def test_print():
# checks that printing does not cause an error
emissions = ["e" + str(x) for x in range(1000)]
chain = bayesian_hmm.Chain(emissions)
print(chain)
repr(chain)
assert True
def test_initialise_chain():
emissions = ["e0", "e1", "e2", "e3"]
states = ["s0", "s1", "s2", "s3", "s4"] # intentionally different length
chain = bayesian_hmm.Chain(emissions)
# check chain initialised correctly
assert chain.emission_sequence == emissions
assert chain.T == len(chain)
assert len(chain) == len(emissions)
assert not chain.initialised_flag
# initialise & check latent states in chain
chain.initialise(states)
assert len(chain) == len(emissions)
assert chain.initialised_flag
assert all(s in states for s in chain.latent_sequence)
def test_chain_resample_emission():
"""
Check that resampled latent sequence conforms to deterministic emission structure
"""
# specify starting emission sequence
len_init = 7
emissions = ["e" + str(x) for x in range(len_init)]
states = ["s" + str(x) for x in range(2 * len_init)]
# specify deterministic latent probabilities
p_initial = {s: 1 / len(states) for s in states}
p_emission = {
s: {emissions[i]: 1 if s == states[i] else 0 for i in range(len_init)}
for s in states
}
p_transition = {s1: {s2: 1 / len(states) for s2 in states} for s1 in states}
# degenerate starting and transition probabilities force the given latent sequence
emission_sequence = emissions * len_init
latent_sequence_resampled = [
states[i % len_init] for i in range(len(emission_sequence))
]
# initialise chain with single emission state
chain = bayesian_hmm.Chain(emission_sequence)
chain.initialise(states)
assert all(s in states for s in chain.latent_sequence)
# resample latent series
chain.latent_sequence = chain.resample_latent_sequence(
(chain.emission_sequence, chain.latent_sequence),
states,
p_initial,
p_emission,
p_transition,
)
# check that engineered latent structure holds
assert all(
chain.latent_sequence[x] == latent_sequence_resampled[x]
for x in range(len(chain))
)
def test_chain_resample_transition():
"""
Check that resampled latent sequence conforms to deterministic transition latent
structure
"""
# specify starting emission sequence
len_init = 10
len_total = 50
emissions = ["e" + str(x % len_init) for x in range(len_total)]
states = ["s" + str(x) for x in range(8)]
# specify deterministic latent probabilities
p_initial = {s: 1 if s == states[-1] else 0 for s in states}
p_emission = {s: {e: 1 / len(set(emissions)) for e in emissions} for s in states}
p_transition = {s1: {s2: 0 for s2 in states} for s1 in states}
for i in range(len(states)):
p_transition[states[i]][states[(i + 1) % len(states)]] = 1
# degenerate starting and transition probabilities force the given latent sequence
latent_sequence_resampled = [states[-1]] + [
states[i % len(states)] for i in range(len(emissions) - 1)
]
# initialise chain with single emission state
chain = bayesian_hmm.Chain(emissions)
chain.initialise([states[0]])
assert all(s == states[0] for s in chain.latent_sequence)
# resample latent series
chain.latent_sequence = chain.resample_latent_sequence(
(chain.emission_sequence, chain.latent_sequence),
states,
p_initial,
p_emission,
p_transition,
)
# check that engineered latent structure holds
assert all(
chain.latent_sequence[x] == latent_sequence_resampled[x]
for x in range(len(chain))
)
def test_empty_chain():
chain_empty = bayesian_hmm.Chain([])
assert len(chain_empty) == 0