def simulate_one():
return pyabc.Particle(m=0,
parameter={},
weight=0,
accepted_sum_stats=[],
accepted_distances=[],
accepted=True)
def simulate_one():
return pyabc.Particle(
m=0,
parameter={},
weight=0.1,
sum_stat={},
distance=42,
accepted=True,
)
def test_predictor_sumstat():
"""Test predictor sumstat."""
sumstat = PredictorSumstat(LinearPredictor(), fit_ixs={3, 5})
assert not sumstat.requires_calibration()
assert sumstat.is_adaptive()
rng = np.random.Generator(np.random.PCG64(0))
n_sample, n_y, n_p = 1000, 100, 3
ys = rng.normal(size=(n_sample, n_y))
ps = rng.normal(size=(n_sample, n_p))
particles = []
for y, p in zip(ys, ps):
particles.append(
pyabc.Particle(
m=0,
parameter=pyabc.Parameter(
{f"p{ix}": val
for ix, val in enumerate(p)}),
sum_stat={f"s{ix}": val
for ix, val in enumerate(y)},
distance=100 + 1 * rng.normal(),
weight=1 + 0.01 * rng.normal(),
))
total_weight = sum(p.weight for p in particles)
for p in particles:
p.weight /= total_weight
sample = pyabc.Sample.from_population(pyabc.Population(particles))
x = particles[0].sum_stat
# nothing should happen in initialize
sumstat.initialize(t=0, get_sample=lambda: sample, x_0=x, total_sims=0)
assert sumstat(x).shape == (n_y, )
assert (sumstat(x) == ys[0]).all()
assert len(sumstat.get_ids()) == n_y
assert sumstat.get_ids() == [f"s{ix}" for ix in range(n_y)]
# 3 is a fit index --> afterwards the output size should have changed
sumstat.update(t=3, get_sample=lambda: sample, total_sims=0)
assert sumstat(x).shape == (n_p, )
assert len(sumstat.get_ids()) == n_p
# change fit indices
sumstat = PredictorSumstat(LinearPredictor(), fit_ixs={0, 1})
sumstat.initialize(t=0, get_sample=lambda: sample, x_0=x, total_sims=0)
assert sumstat(x).shape == (n_p, )
def simulate_one():
accepted = np.random.randint(2)
return pyabc.Particle(0, {}, 0.1, [], [], accepted)
