do the arange thing more efficiently. For larger range with finer accuracy, the amount of required memory is quite high (800kB for range 100, 0, -0.001 with dtype of np.float64, which is the default). Loop with manual decrement by 0.001 would introduce errors, whereas a scaled loop would require division.