Fixup tests for lshmm 0.0.6

[jeromekelleher]

Recent lshmm release breaks some tests in tskit test suite: https://github.com/astheeggeggs/lshmm/releases/tag/v0.0.6

Would you mind look at this please @szhan

[jeromekelleher]

Note we'll need to update the pins in requirements

[szhan]

When doing pytest -vs test_haplotype_matching.py, I encountered a bunch of failed tests that hit value error.

ValueError: First value in the recombination probability array must be zero.

I think it is caused by a line assigning recombination probabilities in, for example:

def check_viterbi(ts, h, recombination=None, mutation=None):
    h = np.array(h).astype(np.int8)
    m = ts.num_sites
    assert len(h) == m
    if recombination is None:
        recombination = np.zeros(ts.num_sites) + 1e-9 # First value is positive.

Similarly, check_forward_matrix and check_backward_matrix.

[szhan]

I'm not sure if this is intended, but mutation probability is set to 0 across all sites by default here.

def check_viterbi(ts, h, recombination=None, mutation=None):
    h = np.array(h).astype(np.int8)
    m = ts.num_sites
    assert len(h) == m
    if recombination is None:
        recombination = np.zeros(ts.num_sites) + 1e-9
    if mutation is None:
        mutation = np.zeros(ts.num_sites)

Similarly, check_forward_matrix and check_backward_matrix.

[szhan]

I also have some other failed tests. Should we investigate these failed tests in separate issues?

[szhan]

I've added stricter requirements on the input data, e.g., having all variable sites in the ref. panel and query combined. It is causing some tests to fail, I think.

[szhan]

I'm encountering these errors when running the tests locally.

• First value in the recomb. prob. array is 0. This can be fixed by setting the first value to 0.
• Some invariant sites in the samples.
• ValueError: More than one root exists. Use tree.roots instead.

[szhan]

Note that we are now using lshmm 0.0.8.