Tweak dynamic parameters

[jeromekelleher]

Trying out running in three passes with a likelihood threshold of 0, mu, and rho^2 * mu^5 (which is about 1e-10, and around the same as 10 mutations). I've arbitrarily chosen this as the most unlikely thing we'd want to characterise exactly.

Just seeing if this gives us the same results as the previous pass with completely unlimited precision and improves run-time.

[jeromekelleher]

One oddity here is that we seem to be sometimes choosing a 3 mutations over a recombination. Eg for ERR4209451 we are choosing three recurrent mutations over a recombination.

Final HMM pass hmm_cost=3.0 ERR4209451 2020-03-10 B path=(0:29904, 194) mutations(3)=1397A>G,14805C>T,17659G>T]

This is done at full precision, so I'm not sure what's going on.

[jeromekelleher]

Solving the above problem required going back to first principles, and I think I've got the right expression for solving num_mismatches. We seem to be reliably choosing recombinants now rather than 3 mutations. There's a bunch of other stuff gone in now too.

[jeromekelleher]

Merging this and running a "long ARG" based on it over weekend. Let's see how it goes.