def test_production2(self):
# single factor attr
for model_subtype in ["pow", "exp"]:
production = models.Model("production", model_subtype,
Test.dataset, "MIGRATIONS",
"O_GEOGRAPHY_CODE",
["HOUSEHOLDS", "JOBS"], "DISTANCE")
self.assertTrue(
rmse(
production(xd=[
Test.dataset.HOUSEHOLDS.values,
Test.dataset.JOBS.values
]), production.impl.yhat) < 1e-10)
Test.dataset["HH_CHANGED"] = Test.dataset.HOUSEHOLDS
Test.dataset.loc[Test.dataset.D_GEOGRAPHY_CODE == "E07000178",
"HH_CHANGED"] = Test.dataset.loc[
Test.dataset.D_GEOGRAPHY_CODE == "E07000178",
"HOUSEHOLDS"] + 300000
Test.dataset["J_CHANGED"] = Test.dataset.JOBS
Test.dataset.loc[Test.dataset.D_GEOGRAPHY_CODE == "E07000178",
"J_CHANGED"] = Test.dataset.loc[
Test.dataset.D_GEOGRAPHY_CODE == "E07000178",
"JOBS"] + 300000
self.assertTrue(
rmse(
production(xd=[
Test.dataset.HH_CHANGED.values,
Test.dataset.HH_CHANGED.values
]), production.impl.yhat) > 1.0)
def test_attraction(self):
# single factor prod
for model_subtype in ["pow", "exp"]:
attraction = models.Model("attraction", model_subtype,
Test.dataset, "MIGRATIONS", "PEOPLE",
"D_GEOGRAPHY_CODE", "DISTANCE")
self.assertTrue(
rmse(attraction(xo=Test.dataset.PEOPLE.values),
attraction.impl.yhat) < 1e-10)
Test.dataset["P_CHANGED"] = Test.dataset.PEOPLE
Test.dataset.loc[Test.dataset.D_GEOGRAPHY_CODE == "E07000178",
"P_CHANGED"] = Test.dataset.loc[
Test.dataset.D_GEOGRAPHY_CODE == "E07000178",
"HOUSEHOLDS"] + 300000
self.assertTrue(
rmse(attraction(xo=Test.dataset.P_CHANGED.values),
attraction.impl.yhat) > 1.0)
def test_stats(self):
np.random.seed(0)
x = np.random.normal(0, 1, 10000)
y = np.random.normal(0, 1, 10000)
# R2
self.assertEqual(r2(x, x), 1.0)
self.assertEqual(r2(x, -x), 1.0) # rho=-1
self.assertTrue(-0.0001 < r2(x, y) < 0.0001)
#
rho = 0.5
res = r2(x, rho * x + np.sqrt(1 - rho**2) * y)
self.assertTrue(rho**2 - 0.015 < res < rho**2 + 0.015)
z = np.zeros(10000)
self.assertEqual(rmse(x, z), np.sqrt(np.mean(x * x)))
self.assertEqual(rmse(x, x), 0.0)
self.assertEqual(rmse(x, -x), np.sqrt(np.mean(4 * x * x)))
def test_gravity(self):
# single factor prod and attr
for model_subtype in ["pow", "exp"]:
gravity = models.Model("gravity", model_subtype, Test.dataset,
"MIGRATIONS", "PEOPLE", "HOUSEHOLDS",
"DISTANCE")
self.assertTrue(
rmse(
gravity(Test.dataset.PEOPLE.values, Test.dataset.
HOUSEHOLDS.values), gravity.impl.yhat) < 1e-10)
Test.dataset["HH_CHANGED"] = Test.dataset.HOUSEHOLDS
Test.dataset.loc[Test.dataset.D_GEOGRAPHY_CODE == "E07000178",
"HH_CHANGED"] = Test.dataset.loc[
Test.dataset.D_GEOGRAPHY_CODE == "E07000178",
"HOUSEHOLDS"] + 300000
self.assertTrue(
rmse(
gravity(Test.dataset.PEOPLE.values, Test.dataset.
HH_CHANGED.values), gravity.impl.yhat) > 1.0)
def test_gravity2(self):
# multi factor emission
for model_subtype in ["pow", "exp"]:
gravity = models.Model("gravity", model_subtype, Test.dataset,
"MIGRATIONS", "PEOPLE",
["HOUSEHOLDS", "JOBS"], "DISTANCE")
self.assertTrue(
rmse(
gravity(Test.dataset.PEOPLE.values, [
Test.dataset.HOUSEHOLDS.values,
Test.dataset.JOBS.values
]), gravity.impl.yhat) < 1e-10)
# multi factor attraction
for model_subtype in ["pow", "exp"]:
gravity = models.Model("gravity", model_subtype, Test.dataset,
"MIGRATIONS", ["PEOPLE", "JOBS"],
"HOUSEHOLDS", "DISTANCE")
self.assertTrue(
rmse(
gravity(
[Test.dataset.PEOPLE.values, Test.dataset.JOBS.values],
Test.dataset.HOUSEHOLDS.values), gravity.impl.yhat) <
1e-10)