Shing Hei Zhan

I've rerun BEAGLE using different query haplotypes by setting Ne to 10. BEAGLE sets Ne to 1 million by default. I don't fully understand why they set it so big,...

In case I forget, just jogging down some notes about BEAGLE 4.1's approach to calculating allele probabilities, which are used to get the MAP alleles at each imputed site of...

I think I understand the difference between state probability matrix and allele probability matrix. The former is what you get from forward-backward algorithm, and it contains the allele probabilities at...

So, we should mimic what BEAGLE is doing, right? And interpolate the allele probabilities of the ungenotyped markers? So, we are really implementing BEAGLE 4.1's algorithm in tskit.

Then, getting the MAP alleles at each ungenotyped marker site for each query haplotype is straightforward.

I've added a separate file (`python/tests/test_beagle.py`) to contain the code to replicate the BEAGLE 4.1 algorithm. It is temporary. I'll reorganise it later. I still don't have all the little...

Some notes about some concepts used in the BEAGLE 4.1 algorithm. * `Marker` refers to a genotyped site. * `Cluster` refers to a set of adjacent markers (defined by a...

Most of the implementation is there. I'm corresponding with Brian Browning to understand how reference haplotype segments are indiced.

In the BEAGLE 4.1, the reference haplotype segments used to index the HMM state probability matrix are defined in the function `refHapSegs()` in `ImputationData.java`. In `refHapSegs()`, the reference haplotype segments...

It's simpler if we don't follow their way of normalizing the probability values in the matrices. We really just want to use Equation 1 in BB2016, but in their implementation...