def test_uniform_acceptor():
"""Test the uniform acceptor."""
def dist(x, x_0):
return sum(abs(x[key] - x_0[key]) for key in x_0)
distance = pyabc.SimpleFunctionDistance(dist)
acceptor = pyabc.UniformAcceptor()
eps = pyabc.ListEpsilon([1, 4, 2])
x = {'s0': 1.5}
x_0 = {'s0': 0}
ret = acceptor(distance_function=distance,
eps=eps,
x=x,
x_0=x_0,
t=2,
par=None)
assert ret.accept
# now let's test again, including previous time points
acceptor = pyabc.UniformAcceptor(use_complete_history=True)
ret = acceptor(distance_function=distance,
eps=eps,
x=x,
x_0=x_0,
t=2,
par=None)
assert not ret.accept
os.path.join(database_dir, "test_pyabc_" + dbid + ".db"))
# make a file to hold onto these NSEs for our own record
with open(os.path.join(temp_path, "NSEs_" + dbid + ".txt"),
"w") as nse_file:
nse_file.write("NSEs\n")
# define NSE Distance function: Calculate NSE with hydroeval library and the subtract 1 from it to get NSED
def nse(x, x_0):
nse = he.evaluator(he.nse,
simulation_s=np.array(list(x.values())),
evaluation=np.array(list(x_0.values())))[0]
print("nse ", nse)
# make record
with open(os.path.join(temp_path, "NSEs_" + dbid + ".txt"),
"a") as nse_file:
nse_file.write(str(nse) + "\n")
return nse
# NSEs live on [-infinity, 1] and the best NSE is 1
# Distances live on [0, infinity and the best distance is 0
# Therefore, let's make a formula that's informed by NSE, but has range and properties of a distance function
NSED = pyabc.SimpleFunctionDistance(fun=lambda x, x_0: 1 - nse(x, x_0))
# make the pyabc Seq Monte Carlo object
abc = pyabc.ABCSMC(model,
prior,
population_size=pyabc.ConstantPopulationSize(npop),
sampler=sampler,
distance_function=NSED)
# initialize a new run
abc.new(db_path, obs_dict)
# run it!
history = abc.run(max_nr_populations=ngen, minimum_epsilon=0.2)