def _run_family(self, family_name):
family = self.families[family_name]
glm_res = glm.fit_glm(self.input, self.exog, family)
f = glm.f_test(glm_res.beta, self.contrast, glm_res.normalized_cov_params, glm_res.scale)
f = np.reshape(f, (len(f),))
smoothed_f = glm.f_test(glm_res.beta, self.contrast, glm_res.normalized_cov_params, glm_res.scale, smoothing=np.zeros(3))
got = {
'beta' : glm_res.beta,
'mu' : glm_res.mu,
'weights' : glm_res.weights,
'f_values' : f }
for (name, values) in got.items():
print("Doing " + name)
filename = 'pade/test/glm/{0}_{1}.txt'.format(family_name, name)
expected = np.genfromtxt(filename)
np.testing.assert_almost_equal(expected, values)
for i in range(3):
np.testing.assert_almost_equal(expected, values)
filename = 'pade/test/glm/{0}_{1}.txt'.format(family_name, 'f_values')
expected = np.genfromtxt(filename)
for i in range(3):
np.testing.assert_almost_equal(expected, smoothed_f[i].reshape(len(smoothed_f[i])))
def new_glm(y, x, family, contrast):
models = []
fitteds = []
fs = []
f = None
(params, mu, weights, cov_p, scale) = glm.fit_glm(y, x, family)
f = glm.f_test(params, contrast, cov_p, scale, smoothing=np.arange(10))
def new_glm(y, x, family, contrast):
models = []
fitteds = []
fs = []
f = None
(params, mu, weights, cov_p, scale) = glm.fit_glm(y, x, family)
f = glm.f_test(params, contrast, cov_p, scale, smoothing=np.arange(10))
def new_glm(y, x, family, contrast):
models = []
fitteds = []
fs = []
f = None
(params, mu, weights, cov_p, scale) = glm.fit_glm(y, x, family)
f = glm.f_test(params, contrast, cov_p, scale)
return GlmResults(y, x, family, contrast,
params, mu, weights, f)
def __call__(self, data):
num_regressors = len(self.layout_full)
num_restrictions = num_regressors - 1
x = self.x
contrast = self.contrast
y = data
family = self.family
if np.ndim(y) == 1:
raise Exception("I only do 2d arrays")
shrink = self.shrink
if shrink:
ctor = GLM_FAMILIES['negative_binomial']
family = ctor()
print(
"Using shrinkage estimator of dispersal in negative binomial GLM."
)
raise Exception("yay!")
m = len(y)
alphas = self.alphas
if alphas is None:
res = np.zeros(m)
else:
res = np.zeros((len(alphas), m))
glm_res = glm.fit_glm(y, x, family)
kwargs = {}
if self.alphas is not None:
kwargs['smoothing'] = alphas
res = glm.f_test(glm_res.beta, contrast, glm_res.normalized_cov_params,
glm_res.scale, **kwargs)
# For some reason the f-test returns an array with two extra dims
return res.reshape(res.shape[:-2])
def __call__(self, data):
num_regressors = len(self.layout_full)
num_restrictions = num_regressors - 1
x = self.x
contrast = self.contrast
y = data
family = self.family
if np.ndim(y) == 1:
raise Exception("I only do 2d arrays")
shrink = self.shrink
if shrink:
ctor = GLM_FAMILIES['negative_binomial']
family = ctor()
print("Using shrinkage estimator of dispersal in negative binomial GLM.")
raise Exception("yay!")
m = len(y)
alphas = self.alphas
if alphas is None:
res = np.zeros(m)
else:
res = np.zeros((len(alphas), m))
glm_res = glm.fit_glm(y, x, family)
kwargs = {}
if self.alphas is not None:
kwargs['smoothing'] = alphas
res = glm.f_test(glm_res.beta, contrast, glm_res.normalized_cov_params, glm_res.scale, **kwargs)
# For some reason the f-test returns an array with two extra dims
return res.reshape(res.shape[:-2])
def _run_family(self, family_name):
family = self.families[family_name]
glm_res = glm.fit_glm(self.input, self.exog, family)
f = glm.f_test(glm_res.beta, self.contrast,
glm_res.normalized_cov_params, glm_res.scale)
f = np.reshape(f, (len(f), ))
smoothed_f = glm.f_test(glm_res.beta,
self.contrast,
glm_res.normalized_cov_params,
glm_res.scale,
smoothing=np.zeros(3))
got = {
'beta': glm_res.beta,
'mu': glm_res.mu,
'weights': glm_res.weights,
'f_values': f
}
for (name, values) in got.items():
print("Doing " + name)
filename = 'pade/test/glm/{0}_{1}.txt'.format(family_name, name)
expected = np.genfromtxt(filename)
np.testing.assert_almost_equal(expected, values)
for i in range(3):
np.testing.assert_almost_equal(expected, values)
filename = 'pade/test/glm/{0}_{1}.txt'.format(family_name, 'f_values')
expected = np.genfromtxt(filename)
for i in range(3):
np.testing.assert_almost_equal(
expected, smoothed_f[i].reshape(len(smoothed_f[i])))
def fittedvalues(self, y):
return glm.fit_glm(y, self.x, self.family).mu
def fittedvalues(self, y):
return glm.fit_glm(y, self.x, self.family).mu