def fit(self, framework='GLM', Quasi=False):
"""
Method that fits a particular count model usign the appropriate
estimation technique. Models include Poisson GLM, Negative Binomial GLM,
Quasi-Poisson - at the moment Poisson GLM is the only option.
TODO: add zero inflated variants and hurdle variants.
Parameters
----------
framework : string
estimation framework; default is GLM
"GLM" | "QUASI" |
"""
if (framework.lower() == 'glm'):
if not Quasi:
results = GLM(self.y,
self.X,
family=Poisson(),
constant=self.constant).fit()
else:
results = GLM(self.y,
self.X,
family=QuasiPoisson(),
constant=self.constant).fit()
return CountModelResults(results)
else:
raise NotImplemented(
'Poisson GLM is the only count model currently implemented')
def test_PoissonGLM(self):
model = CountModel(self.y, self.X, family=Poisson())
results = model.fit('GLM')
np.testing.assert_allclose(results.params, [3.92159085, 0.01183491,
-0.01371397], atol=1.0e-8)
self.assertIsInstance(results.family, Poisson)
self.assertEqual(results.n, 49)
self.assertEqual(results.k, 3)
self.assertEqual(results.df_model, 2)
self.assertEqual(results.df_resid, 46)
np.testing.assert_allclose(results.yhat,
[ 51.26831574, 50.15022766, 40.06142973, 34.13799739,
28.76119226, 42.6836241 , 55.64593703, 34.08277997,
40.90389582, 37.19727958, 23.47459217, 26.12384057,
29.78303507, 25.96888223, 29.14073823, 26.04369592,
34.18996367, 32.28924005, 27.42284396, 72.69207879,
33.05316347, 36.52276972, 49.2551479 , 35.33439632,
24.07252457, 31.67153709, 27.81699478, 25.38021219,
24.31759259, 23.13586161, 48.40724678, 48.57969818,
31.92596006, 43.3679231 , 34.32925819, 51.78908089,
34.49778584, 27.56236198, 48.34273194, 57.50829097,
50.66038226, 54.68701352, 35.77103116, 43.21886784,
40.07615759, 49.98658004, 43.13352883, 40.28520774,
46.28910294])
np.testing.assert_allclose(results.cov_params,
[[ 1.70280610e-02, -6.18628383e-04, -2.21386966e-04],
[ -6.18628383e-04, 2.61733917e-05, 6.77496445e-06],
[ -2.21386966e-04, 6.77496445e-06, 3.75463502e-06]])
np.testing.assert_allclose(results.std_err, [ 0.13049161, 0.00511599,
0.00193769], atol=1.0e-8)
np.testing.assert_allclose(results.pvalues, [ 2.02901657e-198,
2.07052532e-002, 1.46788805e-012])
np.testing.assert_allclose(results.tvalues, [ 30.0524361 , 2.31331634,
-7.07748998])
np.testing.assert_allclose(results.resid,
[ 28.73168426, -5.15022766, -14.06142973, -1.13799739,
-5.76119226, -13.6836241 , 19.35406297, 2.91722003,
12.09610418, 58.80272042, -3.47459217, -6.12384057,
12.21696493, 17.03111777, -11.14073823, -7.04369592,
7.81003633, 27.71075995, 3.57715604, 8.30792121,
-13.05316347, -6.52276972, -1.2551479 , 17.66560368,
-6.07252457, -11.67153709, 6.18300522, -2.38021219,
7.68240741, -1.13586161, -16.40724678, -8.57969818,
-7.92596006, -15.3679231 , -7.32925819, -15.78908089,
8.50221416, -4.56236198, -8.34273194, 4.49170903,
-8.66038226, -10.68701352, -9.77103116, -9.21886784,
-12.07615759, 26.01341996, -1.13352883, -13.28520774,
-10.28910294])
self.assertAlmostEqual(results.deviance, 230.46013824817649)
self.assertAlmostEqual(results.llf, -247.42592089969378)
self.assertAlmostEqual(results.AIC, 500.85184179938756)
self.assertAlmostEqual(results.D2, 0.388656011675)
self.assertAlmostEqual(results.adj_D2, 0.36207583826952761)
def test_GS_NN(self):
est_Int = self.GS_NN.by_col(' est_Intercept')
se_Int = self.GS_NN.by_col(' se_Intercept')
t_Int = self.GS_NN.by_col(' t_Intercept')
est_OCC = self.GS_NN.by_col(' est_OCC_TEC')
se_OCC = self.GS_NN.by_col(' se_OCC_TEC')
t_OCC = self.GS_NN.by_col(' t_OCC_TEC')
est_OWN = self.GS_NN.by_col(' est_OWNH')
se_OWN = self.GS_NN.by_col(' se_OWNH')
t_OWN = self.GS_NN.by_col(' t_OWNH')
est_POP = self.GS_NN.by_col(' est_POP65')
se_POP = self.GS_NN.by_col(' se_POP65')
t_POP = self.GS_NN.by_col(' t_POP65')
est_UNEMP = self.GS_NN.by_col(' est_UNEMP')
se_UNEMP = self.GS_NN.by_col(' se_UNEMP')
t_UNEMP = self.GS_NN.by_col(' t_UNEMP')
yhat = self.GS_NN.by_col(' yhat')
pdev = np.array(self.GS_NN.by_col(' localpdev')).reshape((-1, 1))
model = GWR(self.coords,
self.y,
self.X,
bw=50,
family=Poisson(),
kernel='gaussian',
fixed=False)
rslt = model.fit()
AICc = get_AICc(rslt)
AIC = get_AIC(rslt)
BIC = get_BIC(rslt)
self.assertAlmostEquals(np.floor(AICc), 21070.0)
self.assertAlmostEquals(np.floor(AIC), 21069.0)
self.assertAlmostEquals(np.floor(BIC), 21111.0)
np.testing.assert_allclose(est_Int, rslt.params[:, 0], rtol=1e-04)
np.testing.assert_allclose(se_Int, rslt.bse[:, 0], rtol=1e-02)
np.testing.assert_allclose(t_Int, rslt.tvalues[:, 0], rtol=1e-02)
np.testing.assert_allclose(est_OCC, rslt.params[:, 1], rtol=1e-03)
np.testing.assert_allclose(se_OCC, rslt.bse[:, 1], rtol=1e-02)
np.testing.assert_allclose(t_OCC, rslt.tvalues[:, 1], rtol=1e-02)
np.testing.assert_allclose(est_OWN, rslt.params[:, 2], rtol=1e-04)
np.testing.assert_allclose(se_OWN, rslt.bse[:, 2], rtol=1e-02)
np.testing.assert_allclose(t_OWN, rslt.tvalues[:, 2], rtol=1e-02)
np.testing.assert_allclose(est_POP, rslt.params[:, 3], rtol=1e-02)
np.testing.assert_allclose(se_POP, rslt.bse[:, 3], rtol=1e-02)
np.testing.assert_allclose(t_POP, rslt.tvalues[:, 3], rtol=1e-02)
np.testing.assert_allclose(est_UNEMP, rslt.params[:, 4], rtol=1e-02)
np.testing.assert_allclose(se_UNEMP, rslt.bse[:, 4], rtol=1e-02)
np.testing.assert_allclose(t_UNEMP, rslt.tvalues[:, 4], rtol=1e-02)
np.testing.assert_allclose(yhat, rslt.mu, rtol=1e-04)
np.testing.assert_allclose(pdev, rslt.pDev, rtol=1e-05)
def test_BS_F(self):
est_Int = self.BS_F.by_col(' est_Intercept')
se_Int = self.BS_F.by_col(' se_Intercept')
t_Int = self.BS_F.by_col(' t_Intercept')
est_OCC = self.BS_F.by_col(' est_OCC_TEC')
se_OCC = self.BS_F.by_col(' se_OCC_TEC')
t_OCC = self.BS_F.by_col(' t_OCC_TEC')
est_OWN = self.BS_F.by_col(' est_OWNH')
se_OWN = self.BS_F.by_col(' se_OWNH')
t_OWN = self.BS_F.by_col(' t_OWNH')
est_POP = self.BS_F.by_col(' est_POP65')
se_POP = self.BS_F.by_col(' se_POP65')
t_POP = self.BS_F.by_col(' t_POP65')
est_UNEMP = self.BS_F.by_col(' est_UNEMP')
se_UNEMP = self.BS_F.by_col(' se_UNEMP')
t_UNEMP = self.BS_F.by_col(' t_UNEMP')
yhat = self.BS_F.by_col(' yhat')
pdev = np.array(self.BS_F.by_col(' localpdev')).reshape((-1, 1))
model = GWR(self.coords,
self.y,
self.X,
bw=26029.625,
family=Poisson(),
kernel='bisquare',
fixed=True)
rslt = model.fit()
AICc = get_AICc(rslt)
AIC = get_AIC(rslt)
BIC = get_BIC(rslt)
self.assertAlmostEquals(np.floor(AICc), 13294.0)
self.assertAlmostEquals(np.floor(AIC), 13247.0)
self.assertAlmostEquals(np.floor(BIC), 13485.0)
np.testing.assert_allclose(est_Int, rslt.params[:, 0], rtol=1e-05)
np.testing.assert_allclose(se_Int, rslt.bse[:, 0], rtol=1e-03)
np.testing.assert_allclose(t_Int, rslt.tvalues[:, 0], rtol=1e-03)
np.testing.assert_allclose(est_OCC, rslt.params[:, 1], rtol=1e-04)
np.testing.assert_allclose(se_OCC, rslt.bse[:, 1], rtol=1e-02)
np.testing.assert_allclose(t_OCC, rslt.tvalues[:, 1], rtol=1e-02)
np.testing.assert_allclose(est_OWN, rslt.params[:, 2], rtol=1e-04)
np.testing.assert_allclose(se_OWN, rslt.bse[:, 2], rtol=1e-03)
np.testing.assert_allclose(t_OWN, rslt.tvalues[:, 2], rtol=1e-03)
np.testing.assert_allclose(est_POP, rslt.params[:, 3], rtol=1e-04)
np.testing.assert_allclose(se_POP, rslt.bse[:, 3], rtol=1e-02)
np.testing.assert_allclose(t_POP, rslt.tvalues[:, 3], rtol=1e-02)
np.testing.assert_allclose(est_UNEMP, rslt.params[:, 4], rtol=1e-04)
np.testing.assert_allclose(se_UNEMP, rslt.bse[:, 4], rtol=1e-02)
np.testing.assert_allclose(t_UNEMP, rslt.tvalues[:, 4], rtol=1e-02)
np.testing.assert_allclose(yhat, rslt.mu, rtol=1e-05)
np.testing.assert_allclose(pdev, rslt.pDev, rtol=1e-05)
def test_Dispersion(self):
model = CountModel(self.y, self.X, family=Poisson())
results = model.fit('GLM')
phi = phi_disp(results)
alpha1 = alpha_disp(results)
alpha2 = alpha_disp(results, lambda x: x**2)
np.testing.assert_allclose(phi, [5.39968689, 2.3230411, 0.01008847],
atol=1.0e-8)
np.testing.assert_allclose(alpha1, [4.39968689, 2.3230411, 0.01008847],
atol=1.0e-8)
np.testing.assert_allclose(alpha2,
[0.10690133, 2.24709978, 0.01231683],
atol=1.0e-8)
def __init__(self, y, X, family=Poisson(), constant=True):
self.y = self._check_counts(y)
self.X = X
self.constant = constant