because it seems like sometimes we blow the stack.
When we perform a split pass (splitting the read to contain the per-allele blow-up and then merge afterwards). We end up having a lot more "jitter" in the final values (each intermediate split, per read, can introduce another point where the rounding matters). Consequently, we end up having many more unique (log) likelihood values.
because it seems like sometimes we blow the stack.
When we perform a split pass (splitting the read to contain the per-allele blow-up and then merge afterwards). We end up having a lot more "jitter" in the final values (each intermediate split, per read, can introduce another point where the rounding matters). Consequently, we end up having many more unique (log) likelihood values.