def test_poisson_objective_diploid_unambig():
lengths = np.array([20])
ploidy = 2
seed = 42
alpha, beta = -3., 1.
random_state = np.random.RandomState(seed=seed)
n = lengths.sum()
X_true = random_state.rand(n * ploidy, 3)
dis = euclidean_distances(X_true)
dis[dis == 0] = np.inf
counts = beta * dis**alpha
counts[np.isnan(counts) | np.isinf(counts)] = 0
counts = np.triu(counts, 1)
counts = sparse.coo_matrix(counts)
counts = _format_counts(counts=counts,
lengths=lengths,
ploidy=ploidy,
beta=beta)
obj = poisson.objective(X=X_true,
counts=counts,
alpha=alpha,
lengths=lengths)
assert obj < -1e4
def test_poisson_objective_diploid_partially_ambig_biased():
lengths = np.array([20])
ploidy = 2
seed = 42
alpha, beta = -3., 1.
random_state = np.random.RandomState(seed=seed)
n = lengths.sum()
X_true = random_state.rand(n * ploidy, 3)
dis = euclidean_distances(X_true)
dis[dis == 0] = np.inf
counts = beta * dis**alpha
counts[np.isnan(counts) | np.isinf(counts)] = 0
counts = counts[:, :n] + counts[:, n:]
np.fill_diagonal(counts[:n, :], 0)
np.fill_diagonal(counts[n:, :], 0)
bias = 0.1 + random_state.rand(n)
counts *= np.tile(bias, 2).reshape(-1, 1)
counts *= bias.reshape(-1, 1).T
counts = sparse.coo_matrix(counts)
counts = _format_counts(counts=counts,
lengths=lengths,
ploidy=ploidy,
beta=beta)
obj = poisson.objective(X=X_true,
counts=counts,
alpha=alpha,
lengths=lengths,
bias=bias)
assert obj < -1e4
def test_poisson_objective_multiscale_ambig():
lengths = np.array([20])
ploidy = 2
seed = 42
alpha, beta = -3., 1.
multiscale_rounds = 4
multiscale_factor = 2 ** multiscale_rounds
random_state = np.random.RandomState(seed=seed)
n = lengths.sum()
X_true = random_state.rand(n * ploidy, 3)
dis = euclidean_distances(X_true)
dis[dis == 0] = np.inf
counts = beta * dis ** alpha
counts[np.isnan(counts) | np.isinf(counts)] = 0
counts = counts[:n, :n] + counts[n:, n:] + counts[:n, n:] + counts[n:, :n]
counts = np.triu(counts, 1)
counts = sparse.coo_matrix(counts)
_, _, _, fullres_torm_for_multiscale = preprocess_counts(
counts_raw=counts, lengths=lengths, ploidy=ploidy, normalize=False,
filter_threshold=0., multiscale_factor=1, beta=beta)
counts, _, torm, _ = preprocess_counts(
counts_raw=counts, lengths=lengths, ploidy=ploidy, normalize=False,
filter_threshold=0., multiscale_factor=multiscale_factor, beta=beta,
fullres_torm=fullres_torm_for_multiscale)
multiscale_variances_true = multiscale_optimization.get_multiscale_variances_from_struct(
X_true, lengths=lengths, multiscale_factor=multiscale_factor)
obj = poisson.objective(
X=X_true, counts=counts, alpha=alpha, lengths=lengths, bias=None,
multiscale_factor=multiscale_factor,
multiscale_variances=multiscale_variances_true)
assert obj < -1e4
