I have a Machine Learning algorithm in python which is affected by randomness. So, everytime I run it, it outputs a different score (based on its predictions).
Let's suppose I run it 10000 times. Then I can calculate the sample average and std for this particular sample.
How many times do I have to run my program in order to approximate, with some degree of certainty, the true efficiency of my algorithm? Can the statistics mentioned before help in any way?