def test_constraint_ordering_1():
#m0 = larch.Model.Example(1)
m1 = larch.Model.Example(1)
# m0.utility_co[2] = P.ASC_SR2 + P('hhinc#23') * X.hhinc
# m0.utility_co[3] = P.ASC_SR3P + P('hhinc#23') * X.hhinc
# m0.remove_unused_parameters()
c3 = OrderingBound("hhinc#3", "hhinc#2")
# m0.load_data()
m1.load_data()
m1.constraints = [c3,]
# r0 = m0.maximize_loglike(
# method='slsqp',
# )
r1 = m1.maximize_loglike(
method='slsqp',
)
assert m1['hhinc#2'].value / m1['hhinc#3'].value == approx(1.0, rel=1e-3)
assert r1.message == 'Optimization terminated successfully.'
assert c3.fun(m1.pf.value) == approx(0, abs=1e-5)
# m0.calculate_parameter_covariance()
m1.calculate_parameter_covariance()
# assert m0.pf.loc['hhinc#23', 'std_err'] == approx(0.001397, rel=5e-3)
assert m1.pf.loc['hhinc#2', 'std_err'] == approx(0.001397, rel=5e-3)
assert m1.pf.loc['hhinc#3', 'std_err'] == approx(0.001397, rel=5e-3)
def test_contraint_interpretation():
assert interpret_contraint("aaaa > bbbb") == OrderingBound('bbbb', 'aaaa')
assert interpret_contraint("aaaa < bbbb") == OrderingBound('aaaa', 'bbbb')
assert interpret_contraint("aaaa/bbbb > 3") == RatioBound('aaaa', 'bbbb', min_ratio=3)
assert interpret_contraint("aaaa/bbbb < 3") == RatioBound('aaaa', 'bbbb', max_ratio=3)
assert interpret_contraint("aaaa / bbbb > 3") == RatioBound('aaaa', 'bbbb', min_ratio=3)
assert interpret_contraint("aaaa / bbbb < 3") == RatioBound('aaaa', 'bbbb', max_ratio=3)
assert interpret_contraint("aaaa / bbbb >= 3") == RatioBound('aaaa', 'bbbb', min_ratio=3)
assert interpret_contraint("aaaa / bbbb <= 3") == RatioBound('aaaa', 'bbbb', max_ratio=3)
assert interpret_contraint("3.1 < aaaa / bbbb <= 3.2") == RatioBound('aaaa', 'bbbb', min_ratio=3.1, max_ratio=3.2)
assert interpret_contraint("aaaa > 3") == FixedBound('aaaa', minimum=3)
assert interpret_contraint("aaaa < 3") == FixedBound('aaaa', maximum=3)
assert interpret_contraint("3.1 < aaaa < 3.5") == FixedBound('aaaa', minimum=3.1, maximum=3.5)
assert interpret_contraint("aaaa/bbbb > 1/2") == RatioBound('aaaa', 'bbbb', min_ratio=1/2)
assert interpret_contraint("aaaa/bbbb < 1/2") == RatioBound('aaaa', 'bbbb', max_ratio=1/2)
assert interpret_contraint("aaaa / bbbb > 1/2") == RatioBound('aaaa', 'bbbb', min_ratio=1/2)
assert interpret_contraint("aaaa / bbbb < 1/2") == RatioBound('aaaa', 'bbbb', max_ratio=1/2)
assert interpret_contraint("aaaa / bbbb >= 1/2") == RatioBound('aaaa', 'bbbb', min_ratio=1/2)
assert interpret_contraint("aaaa / bbbb <= 1/2") == RatioBound('aaaa', 'bbbb', max_ratio=1/2)
assert interpret_contraint("1/4 < aaaa / bbbb <= 3/4") == RatioBound('aaaa', 'bbbb', min_ratio=1/4, max_ratio=3/4)
assert interpret_contraint("aaaa > 3/4") == FixedBound('aaaa', minimum=3/4)
assert interpret_contraint("aaaa < 3/4") == FixedBound('aaaa', maximum=3/4)
assert interpret_contraint("1/4 < aaaa < 3/4") == FixedBound('aaaa', minimum=1/4, maximum=3/4)
def test_parameter_summary():
import larch
from larch.model.constraints import RatioBound, OrderingBound, FixedBound
m0 = larch.Model.Example(1)
m0.lock_value('tottime', -0.05)
m0.set_value('totcost', maximum=-0.005, minimum=-0.02)
m0.constraints.append(OrderingBound("hhinc#3 <= hhinc#2"))
m0.load_data()
r0 = m0.maximize_loglike(
method='slsqp',
options={'ftol': 1e-09},
quiet=True,
)
m0.calculate_parameter_covariance()
ps = m0.parameter_summary('df').data
try:
assert ps.loc[('LOS', 'totcost'), 'Value'] == "-0.00500"
assert ps.loc[('LOS', 'tottime'), 'Value'] == "-0.0500"
assert ps.loc[('Income', 'hhinc#2'), 'Value'] == "-0.00159"
assert ps.loc[('Income', 'hhinc#3'), 'Value'] == "-0.00159"
assert ps.loc[('LOS', 'totcost'), 'Std Err'] == " 0.00"
assert ps.loc[('LOS', 'tottime'), 'Std Err'] == " 0.00"
assert ps.loc[('Income', 'hhinc#2'), 'Std Err'] == " 0.00140"
assert ps.loc[('Income', 'hhinc#3'), 'Std Err'] == " 0.00140"
assert ps.loc[('LOS', 'totcost'), 't Stat'] == " NA"
assert ps.loc[('LOS', 'tottime'), 't Stat'] == " NA"
assert ps.loc[('Income', 'hhinc#2'), 't Stat'] == "-1.14"
assert ps.loc[('Income', 'hhinc#3'), 't Stat'] == "-1.14"
assert ps.loc[('LOS', 'totcost'), 'Signif'] == ""
assert ps.loc[('LOS', 'tottime'), 'Signif'] == ""
assert ps.loc[('Income', 'hhinc#2'), 'Signif'] == ""
assert ps.loc[('Income', 'hhinc#3'), 'Signif'] == ""
assert ps.loc[('LOS', 'totcost'), 'Constrained'] == "totcost ≤ -0.005"
assert ps.loc[('LOS', 'tottime'), 'Constrained'] == "fixed value"
assert ps.loc[('Income', 'hhinc#2'),
'Constrained'] == "hhinc#3 ≤ hhinc#2"
assert ps.loc[('Income', 'hhinc#3'),
'Constrained'] == "hhinc#3 ≤ hhinc#2"
except:
print(ps.iloc[:, :3])
print(ps.iloc[:, 3:])
raise
def test_constrained_optimization():
from larch.numba import example
m = example(1)
from larch.model.constraints import RatioBound, OrderingBound
m.set_value("totcost", -0.001, maximum=0)
m.set_value("tottime", maximum=0)
m.constraints = [
RatioBound(P("totcost"),
P("tottime"),
min_ratio=0.1,
max_ratio=1.0,
scale=1),
OrderingBound(P("ASC_WALK"), P("ASC_BIKE")),
]
m.load_data()
r = m.maximize_loglike(method='slsqp')
assert r.loglike == approx(-3647.76149525901)
x = {
'ASC_BIKE': -0.8087472748965431,
'ASC_SR2': -2.193449976582375,
'ASC_SR3P': -3.744188833076006,
'ASC_TRAN': -0.7603092451373663,
'ASC_WALK': -0.8087472751682576,
'hhinc#2': -0.0021699330391421407,
'hhinc#3': 0.0003696763687090173,
'hhinc#4': -0.00509836274463602,
'hhinc#5': -0.0431749907425252,
'hhinc#6': -0.002373556571769923,
'totcost': -0.004910169034222911,
'tottime': -0.04790588175791953,
}
assert dict(r.x) == approx(x)
assert r.iteration_number == 48
m.set_values("null")
m.set_value("totcost", -0.001, maximum=0)
r2 = m.maximize_loglike(method='slsqp', bhhh_start=3)
assert r2.iteration_number == 24
assert r2.loglike == approx(-3647.76149525901)
assert dict(r2.x) == approx(x, rel=1e-2)
def test_parameter_summary():
import larch
from larch.model.constraints import RatioBound, OrderingBound, FixedBound
m0 = larch.Model.Example(1)
m0.lock_value('tottime', -0.05)
m0.set_value('totcost', maximum=-0.005, minimum=-0.02)
m0.constraints.append(OrderingBound("hhinc#3 <= hhinc#2"))
m0.load_data()
# constraint tests are unstable across platforms
# r0 = m0.maximize_loglike(
# method='slsqp',
# options={'ftol': 1e-09},
# quiet=True,
# )
m0.set_values({
'ASC_BIKE': -2.401574701017008,
'ASC_SR2': -2.2234574452767037,
'ASC_SR3P': -3.630001719635557,
'ASC_TRAN': -0.7000998451196682,
'ASC_WALK': -0.25495834390295924,
'hhinc#2': -0.0015948074747861667,
'hhinc#3': -0.0015948074747729564,
'hhinc#4': -0.005385664916039662,
'hhinc#5': -0.01284021327300144,
'hhinc#6': -0.009658848920781143,
'totcost': -0.005000000000003634,
'tottime': -0.05,
})
m0.calculate_parameter_covariance()
ps = m0.parameter_summary('df').data
stable_df(ps, 'parameter_summary_test')
try:
assert ps.loc[('LOS', 'totcost'), 'Value'] == "-0.00500"
assert ps.loc[('LOS', 'tottime'), 'Value'] == "-0.0500"
assert ps.loc[('Income', 'hhinc#2'), 'Value'] == "-0.00159"
assert ps.loc[('Income', 'hhinc#3'), 'Value'] == "-0.00159"
assert ps.loc[('LOS', 'totcost'), 'Std Err'] == " NA"
assert ps.loc[('LOS', 'tottime'), 'Std Err'] == " NA"
assert ps.loc[('Income', 'hhinc#2'), 'Std Err'] == " 0.00140"
assert ps.loc[('Income', 'hhinc#3'), 'Std Err'] == " 0.00140"
assert ps.loc[('LOS', 'totcost'), 't Stat'] == " NA"
assert ps.loc[('LOS', 'tottime'), 't Stat'] == " NA"
assert ps.loc[('Income', 'hhinc#2'), 't Stat'] == "-1.14"
assert ps.loc[('Income', 'hhinc#3'), 't Stat'] == "-1.14"
assert ps.loc[('LOS', 'totcost'), 'Signif'] == "[***]"
assert ps.loc[('LOS', 'tottime'), 'Signif'] == ""
assert ps.loc[('Income', 'hhinc#2'), 'Signif'] == ""
assert ps.loc[('Income', 'hhinc#3'), 'Signif'] == ""
assert ps.loc[('LOS', 'totcost'), 'Constrained'] == "totcost ≤ -0.005"
assert ps.loc[('LOS', 'tottime'), 'Constrained'] == "fixed value"
assert ps.loc[('Income', 'hhinc#2'), 'Constrained'] == "hhinc#3 ≤ hhinc#2"
assert ps.loc[('Income', 'hhinc#3'), 'Constrained'] == "hhinc#3 ≤ hhinc#2"
except:
print(ps.iloc[:,:3])
print(ps.iloc[:,3:])
raise
def test_multi_constraints():
#m0 = larch.Model.Example(1)
m1 = larch.Model.Example(1)
#m0.lock_value('tottime', -0.1)
m1.set_value('tottime', maximum=-0.1)
#m0.utility_co[2] = P.ASC_SR2 + P('hhinc#23') * X.hhinc
#m0.utility_co[3] = P.ASC_SR3P + P('hhinc#23') * X.hhinc
#m0.remove_unused_parameters()
m1.constraints.append(OrderingBound("hhinc#3 <= hhinc#2"))
#m0.load_data()
m1.load_data()
# r0 = m0.torch.maximize_loglike(
# method='slsqp',
# options={'ftol': 1e-09},
# )
r1 = m1.maximize_loglike(
method='slsqp',
options={'ftol': 1e-09},
quiet=True,
)
# assert r1.message == 'Positive directional derivative for linesearch'
m1.constraints.rescale(0.1)
r1 = m1.maximize_loglike(
method='slsqp',
options={'ftol': 1e-09},
quiet=True,
)
# assert r1.message == 'Positive directional derivative for linesearch'
m1.constraints.rescale(0.01)
r1 = m1.maximize_loglike(
method='slsqp',
options={'ftol': 1e-09},
quiet=True,
)
assert r1.loglike == approx(-3725.439832484451)
assert r1.message == 'Optimization terminated successfully.'
#m0.calculate_parameter_covariance()
m1.calculate_parameter_covariance()
assert dict(m1.pf['value']) == approx({
'ASC_BIKE': -1.8336226490217296,
'ASC_SR2': -2.008575006161787,
'ASC_SR3P': -3.372494233637122,
'ASC_TRAN': 0.23543257975862672,
'ASC_WALK': 1.1106100812079671,
'hhinc#2': -0.0016453210270738865,
'hhinc#3': -0.001645187376205547,
'hhinc#4': -0.0055451755658633,
'hhinc#5': -0.012519484434344206,
'hhinc#6': -0.01079425276590098,
'totcost': -0.005093387068047046,
'tottime': -0.1
}, rel=1e-2)
# Problem: Travis is failing on this test, giving
# tottime as NaN. Disabling until we find a
# similar problem we can test and diagnose
assert dict(m1.pf['std_err']) == approx({
'ASC_BIKE': 0.30013699696981455,
'ASC_SR2': 0.09856612504524623,
'ASC_SR3P': 0.1250682262709841,
'ASC_TRAN': 0.12311806485021345,
'ASC_WALK': 0.17807446535212992,
'hhinc#2': 0.0014236497001240312,
'hhinc#3': 0.001423649700124031,
'hhinc#4': 0.0019212691279916797,
'hhinc#5': 0.005248506439174827,
'hhinc#6': 0.003146916477051659,
'totcost': 0.0002466030726673273,
'tottime': 0
}, abs=1e-10, rel=5e-2)
assert dict(m1.pf['unconstrained_std_err']) == approx({
'ASC_BIKE': 0.3032386370524189,
'ASC_SR2': 0.10695557451038115,
'ASC_SR3P': 0.18033743391454088,
'ASC_TRAN': 0.1439964957540911,
'ASC_WALK': 0.20789603304640025,
'hhinc#2': 0.0015638823624232918,
'hhinc#3': 0.002640071485204856,
'hhinc#4': 0.00192260391409179,
'hhinc#5': 0.0052485070222418155,
'hhinc#6': 0.0031494462988719317,
'totcost': 0.00024742269603570923,
'tottime': 0.004377029904009681
}, rel=5e-2)