Implement Ascertainment Corrections

[SimonGreenhill]

It would be lovely to have the ascertainment corrections in FilteredAlignment available.

[alexeid]

Sounds good. What exactly would you like to see? I imagine something like:

A ~ PhyloCTMC(tree=tree, L=100, Q=jukesCantor());
D = removeInvariableSites(alignment=A);

or

A ~ PhyloCTMC(tree=tree, L=100, Q=jukesCantor());
D = removeUninformativeSites(alignment=A);

[SimonGreenhill]

Nice!

One question, a lot of the language data ascertains all-zero sites (i.e. we don't record data absent in all languages). I guess there could be a 'removeAbsentSites' function too (but then how do you know what site is "absent"? - is it safe to assume "0" is absent? or can you specify it e.g. D = removeAbsentSites(alignment=A, absentstate="0");)

Currently in BEAST2 this is implemented at a pattern level by using a FilteredAlignment which ignores site 1 e.g.:

<data id="x" dataType="standard" name="alignment">
    <sequence id="seq_1" taxon="lang_1" totalcount="2" value="011..."/>
    <sequence id="seq_2" taxon="lang_2" totalcount="2" value="000..."/>
    <!--                                                 HERE ^     -->
</data>

<distribution id="treeLikelihood.x" spec="TreeLikelihood" tree="@Tree.t:x" useAmbiguities="true">
    <data id="orgdata.x" spec="FilteredAlignment" ascertained="true" data="@x" excludefrom="0" excludeto="1" filter="-">
        <userDataType id="TwoStateCovarion.0" spec="beast.evolution.datatype.TwoStateCovarion"/>
    </data>
    ...
</distribution>

...which also means that you could have some weird and wonderful combination of ascertainment if you wanted ("I never bothered to collect 1's in taxon 2 when taxon 1 was already 0" ??) but I don't know of anyone using this, so I can't see of any drawbacks.