def run_simulation(self, simulation_input):
sim = self.init_rebound_simulation(simulation_input)
feature_classifier_model = FeatureClassifier()
deep_regressor_model = DeepRegressor()
stability_probability = feature_classifier_model.predict_stable(sim)
median, lower, upper = deep_regressor_model.predict_instability_time(sim, samples=10000)
return {"star_mass": simulation_input.star_mass,
"periods": ",".join([str(planet_period) for planet_period in simulation_input.planet_periods]),
"masses": ",".join([str(mass_value) for mass_value in simulation_input.mass_arr]),
"inclinations": ",".join([str(ecc_value) for ecc_value in simulation_input.inc_arr]),
"eccentricities": ",".join([str(ecc_value) for ecc_value in simulation_input.ecc_arr]),
"arg_periastron": ",".join([str(ecc_value) for ecc_value in simulation_input.omega_arr]),
"stability_probability": stability_probability, "median_expected_instability_time": median}
def setUp(self):
self.model = DeepRegressor(cuda=False)
class TestRegressorClassification(unittest.TestCase):
def setUp(self):
self.model = DeepRegressor(cuda=False)
def test_list_stable(self): # pass list of sims with same size list of tmax
tmax = [1e4, 1e4, 1, 1e4] # test that unstablesim in middle still classified as stable with tmax=1
stable_target = [0, 0, 1, 1]
stable = self.model.predict_stable([hyperbolicsim(), escapesim(), unstablesim(), longstablesim()], tmax=tmax)
self.assertSequenceEqual(stable.tolist(), stable_target)
def test_list_no_tmax(self): # pass list of sims, tmax = None
stable = self.model.predict_stable([vstablesim(), vstablesim()])
self.assertGreater(stable[0], 0.9)
self.assertGreater(stable[1], 0.9)
def test_single_no_tmax(self): # pass list of sims, tmax = None
stable = self.model.predict_stable(vstablesim())
self.assertGreater(stable, 0.9)
def test_mismatched_lists(self):
tmax = [1e4, 1e4, 1] # test that unstablesim in middle still classified as stable with tmax=1
with self.assertRaises(AssertionError):
stable = self.model.predict_stable([hyperbolicsim(), escapesim(), unstablesim(), longstablesim()], tmax=tmax)
def test_sim_unchanged(self):
sim = rebound.Simulation()
sim.add(m=1.)
sim.add(m=1.e-5, P=1.)
sim.add(m=1.e-5, P=2.)
sim.add(m=1.e-5, P=3.)
sim.integrate(1.2)
x0 = sim.particles[1].x
p1 = self.model.predict_stable(sim, seed=0, **SAMPLE_SETTINGS)
self.assertEqual(sim.particles[1].x, x0)
def test_repeat(self):
sim = rebound.Simulation()
sim.add(m=1.)
sim.add(m=1.e-5, P=1.)
sim.add(m=1.e-5, P=2.)
sim.add(m=1.e-5, P=3.)
p1 = self.model.predict_stable(sim, seed=0, **SAMPLE_SETTINGS)
p2 = self.model.predict_stable(sim, seed=0, **SAMPLE_SETTINGS)
self.assertEqual(p1, p2)
# when chaotic realization matters, probs will vary more (eg t_inst=2e4)
def test_galilean_transformation(self):
sim = longstablesim()
sim.move_to_com()
p_com = self.model.predict_stable(sim, seed=0, **SAMPLE_SETTINGS)
sim = longstablesim()
for p in sim.particles:
p.vx += 1000
p_moving = self.model.predict_stable(sim, seed=0, **SAMPLE_SETTINGS)
self.assertAlmostEqual(p_com, p_moving, delta=1e-2)
def test_rescale_distances(self):
sim = longstablesim()
p0 = self.model.predict_stable(sim, seed=0, **SAMPLE_SETTINGS)
sim = longstablesim()
sim = rescale(sim, dscale=1e10, tscale=1, mscale=1)
p1 = self.model.predict_stable(sim, seed=0, **SAMPLE_SETTINGS)
self.assertAlmostEqual(p0, p1, delta=1e-2)
def test_rescale_times(self):
sim = longstablesim()
p0 = self.model.predict_stable(sim, seed=0, **SAMPLE_SETTINGS)
sim = longstablesim()
sim = rescale(sim, dscale=1, tscale=1e10, mscale=1)
p1 = self.model.predict_stable(sim, seed=0, **SAMPLE_SETTINGS)
self.assertAlmostEqual(p0, p1, delta=1e-2)
def test_rescale_masses(self):
sim = longstablesim()
p0 = self.model.predict_stable(sim, seed=0, **SAMPLE_SETTINGS)
sim = longstablesim()
sim = rescale(sim, dscale=1, tscale=1, mscale=1e10)
p1 = self.model.predict_stable(sim, seed=0, **SAMPLE_SETTINGS)
self.assertAlmostEqual(p0, p1, delta=1e-2)
def test_hyperbolic(self):
sim = rebound.Simulation()
sim.add(m=1.)
sim.add(m=1.e-5, a=-1., e=1.2)
sim.add(m=1.e-5, a=2.)
sim.add(m=1.e-5, a=3.)
self.assertEqual(self.model.predict_stable(sim, seed=0, **SAMPLE_SETTINGS), 0)
def test_escape(self):
sim = rebound.Simulation()
sim.add(m=1.)
sim.add(m=1.e-12, P=3.14, e=0.03, l=0.5)
sim.add(m=1.e-12, P=4.396, e=0.03, l=4.8)
sim.add(m=1.e-12, a=100, e=0.999)
self.assertEqual(self.model.predict_stable(sim, seed=0, **SAMPLE_SETTINGS), 0)
def test_unstable_in_short_integration(self):
sim = unstablesim()
self.assertEqual(self.model.predict_stable(sim, seed=0, **SAMPLE_SETTINGS), 0)
def test_solarsystem(self):
sim = solarsystemsim()
median, lower, upper, t_inst_samples = self.model.predict_instability_time(
sim, seed=0,
return_samples=True,
**SAMPLE_SETTINGS)
log_iqr = np.log10(upper) - np.log10(lower)
correct_estimate = (np.average(t_inst_samples > 1e9) > 0.7 )
very_uncertain = (log_iqr > 3)
self.assertTrue(correct_estimate or very_uncertain)
def test_stable(self):
sim = longstablesim()
self.assertGreater(self.model.predict_stable(sim, seed=0, **SAMPLE_SETTINGS), 0.7)
class TestRegressor(unittest.TestCase):
def setUp(self):
self.model = DeepRegressor(cuda=False)
def relative(self, p1, p2):
return abs(p1 - p2) / (p1 + p2) / 2
def test_seed(self):
sim = rebound.Simulation()
sim.add(m=1.)
sim.add(m=1.e-5, P=1.)
sim.add(m=1.e-5, P=2.)
sim.add(m=1.e-5, P=3.)
p1 = np.log10(self.model.predict_instability_time(sim, seed=0, **SAMPLE_SETTINGS)[0])
p2 = np.log10(self.model.predict_instability_time(sim, seed=1, **SAMPLE_SETTINGS)[0])
self.assertTrue(self.relative(p1, p2) < 0.1)
self.assertTrue(self.relative(p1, p2) > 0.0)
def test_prediction(self):
times = []
sims = []
for mass in [1e-4, 5e-5, 3e-5, 1e-5]:
sim = rebound.Simulation()
sim.add(m=1.)
sim.add(m=mass, P=1)
sim.add(m=mass, P=1.3)
sim.add(m=mass, P=1.6)
sims.append(sim)
times = np.log10(self.model.predict_instability_time(sims, **SAMPLE_SETTINGS)[0])
# First one is unstable:
self.assertTrue(times[0] < 4.0)
# Should get more stable:
self.assertTrue(np.all(times[1:] > times[:-1]))
def test_rescale_distances(self):
sim = longstablesim()
t, upper, lower = self.model.predict_instability_time(sim, seed=0, **SAMPLE_SETTINGS)
simr = rescale(sim, dscale=1e10, tscale=1, mscale=1)
tr, upperr, lowerr = self.model.predict_instability_time(simr, seed=0, **SAMPLE_SETTINGS)
self.assertAlmostEqual(t/sim.particles[1].P, tr/simr.particles[1].P, delta=np.abs((upper-lower)/10/sim.particles[1].P))
def test_rescale_times(self):
sim = longstablesim()
t, upper, lower = self.model.predict_instability_time(sim, seed=0, **SAMPLE_SETTINGS)
simr = rescale(sim, dscale=1, tscale=1e10, mscale=1)
tr, upperr, lowerr = self.model.predict_instability_time(simr, seed=0, **SAMPLE_SETTINGS)
self.assertAlmostEqual(t/sim.particles[1].P, tr/simr.particles[1].P, delta=np.abs((upper-lower)/10/sim.particles[1].P))
def test_rescale_masses(self):
sim = longstablesim()
t, upper, lower = self.model.predict_instability_time(sim, seed=0, **SAMPLE_SETTINGS)
simr = rescale(sim, dscale=1, tscale=1, mscale=1e10)
tr, upperr, lowerr = self.model.predict_instability_time(simr, seed=0, **SAMPLE_SETTINGS)
self.assertAlmostEqual(t/sim.particles[1].P, tr/simr.particles[1].P, delta=np.abs((upper-lower)/10/sim.particles[1].P))
def test_time_scaling(self):
times = []
sims = []
mass = 3e-5
for P in [1, 10]:
sim = rebound.Simulation()
sim.add(m=1.)
sim.add(m=mass, P=1*P)
sim.add(m=mass, P=1.3*P)
sim.add(m=mass, P=1.6*P)
sims.append(sim)
# Second time should have ~10x larger inst time.
times = self.model.predict_instability_time(sims, seed=0, **SAMPLE_SETTINGS)[0]
# Should be much larger time:
self.assertGreater(times[1], 5*times[0])
def test_time_scaling_from_integration(self):
times = []
sims = []
mass = 1e-3
for P in [1, 10]:
sim = rebound.Simulation()
sim.add(m=1.)
sim.add(m=mass, P=1*P)
sim.add(m=mass, P=1.3*P)
sim.add(m=mass, P=1.6*P)
sims.append(sim)
# Second time should have ~10x larger inst time.
times = self.model.predict_instability_time(sims, seed=0, **SAMPLE_SETTINGS)[0]
# Should be much larger time:
self.assertGreater(times[1], 5*times[0])
def test_custom_prior(self):
mass = 1e-7
sim = rebound.Simulation()
sim.add(m=1.)
sim.add(m=mass, P=1)
sim.add(m=mass, P=1.3)
sim.add(m=mass, P=1.6)
expected_center = 13.0
prior = lambda logT: np.exp(-(logT - expected_center)**2/2/0.1**2)
times = np.log10(self.model.predict_instability_time(sim, prior_above_9=prior, **SAMPLE_SETTINGS)[0])
self.assertAlmostEqual(times, expected_center, delta=1e-1)
def test_list_time(self):
tinst, lower, upper = self.model.predict_instability_time([hyperbolicsim(), escapesim(), unstablesim(), longstablesim()])
self.assertTrue(np.isnan(tinst[0]))
self.assertLess(tinst[1], 1e4)
self.assertLess(tinst[2], 1e4)
self.assertGreater(tinst[3], 1e4)
class TestRegressor(unittest.TestCase):
def setUp(self):
self.model = DeepRegressor(cuda=False)
def relative(self, p1, p2):
return abs(p1 - p2) / (p1 + p2) / 2
def test_seed(self):
sim = rebound.Simulation()
sim.add(m=1.)
sim.add(m=1.e-5, P=1.)
sim.add(m=1.e-5, P=2.)
sim.add(m=1.e-5, P=3.)
p1 = np.log10(
self.model.predict_instability_time(sim, seed=0,
**SAMPLE_SETTINGS)[0])
p2 = np.log10(
self.model.predict_instability_time(sim, seed=1,
**SAMPLE_SETTINGS)[0])
self.assertTrue(self.relative(p1, p2) < 0.1)
self.assertTrue(self.relative(p1, p2) > 0.0)
def test_prediction(self):
times = []
sims = []
for mass in [1e-4, 5e-5, 3e-5, 1e-5]:
sim = rebound.Simulation()
sim.add(m=1.)
sim.add(m=mass, P=1)
sim.add(m=mass, P=1.3)
sim.add(m=mass, P=1.6)
sims.append(sim)
times = np.log10(
self.model.predict_instability_time(sims, **SAMPLE_SETTINGS)[0])
# First one is unstable:
self.assertTrue(times[0] < 4.0)
# Should get more stable:
self.assertTrue(np.all(times[1:] > times[:-1]))
def test_custom_prior(self):
mass = 1e-7
sim = rebound.Simulation()
sim.add(m=1.)
sim.add(m=mass, P=1)
sim.add(m=mass, P=1.3)
sim.add(m=mass, P=1.6)
expected_center = 13.0
prior = lambda logT: np.exp(-(logT - expected_center)**2 / 2 / 0.1**2)
times = np.log10(
self.model.predict_instability_time(sim,
prior_above_9=prior,
**SAMPLE_SETTINGS)[0])
self.assertAlmostEqual(times, expected_center, delta=1e-2)
def test_list_time(self):
tinst, lower, upper = self.model.predict_instability_time(
[hyperbolicsim(),
escapesim(),
unstablesim(),
longstablesim()])
self.assertTrue(np.isnan(tinst[0]))
self.assertLess(tinst[1], 1e4)
self.assertLess(tinst[2], 1e4)
self.assertGreater(tinst[3], 1e4)