ProbabilitySamplingScheduler might be inverted

[DanBlackwell]

According to the comment for TestcaseScore::compute below, returning a higher value should make it more likely that a queue member is selected:

LibAFL/libafl/src/schedulers/testcase_score.rs

Line 23 in 1dcfe8e

/// Computes the favor factor of a [`Testcase`]. Higher is better.

It looks to me like the reciprocal of this number is used in the ProbabilitySamplingScheduler, resulting in bigger compute scores ending up with lower probability of selection:

LibAFL/libafl/src/schedulers/probabilistic_sampling.rs

Line 87 in 1dcfe8e

let prob = 1.0 / factor;

In my local version I have changed the above line to just let prob = factor, and it seems to behave more like I would expect. Perhaps I'm missing something though.

I'll also add that I am updating testcase scores, so I also added the following to the end of the function (if this is a new key then the subtraction never happens, so works for new insert and replace):

        if let Some(old_prob) = meta.map.insert(idx, prob) {
            // subtract the old probability from total first
            meta.total_probability -= old_prob;
        }
        meta.total_probability += prob;

[domenukk]

@tokatoka can you take a look?

[tokatoka]

yes i think you are right
can you send a PR?

[tokatoka]

I'll also add that I am updating testcase scores, so I also added the following to the end of the function (if this is a new key then the subtraction never happens, so works for new insert and replace):

To implement this, can you impl RemovableScheduler for Probabilistic scheduler