# Shall I derandomize a randomized algorithm in real application?

In general (and in real application), suppose I am using a randomized algorithm (e.g. Use MCMC to sample from a distribution and then compute $E(f(x))$ for some function $f$)

Assume my algorithm will face every possible input and I use a pseudo-random generator to provide randomness.

Shall I derandomize my algorithm by fixing a random seed every time the algorithm start?

The reason that I come up with this question is that by derandomizing it, I can make sure every input will have a corresponding output, which makes me feel more comfortable for practical use. However, I believe I will lose something (maybe the algorithm works very bad for worst case input) if I derandomizing it like this.