John Lees

Hard to set a sensible default, require scree plot to be used

It would be more general to allow any order of sam file, by using the k-mer as a unique ID between the SEER output and mapping co-ordinates.

Need to use a machine with cmake 2.7.12 or above. Setting "dist": "trusty" in .travis.yaml might work, but need to test. For now workaround in ca03e4d07b391a4bb89e6efe50565a59b2ca8891

Try using BOBYQA algorithm to maximise likelihood for Firth regression, rather than N-R iterations. I wonder if it might be faster...

Read file twice rather than into memory. Don't read sample vector in

Currently uses the most naive approach. Use an external sort instead: https://en.wikipedia.org/wiki/External_sorting

Calculate erfc exactly using arbitrary precision

1

I have written the code for this, but I think it should be added as a compile option rather than enabled by default

Should check that parameters aren't diverging, and that likelihood is constantly increasing (using step-halving if it isn't)

Possible changes to fastq input

11

Two possible additions ### An equivalent to the `-m` option in ska1: > Finally, the base call for the middle base in the split kmer is filtered to remove any...

Alternatives for .skf format

4

The current .skf format has the following issues: - It can become very large, especially for diverse datasets. As it is all deserialised (see #22 for attempts at altering this)...